# GridToChip — full catalog

> Every physical part of an AI data center, and the companies that make each one.
> Confidence tags: disclosed = company/filing statement with source; reported = trade press
> or partner list; inferred = uncited lead from the taxonomy seed. Do not present reported or
> inferred placements as confirmed supplier relationships.

## Layer 1 — Compute silicon & packaging

### 1.01 Merchant GPUs / accelerators
https://gridtochip.site/parts/compute-silicon-and-packaging/merchant-gpus-accelerators
GPUs and other accelerator chips designed by merchant vendors and sold to any buyer, rather than designed in-house by a single hyperscaler; the workhorse silicon for AI training and inference.
AI delta: AI training and inference workloads are the primary demand driver for merchant accelerators, which is why unit volumes and generational upgrade cadence have both accelerated well beyond a typical chip replacement cycle.
Companies: AMD [disclosed], Intel [inferred], Nvidia [disclosed]

### 1.02 Hyperscaler custom ASICs (XPUs)
https://gridtochip.site/parts/compute-silicon-and-packaging/hyperscaler-custom-asics-xpus
AI accelerator chips (often called XPUs) designed in-house by a hyperscaler for its own fleet, typically co-developed with an outside ASIC design partner and fabbed by a merchant foundry.
AI delta: Hyperscalers commission custom ASICs specifically to cut the cost of AI training and inference at their own fleet scale, which is why this category exists at all outside of a few prior-generation exceptions.
Companies: Alchip [inferred], Alphabet (Google) [inferred], Amazon (AWS) [inferred], Broadcom [inferred], Marvell [inferred], MediaTek [reported], Meta Platforms [inferred], Microsoft [inferred], OpenAI [inferred]

### 1.03 ASIC design services, IP & EDA
https://gridtochip.site/parts/compute-silicon-and-packaging/asic-design-services-ip-and-eda
Outside firms and licensors — design-services houses, IP blocks, and EDA software — that hyperscalers and accelerator startups use to turn a chip specification into a manufacturable design.
AI delta: Every hyperscaler and well-funded AI startup now wants its own accelerator design, which has pulled design-services, IP licensing, and EDA demand up well beyond what general-purpose chip design alone would generate.
Companies: Alchip [inferred], Arm [inferred], Broadcom [inferred], Cadence [inferred], Credo [inferred], Faraday [inferred], Global Unichip [inferred], Marvell [inferred], MediaTek [inferred], Rambus [inferred], Socionext [inferred], Synopsys [inferred], Alphawave [inferred]

### 1.04 Startup accelerators (non-GPU architectures)
https://gridtochip.site/parts/compute-silicon-and-packaging/startup-accelerators-non-gpu-architectures
Non-GPU accelerator architectures from mostly-private startups, built as an alternative to merchant GPUs for specific AI training or inference workloads.
AI delta: The search for AI training or inference compute that is cheaper, faster, or more efficient than merchant GPUs for a specific workload is what funds this entire category of startups.
Companies: Graphcore [inferred], Cerebras [inferred], d-Matrix [inferred], Etched [inferred], Furiosa [inferred], Groq [inferred], MatX [inferred], Positron [inferred], Rebellions [inferred], SambaNova [inferred], Tenstorrent [inferred]

### 1.05 Server CPUs
https://gridtochip.site/parts/compute-silicon-and-packaging/server-cpus
The host CPU in an AI server — from AMD, Intel, or Arm-based designs including Nvidia's own and hyperscaler in-house parts — that orchestrates the GPUs and handles work the accelerators don't do.
AI delta: AI servers still need a host CPU to orchestrate GPUs and run non-matrix workloads, so CPU volume rises with every GPU-dense server shipped even though the CPU itself does little of the AI math.
Companies: AMD [disclosed], Ampere [inferred], Arm [inferred], Intel [inferred], Nvidia [disclosed]
- 1.05a Hyperscaler Arm CPUs: Alphabet (Google) [inferred], Amazon (AWS) [inferred], Microsoft [inferred]

### 1.06 NICs / SuperNICs
https://gridtochip.site/parts/compute-silicon-and-packaging/nics-supernics
Network interface cards, including Nvidia's higher-bandwidth "SuperNIC" variants, that connect a server's CPU and GPUs to the cluster network.
AI delta: Training and inference workloads depend on moving large tensors between accelerators at high bandwidth and low latency, which pulls NIC and SuperNIC speed and count up in step with cluster and GPU generation size.
Companies: AMD [inferred], Broadcom [inferred], Intel [inferred], Marvell [inferred], Nvidia [inferred]

### 1.07 DPUs
https://gridtochip.site/parts/compute-silicon-and-packaging/dpus
Data Processing Units that offload networking, storage, and security functions from the host CPU onto a dedicated, programmable chip.
AI delta: Offloading networking, storage, and security work from the host CPU matters more as GPU-dense AI servers push CPU cycles toward scarcity, since every cycle not spent on infrastructure work is a cycle available for orchestrating accelerators.
Companies: AMD [inferred], Intel [inferred], Marvell [inferred], Microsoft [inferred], Nvidia [inferred]

### 1.08 FPGAs (inference, networking, test)
https://gridtochip.site/parts/compute-silicon-and-packaging/fpgas-inference-networking
Field-programmable chips used in AI clusters for inference offload, networking functions, and hardware test and bring-up, rather than as the primary training accelerator.
AI delta: FPGAs pick up AI inference and networking tasks where a fixed accelerator design doesn't fit, and see added demand from the test and bring-up work that a growing volume of new AI silicon generates.
Companies: AMD [inferred], Lattice [inferred], Achronix [inferred], Altera [inferred]

### 1.09 Leading-edge foundry
https://gridtochip.site/parts/compute-silicon-and-packaging/leading-edge-foundry
The advanced-node semiconductor fabs that manufacture the compute dies for essentially every AI accelerator on the market.
AI delta: Every additional AI accelerator design and unit sold is another wafer order at the most advanced process nodes, concentrating incremental demand at a small number of qualified fabs.
Supply status: constrained
Companies: Intel [inferred], Samsung Electronics [disclosed], TSMC [reported]

### 1.10 Silicon wafers
https://gridtochip.site/parts/compute-silicon-and-packaging/silicon-wafers
The raw silicon wafers that foundries process into finished chips, produced by a small group of specialty wafer manufacturers.
AI delta: Wafer demand rises directly with foundry wafer starts for AI accelerators, and the larger, more complex AI dies add further pressure on the most advanced wafer specifications.
Companies: GlobalWafers [disclosed], Shin-Etsu [disclosed], Siltronic [disclosed], SK Siltron [disclosed], SUMCO [disclosed]

### 1.11 2.5D/3D advanced packaging (CoWoS-S/L/R, SoIC, EMIB/Foveros, I-Cube)
https://gridtochip.site/parts/compute-silicon-and-packaging/2-5d-3d-advanced-packaging-cowos-s-l-r-soic
The 2.5D/3D processes that combine a compute die, HBM memory stacks, and a substrate into one finished package — the step that turns silicon into a shippable AI accelerator.
AI delta: Multi-die AI accelerator packages — a compute die plus HBM stacks — require advanced 2.5D/3D packaging that ordinary logic chips never needed, and packaging capacity has repeatedly been the first-order constraint on how many finished accelerators can ship.
Supply status: constrained
Companies: Intel [inferred], Samsung Electronics [inferred], TSMC [inferred]
- 1.11a OSATs taking CoWoS back-end and overflow: Amkor [inferred], ASE Technology (ASE/SPIL) [inferred], KYEC [inferred], Powertech [inferred]

### 1.12 Packaging & bonding equipment
https://gridtochip.site/parts/compute-silicon-and-packaging/packaging-and-bonding-equipment
The bonding, dicing, inspection, and molding equipment used to physically assemble and verify advanced chip packages.
AI delta: Growth in advanced packaging volume for AI accelerators flows directly through to demand for the bonding, dicing, and inspection tools used to build that packaging.
Companies: Applied Materials [inferred], ASMPT [inferred], BESI [inferred], Camtek [inferred], Disco [inferred], Hanmi Semiconductor [inferred], Hanwha Semitech [inferred], KLA [inferred], Lam Research [inferred], Nova [inferred], Onto Innovation [inferred], Shibaura [inferred], SUSS MicroTec [inferred], Tokyo Electron [inferred], Towa [inferred], EV Group [inferred]

### 1.13 ABF / high-layer package substrates
https://gridtochip.site/parts/compute-silicon-and-packaging/abf-high-layer-package-substrates
The high-layer-count ABF substrates that an advanced chip package is mounted on, sitting between the interposer and the server board.
AI delta: Larger, more complex AI accelerator packages need bigger, higher-layer-count substrates than prior chip generations, straining the segment of substrate capacity qualified to build them.
Supply status: constrained
Companies: AT&S [disclosed], Daeduck [reported], Ibiden [disclosed], Kinsus [inferred], LG Innotek [disclosed], Nan Ya PCB [disclosed], Samsung Electro-Mechanics [disclosed], Unimicron [disclosed], Zhen Ding [disclosed], Shinko [disclosed]

### 1.14 Substrate materials
https://gridtochip.site/parts/compute-silicon-and-packaging/substrate-materials
The films, resins, glass cloth, and other specialty chemicals that go into building an ABF package substrate.
AI delta: Rising AI substrate volume and larger, higher-layer-count substrate designs pull directly on the limited capacity of the specialty films, resins, and cloth that go into them.
Supply status: constrained
Companies: Ajinomoto [disclosed], Asahi Kasei [disclosed], Mitsubishi Gas Chemical [disclosed], Nittobo [disclosed], Sumitomo Bakelite [disclosed]

### 1.15 Glass-core substrates (emerging)
https://gridtochip.site/parts/compute-silicon-and-packaging/glass-core-substrates-emerging
Substrates that use a glass core instead of an organic laminate, an emerging alternative aimed at the largest, most demanding AI packages.
AI delta: AI package sizes and layer counts are pushing conventional substrate materials toward their physical limits, which is the specific pressure pulling glass-core substrates from lab work toward qualification.
Companies: Absolics [inferred], AGC [inferred], Corning [inferred], Intel [inferred], Samsung Electro-Mechanics [inferred], Unimicron [inferred]

### 1.16 Interposers, silicon bridges, RDL
https://gridtochip.site/parts/compute-silicon-and-packaging/interposers-silicon-bridges-rdl
The silicon interposers, bridges, and redistribution layers that route signals between a compute die, HBM stacks, and the substrate inside an advanced package.
AI delta: More die per package and more instances of 2.5D/3D packaging per accelerator generation both require more interposer or bridge area, independent of overall unit growth.
Companies: Intel [inferred], Samsung Electronics [inferred], TSMC [inferred], UMC [inferred]

### 1.17 Chip-level thermal: lids, heat spreaders, TIMs
https://gridtochip.site/parts/compute-silicon-and-packaging/chip-level-thermal-lids-heat-spreaders-tims
The metal lids, heat spreaders, and thermal interface materials applied directly at the die level to move heat off an accelerator package before it ever reaches a cold plate or heat sink.
AI delta: Rising AI accelerator power density raises the performance bar on die-level lids, heat spreaders, and thermal interface materials generation over generation.
Companies: Dow [inferred], Henkel [inferred], Honeywell [inferred], Jentech [inferred], Laird [inferred], Shin-Etsu [inferred], Boyd [inferred], Fujipoly [inferred], Indium [inferred]

### 1.18 Semiconductor test & burn-in
https://gridtochip.site/parts/compute-silicon-and-packaging/semiconductor-test-and-burn-in
The automated test equipment, probe cards, sockets, and burn-in systems used to verify a chip works before and after it's packaged.
AI delta: Larger, more complex, and higher-value AI dies raise the cost of letting a defective unit through, pushing more test time and more test insertions onto each unit shipped.
Companies: Advantest [disclosed], Aehr [disclosed], Cohu [disclosed], FormFactor [disclosed], ISC [inferred], KYEC [inferred], Leeno [inferred], MPI [inferred], Technoprobe [disclosed]

### 1.19 Fab equipment (upstream)
https://gridtochip.site/parts/compute-silicon-and-packaging/fab-equipment-upstream
The lithography, deposition, etch, and inspection tools that fabs use to manufacture chips — upstream capital equipment rather than a part that ships in any finished accelerator.
AI delta: Every increment of new advanced-node wafer capacity added for AI accelerators is new fab equipment demand somewhere upstream in the process, ordered years ahead of the wafers it will eventually produce.
Supply status: tight
Companies: Applied Materials [inferred], ASM International [disclosed], ASML [disclosed], KLA [disclosed], Kokusai [disclosed], Lam Research [disclosed], Screen [inferred], Tokyo Electron [disclosed]

### 1.20 Fab materials, masks, gases (upstream)
https://gridtochip.site/parts/compute-silicon-and-packaging/fab-materials-masks-gases-upstream
The photoresist, masks, process chemicals, sputtering targets, and specialty gases consumed in volume during chip fabrication.
AI delta: Fab materials consumption scales with wafer starts, and AI accelerator dies' larger size and greater process-step complexity add further consumption per wafer on top of that.
- 1.20a Photoresist: DuPont [inferred], Fujifilm [inferred], Shin-Etsu [inferred], Tokyo Ohka [inferred], JSR [inferred]
- 1.20b Masks and blanks: DNP [inferred], Hoya [inferred], Photronics [disclosed], Toppan [inferred]
- 1.20c CMP and chemicals: Entegris [disclosed], Fujimi [inferred], Merck KGaA [inferred], Resonac [disclosed], Soulbrain [inferred], Stella Chemifa [inferred]
- 1.20d Sputtering targets: JX Advanced Metals [disclosed], Materion [disclosed]
- 1.20e Gases: Air Liquide [disclosed], Air Products [disclosed], Linde [inferred], Taiyo Nippon Sanso [disclosed]

## Layer 2 — Memory & storage

### 2.01 HBM (HBM3E, HBM4)
https://gridtochip.site/parts/memory-and-storage/hbm-hbm3e-hbm4
High-bandwidth memory stacked directly on the same package as an AI accelerator, giving the GPU or ASIC very fast, close-range access to the memory it needs to stay fed.
AI delta: Larger models and longer context windows push accelerator designs toward more HBM stacks and higher per-stack capacity with each generation, keeping HBM the tightest link in the memory chain.
Supply status: constrained
Companies: BESI [inferred], Disco [inferred], Hanmi Semiconductor [disclosed], Micron [disclosed], Samsung Electronics [disclosed], SK Hynix [disclosed]

### 2.02 DRAM (DDR5 RDIMM/MRDIMM, LPDDR5X for Grace)
https://gridtochip.site/parts/memory-and-storage/dram-ddr5-rdimm-mrdimm-lpddr5x-for-grace
Off-package system memory — DDR5 modules for host CPUs and LPDDR5X soldered near Nvidia's Grace CPU — that holds the working data a server's processors operate on.
AI delta: AI servers' data-staging and checkpointing needs raise host DRAM content per node, and HBM production competing for the same fab capacity tightens standard DRAM supply alongside it.
Supply status: constrained
Companies: Micron [disclosed], Nanya [disclosed], Samsung Electronics [disclosed], SK Hynix [disclosed], Winbond [disclosed], CXMT [disclosed]

### 2.03 Memory modules
https://gridtochip.site/parts/memory-and-storage/memory-modules
The physical DIMM (or module) that packages raw DRAM die, substrate, and management circuitry into the pluggable unit a server actually sockets.
AI delta: Higher memory-per-socket requirements in AI servers push module makers toward higher-capacity, higher-rank designs that carry more assembly and qualification content per unit.
Companies: ADATA [inferred], Micron [disclosed], Netlist [disclosed], Penguin Solutions/SMART [disclosed], Samsung Electronics [disclosed], SK Hynix [disclosed], Kingston [inferred]

### 2.04 Memory interface / buffer chips (RCD, DB, MRCD/MDB)
https://gridtochip.site/parts/memory-and-storage/memory-interface-buffer-chips-rcd-db
Small companion chips on a memory module that buffer and re-drive command, address, and data signals so many DRAM chips can be packed per module without losing signal integrity.
AI delta: Higher-capacity, higher-rank modules needed to keep memory in step with AI-era core and accelerator counts make buffer and register chips a near-mandatory part of the module rather than an optional add-on.
Companies: Montage [disclosed], Rambus [disclosed], Renesas [disclosed]

### 2.05 CXL controllers & memory expanders
https://gridtochip.site/parts/memory-and-storage/cxl-controllers-and-memory-expanders
Chips that let a server attach an extra pool of DRAM over a CXL link rather than only through the CPU's own memory channels, expanding usable capacity beyond what fits in local DIMM slots.
AI delta: AI workloads' appetite for memory beyond what fits in a single node's DIMM slots is what makes pooled, CXL-attached memory worth the added system complexity.
Companies: Astera Labs [disclosed], Marvell [inferred], Microchip [inferred], Micron [inferred], Montage [disclosed], Rambus [disclosed], Samsung Electronics [disclosed], SK Hynix [disclosed]

### 2.06 PCIe/CXL switches & retimers
https://gridtochip.site/parts/memory-and-storage/pcie-cxl-switches-and-retimers
Fabric silicon — switches that fan a single PCIe or CXL root out to many downstream devices, and retimers that clean up the signal over longer runs — connecting accelerators, SSDs, and pooled memory back to the host.
AI delta: Packing many more PCIe and CXL endpoints per AI server than a general-purpose server requires drives up switch and retimer content per system.
Companies: Astera Labs [inferred], Broadcom [inferred], Marvell [inferred], Microchip [inferred], Xconn [inferred]

### 2.07 NAND flash
https://gridtochip.site/parts/memory-and-storage/nand-flash
Non-volatile flash memory die — the raw storage medium later packaged into SSDs — that holds data and model artifacts when the system is powered off.
AI delta: Frequent checkpointing during training and local weight caching for inference push NAND endurance and capacity needs above typical enterprise storage workloads.
Companies: Kioxia [disclosed], Micron [disclosed], Samsung Electronics [disclosed], SanDisk [disclosed], SK Hynix [disclosed], YMTC [inferred]

### 2.08 Enterprise SSDs & controllers
https://gridtochip.site/parts/memory-and-storage/enterprise-ssds-and-controllers
Solid-state drives built for data center duty cycles, and the controller chips that manage them — the fast local storage tier that feeds checkpoints, staged datasets, and cached weights to accelerators.
AI delta: GPU-direct storage architectures and dense compute-tray form factors have pulled SSD content per AI server well above what general enterprise storage designs carry.
- 2.08a Drives: Kioxia [disclosed], Micron [disclosed], Phison [disclosed], Samsung Electronics [disclosed], SanDisk [disclosed], Seagate [inferred], SK Hynix [disclosed], Kingston [inferred]
- 2.08b Controllers: Kioxia [inferred], Marvell [inferred], Microchip [inferred], Micron [inferred], Phison [disclosed], Samsung Electronics [inferred], Silicon Motion [disclosed]

### 2.09 HDD (nearline / object tiers)
https://gridtochip.site/parts/memory-and-storage/hdd-nearline-object-tiers
Spinning hard disk drives used for the bulk, lower-cost nearline and object storage tier that holds data too large or too cold to justify keeping on flash.
AI delta: The sheer volume of training data, checkpoints, and generated content needing affordable long-term retention keeps nearline HDD demand growing alongside, not despite, flash's rise in performance-sensitive tiers.
Companies: Seagate [inferred], Western Digital [disclosed], Toshiba [disclosed]

### 2.10 AI storage systems (parallel file systems, object, GPU-direct)
https://gridtochip.site/parts/memory-and-storage/ai-storage-systems-parallel-file-systems-object
Purpose-built storage software and appliances — parallel file systems, object stores, and GPU-direct pipelines — that aggregate many drives into one system capable of feeding a large number of accelerators at once.
AI delta: Large training jobs need read throughput that general enterprise storage was not designed to sustain, which is what has created a distinct category of storage system built around GPU-direct access.
Companies: Dell [disclosed], Everpure [disclosed], Hitachi Vantara [disclosed], HPE [inferred], IBM [disclosed], NetApp [inferred], Supermicro [inferred], DDN [disclosed], Hammerspace [disclosed], Huawei [inferred], MinIO [disclosed], Qumulo [disclosed], VAST Data [disclosed], Weka [disclosed]

## Layer 3 — Networking & interconnect

### 3.01 Scale-up fabric (GPU-to-GPU inside the rack/pod)
https://gridtochip.site/parts/networking-and-interconnect/scale-up-fabric-gpu-to-gpu-inside-the-rack-pod
The fabric linking GPUs to each other inside a single rack or pod, separate from the switching fabric that connects racks to each other.
AI delta: Larger AI models split across more GPUs per training or inference job, which raises the bandwidth and reach demanded of the scale-up fabric linking them inside a rack or pod.
Companies: Alphabet (Google) [disclosed], AMD [inferred], Astera Labs [disclosed], Broadcom [inferred], Nvidia [disclosed], UALink consortium [disclosed]

### 3.02 Scale-out switch ASICs
https://gridtochip.site/parts/networking-and-interconnect/scale-out-switch-asics
The switch ASICs that route traffic between racks and pods, distinct from the scale-up fabric inside a single rack.
AI delta: AI clusters need far more inter-rack bandwidth than general-purpose data centers, pushing scale-out switch ASICs toward higher port counts and per-port speeds sooner.
Companies: Broadcom [disclosed], Cisco [inferred], Marvell [inferred], Nvidia [disclosed], Huawei [inferred]

### 3.03 Switch systems & white-box ODMs
https://gridtochip.site/parts/networking-and-interconnect/switch-systems-and-white-box-odms
The physical switches — built by branded vendors and white-box ODMs — that house scale-out switch ASICs and connect racks into a cluster fabric.
AI delta: Hyperscale AI clusters increasingly specify white-box switch systems built to their own designs around merchant silicon, shifting share away from fully branded switch platforms.
Companies: Accton (Edgecore) [inferred], Alpha Networks [inferred], Arista [inferred], Celestica [disclosed], Cisco [inferred], Dell [inferred], Delta Electronics [inferred], HPE [inferred], Nvidia [inferred], Quanta Computer [inferred], Wistron NeWeb [inferred], Huawei [inferred], Ufispace [disclosed]

### 3.04 Optical transceivers (400G / 800G / 1.6T)
https://gridtochip.site/parts/networking-and-interconnect/optical-transceivers-400g-800g-1-6t
Pluggable optical modules that convert electrical signals to light for transmission over fiber between switches and servers, sold at 400G, 800G, and 1.6T speeds.
AI delta: AI clusters need far more optical ports per GPU than general-purpose data centers, and the push to higher per-GPU bandwidth accelerates the transition from 400G to 800G to 1.6T modules.
Companies: Accelink [disclosed], Applied Optoelectronics [disclosed], Broadcom [inferred], Cisco [inferred], Coherent [disclosed], Eoptolink [disclosed], HG Genuine [inferred], Innolight [disclosed], Jabil [inferred], Lumentum [disclosed], Nvidia [disclosed], TFC [inferred], Hisense Broadband [inferred], Molex [inferred], Source Photonics [inferred]

### 3.05 Optics contract manufacturing
https://gridtochip.site/parts/networking-and-interconnect/optics-contract-manufacturing
Contract manufacturers that assemble and test optical transceivers on behalf of module brands and switch vendors.
AI delta: Surging optical module demand from AI clusters has concentrated assembly volume in the small group of contract manufacturers already qualified at the highest optical speeds.
Companies: Celestica [inferred], Fabrinet [inferred], Flex [inferred], Jabil [inferred], Sanmina [inferred]

### 3.06 Lasers (EML, CW/DFB, VCSEL)
https://gridtochip.site/parts/networking-and-interconnect/lasers-eml-cw-dfb
The laser sources — EML, CW/DFB, and VCSEL types — that generate the light signal inside optical transceivers and co-packaged optics.
AI delta: Rising optical transceiver and co-packaged optics volume from AI clusters increases demand for qualified laser sources across every reach and speed grade.
Companies: Broadcom [inferred], Coherent [inferred], Lumentum [inferred], Mitsubishi Electric [inferred], POET [disclosed], Sivers [disclosed], Sumitomo Electric [disclosed], Source Photonics [inferred]

### 3.07 Optical DSPs, drivers, TIAs
https://gridtochip.site/parts/networking-and-interconnect/optical-dsps-drivers-tias
The signal-processing silicon inside optical modules — DSPs plus the drivers and TIAs that condition the electrical signal on either side of the laser and photodetector.
Companies: Broadcom [inferred], Credo [disclosed], Marvell [inferred], MaxLinear [inferred]
- 3.07a Drivers and TIAs: Analog Devices [inferred], Macom [inferred], Semtech [disclosed]

### 3.08 Silicon photonics & co-packaged optics (CPO)
https://gridtochip.site/parts/networking-and-interconnect/silicon-photonics-and-co-packaged-optics-cpo
An approach that integrates optical components directly onto or next to the switch or GPU package, replacing some pluggable transceivers with light routed closer to the silicon.
AI delta: Rising per-GPU bandwidth is what makes co-packaged optics attractive, since the power and cable-count savings only outweigh pluggable optics' simplicity at AI-scale link speeds.
Companies: Broadcom [inferred], GlobalFoundries [disclosed], Intel [inferred], Marvell [inferred], Nvidia [inferred], Tower Semiconductor [disclosed], TSMC [inferred], Ayar Labs [inferred], Celestial AI [inferred], Lightmatter [disclosed], Nubis [inferred], Ranovus [inferred], Xscape [disclosed]
- 3.08a External laser sources: Coherent [inferred], Lumentum [inferred]
- 3.08b Fiber attach and connectors: Corning [inferred], Senko [inferred], US Conec [inferred]

### 3.09 Copper: DAC / ACC / AEC cables
https://gridtochip.site/parts/networking-and-interconnect/copper-dac-acc-aec-cables
High-speed copper cable assemblies — DAC, ACC, and AEC types — that link GPUs, NICs, and switches over short distances as a lower-cost alternative to optical transceivers.
AI delta: AI racks pack far more high-speed copper links per GPU than general-purpose servers, and rising per-link speeds push volume from passive DAC toward active ACC and AEC assemblies.
Companies: Amphenol [inferred], Bizlink [inferred], Credo [disclosed], Foxconn Interconnect [inferred], Luxshare [inferred], TE Connectivity [inferred], Volex [inferred], Leoni [inferred], Molex [inferred]
- 3.09a Retimers inside AECs: Astera Labs [inferred], Broadcom [inferred], Marvell [inferred], Spectra7 [inferred]

### 3.10 NVLink copper spine / backplane cartridges
https://gridtochip.site/parts/networking-and-interconnect/nvlink-copper-spine-backplane-cartridges
The copper cable assemblies and backplane cartridges that physically wire the scale-up fabric together inside a rack, most visible in Nvidia's NVL72 architecture.
AI delta: Larger scale-up domains push cable count per rack higher, up to the point where copper's reach and density limits at the highest speeds make an optical alternative necessary.
Companies: Amphenol [reported], Bizlink [reported], Foxconn Interconnect [reported], Luxshare [reported], TE Connectivity [reported]

### 3.11 Fiber, connectors, structured cabling
https://gridtochip.site/parts/networking-and-interconnect/fiber-connectors-structured-cabling
The fiber optic cable, connectors, and structured cabling that carry optical signals between transceivers across a data hall or campus.
AI delta: Higher optical port counts per GPU and per rack in AI clusters increase the volume of fiber, connectors, and structured cabling needed per unit of compute deployed.
Companies: Belden [inferred], Corning [inferred], Fujikura (AFL) [inferred], Furukawa [inferred], Hengtong [inferred], Nexans [inferred], Prysmian [disclosed], Sumitomo Electric [inferred], YOFC [inferred], Leviton [disclosed], Panduit [inferred], Siemon [disclosed]
- 3.11a MPO/MTP ferrules and connectors: Senko [inferred], US Conec [inferred]

### 3.12 DCI / coherent optics (campus-to-campus, metro)
https://gridtochip.site/parts/networking-and-interconnect/dci-coherent-optics-campus-to-campus-metro
Coherent optical systems that carry data center interconnect (DCI) traffic between buildings on a campus or between metro sites, distinct from the shorter-reach optics inside a data hall.
AI delta: Multi-building AI campuses and multi-site training clusters push more traffic onto campus-to-campus and metro coherent links than single-building data centers required.
Companies: Adtran [inferred], Ciena [inferred], Cisco [inferred], Coherent [inferred], Lumentum [inferred], Marvell [inferred], Nokia [inferred], ZTE [inferred], Huawei [inferred]

### 3.13 Timing & telemetry silicon
https://gridtochip.site/parts/networking-and-interconnect/timing-and-telemetry-silicon
Precision timing and telemetry chips that synchronize clocks and monitor signal integrity across the network fabric.
AI delta: Higher network speeds and larger, more complex AI cluster topologies tighten timing tolerances and increase the value of telemetry coverage across the fabric.
Companies: Broadcom [inferred], Marvell [inferred], Microchip [inferred], Renesas [disclosed], SiTime [disclosed]

## Layer 4 — Servers, racks & integration

### 4.01 Rack-scale systems (NVL72-class, MI400 Helios-class)
https://gridtochip.site/parts/servers-racks-and-integration/rack-scale-systems-nvl72-class-mi400-helios-class
A rack-scale system is the fully integrated cabinet of compute trays, switch trays, power shelves, cooling manifolds and busbar that ships as one wired unit rather than as discrete servers.
AI delta: AI training and inference workloads are the reason rack-scale systems exist at all — no other server class assembles dozens of GPUs into one wired, liquid-cooled domain.
Companies: AMD [inferred], ASRock Rack [inferred], ASUS [inferred], Cisco [inferred], Compal [inferred], Dell [disclosed], Eviden/Atos [inferred], Foxconn (Hon Hai) [inferred], Fujitsu [inferred], Gigabyte [inferred], H3C [inferred], HPE [inferred], Inspur [inferred], Inventec [disclosed], Lenovo [inferred], MiTAC (Tyan) [inferred], NEC [inferred], Nvidia [disclosed], Pegatron [inferred], Quanta Computer [inferred], Supermicro [inferred], Wiwynn [inferred], xFusion [inferred]

### 4.02 EMS, system integration & rack-level test
https://gridtochip.site/parts/servers-racks-and-integration/ems-system-integration-and-rack-level-test
Electronics manufacturing services firms assemble, wire and burn-in the finished server or rack before it ships, standing between the ODM board build and the customer's data center floor.
AI delta: AI rack complexity — liquid cooling, dense internal cabling, tight power budgets — raises the bar for pre-ship test and burn-in well past what a standard server requires.
Companies: Benchmark [inferred], Celestica [inferred], Compal [inferred], Flex [disclosed], Foxconn (Hon Hai) [inferred], Inventec [inferred], Jabil [inferred], Kimball [inferred], Plexus [inferred], Quanta Computer [inferred], Sanmina [disclosed], Wistron [inferred]
- 4.02a Rack test: Chroma [inferred], Keysight [inferred]

### 4.03 GPU baseboards & compute trays (HGX, Bianca, UBB/OAM)
https://gridtochip.site/parts/servers-racks-and-integration/gpu-baseboards-and-compute-trays-hgx-bianca
The baseboard is the printed circuit assembly that carries the GPUs (and often paired CPUs) and their direct power and interconnect, packaged into a tray that slides into the rack.
AI delta: A compute tray carrying multiple high-power GPUs with coherent CPU links and dense scale-up interconnect is a fundamentally different design than a general-purpose server motherboard, which is why this category exists as its own part class.
Companies: Foxconn (Hon Hai) [inferred], Inventec [inferred], Nvidia [disclosed], Pegatron [inferred], Quanta Computer [inferred], Supermicro [inferred], Wistron [inferred], Wiwynn [inferred]

### 4.04 Motherboards, CPU trays, switch trays
https://gridtochip.site/parts/servers-racks-and-integration/motherboards-cpu-trays-switch-trays
The non-GPU trays in a rack — CPU-only compute trays, host motherboards and the switch trays carrying scale-up switch silicon — built by the same board makers as the GPU trays.
AI delta: Rack-scale GPU systems require dedicated switch trays purely to fan out the scale-up fabric between GPU trays, a tray category that has little equivalent in conventional servers.
Companies: ASRock Rack [inferred], ASUS [inferred], Gigabyte [inferred], MiTAC (Tyan) [inferred], MSI [inferred]

### 4.05 High-layer-count PCBs (HDI, backdrilled, 20–40 layers)
https://gridtochip.site/parts/servers-racks-and-integration/high-layer-count-pcbs-hdi-backdrilled
The bare printed circuit boards themselves — high-density interconnect, backdrilled, 20 to 40 copper layers — that GPU and switch trays are built on.
AI delta: AI boards need far more routing layers and tighter fabrication tolerances than a standard server motherboard, because dense scale-up interconnect and high-pin-count sockets have to be routed across the same board.
Companies: AT&S [disclosed], Compeq [inferred], Dongshan [inferred], Gold Circuit GCE [inferred], Ibiden [inferred], ISU Petasys [inferred], Kingboard [inferred], Nan Ya PCB [inferred], Shennan [inferred], Tripod [inferred], TTM [inferred], Unimicron [inferred], Victory Giant [inferred], WUS [inferred], Zhen Ding [inferred]

### 4.06 Copper-clad laminate & PCB materials
https://gridtochip.site/parts/servers-racks-and-integration/copper-clad-laminate-and-pcb-materials
The laminate — resin, copper foil and glass-cloth reinforcement — that high-layer-count boards (4.05) are fabricated from.
AI delta: Higher board layer counts and faster signaling push demand toward low-loss, low-dielectric-constant laminate grades instead of commodity material, shifting the mix even where total board area doesn't grow as fast.
Companies: Doosan [inferred], Elite Material [inferred], Iteq [inferred], Nanya Plastics [inferred], Panasonic [inferred], Rogers [disclosed], Shengyi [inferred], Taiwan Union TUC [inferred], Isola [inferred]
- 4.06a Copper foil: Furukawa [inferred], Mitsui Kinzoku [inferred], Chang Chun [inferred]
- 4.06b Low-Dk glass cloth: Asahi Kasei [inferred], Nittobo [inferred]

### 4.07 Board-level connectors & sockets
https://gridtochip.site/parts/servers-racks-and-integration/board-level-connectors-and-sockets
The physical connectors — CPU and GPU sockets, board-to-board and cable connectors, and I/O cages — that link chips, trays and cables together on and between boards.
AI delta: Higher pin counts, higher currents and faster signaling with each GPU generation make socket and connector design a harder mechanical problem, and AI racks need more connectors per system than a conventional server simply because there are more trays to link.
Companies: Aces [disclosed], Amphenol [inferred], Bizlink [disclosed], Foxconn Interconnect [inferred], Hirose [disclosed], JAE [inferred], Lotes [inferred], Luxshare [inferred], TE Connectivity [inferred], Yamaichi [disclosed], Molex [inferred], Samtec [disclosed]
- 4.07a OSFP/QSFP cages: Amphenol [inferred], TE Connectivity [inferred], Molex [inferred]

### 4.08 Board-level power: VRMs, power stages, PoL, power semis
https://gridtochip.site/parts/servers-racks-and-integration/board-level-power-vrms-power-stages-pol-power-semis
The voltage regulator modules and point-of-load converters that step rack-level DC power down to the precise, high-current voltages each GPU and chip needs at the board.
AI delta: GPU power draw is both higher and far more transient than CPU power draw, which pushes VRM design toward faster response and higher current density than board-level power delivery has historically needed.
Companies: Alpha & Omega [inferred], Analog Devices [inferred], Delta Electronics [disclosed], Infineon [disclosed], Monolithic Power Systems [inferred], onsemi [inferred], Renesas [disclosed], Rohm [disclosed], STMicroelectronics [inferred], Texas Instruments [disclosed], Vicor [disclosed], Empower [inferred]
- 4.08a GaN and SiC: Coherent [inferred], Infineon [inferred], Navitas [inferred], Wolfspeed [inferred], EPC [inferred]

### 4.09 Passives (MLCC, inductors, polymer/tantalum, resistors)
https://gridtochip.site/parts/servers-racks-and-integration/passives-mlcc-inductors
The multilayer ceramic capacitors, inductors and other passive components that filter and stabilize power everywhere on a board, populated in far greater numbers on AI boards than on standard servers.
AI delta: roughly 10–15× the passive count of a standard server; high-spec MLCC lead times around 24 weeks (mid-2026 trade press).
Supply status: tight
Companies: Chilisin [inferred], Fenghua [inferred], Kyocera (AVX) [disclosed], Murata [disclosed], Panasonic [inferred], Samsung Electro-Mechanics [disclosed], Sunlord [inferred], Taiyo Yuden [disclosed], TDK [inferred], Vishay [disclosed], Walsin [inferred], Yageo [disclosed]

### 4.10 BMC & platform firmware
https://gridtochip.site/parts/servers-racks-and-integration/bmc-and-platform-firmware
The baseboard management controller chip and its firmware, which independently monitor power, temperature and health on a server board and let it be managed remotely, separate from the host CPU or GPU.
AI delta: Liquid cooling and tighter thermal margins in AI racks make independent, out-of-band health monitoring more operationally critical than it was in lower-density air-cooled servers, since a leak or thermal excursion has to be caught by something other than the host OS.
Companies: Aspeed [inferred], Insyde [disclosed], Nuvoton [inferred], Nvidia [inferred], AMI [disclosed], Phoenix [inferred]

### 4.11 Timing, security & misc board silicon
https://gridtochip.site/parts/servers-racks-and-integration/timing-security-and-misc-board-silicon
Clock and timing chips, TPM security modules, and boot flash — the smaller fixed-function silicon that supports the CPU, GPU and BMC on every board without doing the compute itself.
AI delta: Higher signaling rates on scale-up and scale-out interconnect need tighter clock jitter and synchronization tolerances than lower-speed server interfaces required, raising the specification bar on timing silicon specifically.
Companies: Analog Devices [inferred], Infineon [inferred], Microchip [inferred], Renesas [disclosed], SiTime [inferred], Texas Instruments [inferred], Winbond [inferred]

### 4.12 Power shelves & PSUs (33 kW-class ORV3 shelves, rack-level)
https://gridtochip.site/parts/servers-racks-and-integration/power-shelves-and-psus-33-kw-class-orv3-shelves-rack-level
The rack-mounted power shelves that convert facility AC or DC input into the regulated rail distributed to trays across the rack, built to the Open Rack V3 33 kW-class spec used in rack-scale AI systems.
AI delta: GPU trays draw far more continuous power than general-purpose compute, which is why rack-scale AI systems need substantially more power-shelf capacity per rack than a conventional server rack, and why wide-bandgap semiconductors are displacing silicon inside the shelf.
Companies: AcBel [inferred], Advanced Energy/Artesyn [disclosed], Bel Fuse [disclosed], Chicony Power [inferred], Cosel [inferred], Delta Electronics [inferred], Eaton [inferred], Flex [inferred], FSP [inferred], Lite-On [inferred], Murata [inferred], Vertiv [inferred]
- 4.12a GaN and SiC inside: Infineon [inferred], Navitas [inferred], onsemi [inferred], Wolfspeed [inferred], EPC [inferred]

### 4.13 Fans (air-cooled servers, hybrid trays)
https://gridtochip.site/parts/servers-racks-and-integration/fans-air-cooled-servers-hybrid-trays
The axial and centrifugal fans mounted in server chassis and rack trays that move air across heatsinks and boards, still needed even in racks that use liquid cooling for the highest-power chips.
AI delta: Even in a liquid-cooled rack, cold plates typically cool only the GPUs, so fans remain necessary for memory, VRMs and networking silicon on the same tray, making air-cooling components a hybrid-era holdover rather than something liquid cooling eliminates outright.
Companies: AVC [inferred], Delta Electronics [inferred], Foxconn (Hon Hai) [inferred], Nidec [inferred], Sanyo Denki [disclosed], Sunon [inferred], Cooler Master [inferred], ebm-papst [inferred]

### 4.14 Chassis, rails, rack enclosures
https://gridtochip.site/parts/servers-racks-and-integration/chassis-rails-rack-enclosures
The sheet-metal chassis and mounting rails that trays slide into, and the enclosure (the rack cabinet itself) that houses the whole assembly.
AI delta: Liquid-cooling plumbing routed through the chassis, plus the added weight of cold plates, manifolds and denser boards, puts more mechanical load on rails and enclosures than a conventional air-cooled server rack requires.
Companies: Chenbro [inferred], Foxconn (Hon Hai) [inferred], King Slide [inferred], Wistron [inferred], AIC [inferred]
- 4.14a Enclosures: Eaton [inferred], Legrand [inferred], nVent [inferred], Schneider Electric [inferred], Vertiv [inferred], Chatsworth [inferred], Panduit [inferred], Rittal [inferred]

### 4.15 In-rack busbar & power whips
https://gridtochip.site/parts/servers-racks-and-integration/in-rack-busbar-and-power-whips
The high-current busbar and cabling that carries power from the rack's power shelves to each tray, including Nvidia's specified ±48 V / 1,400 A busbar used in rack-scale liquid-cooled systems.
AI delta: Fully populated GPU racks need to move more current internally than a bundle of discrete cables handles well, which is why rack-scale systems increasingly use a solid busbar rather than conventional power whips.
Companies: Amphenol [reported], Bizlink [reported], Delta Electronics [reported], Eaton [reported], Lite-On [reported], Schneider Electric [reported]

### 4.16 Rack PDUs
https://gridtochip.site/parts/servers-racks-and-integration/rack-pdus
The power distribution units mounted in or beside a rack that meter and distribute power to individual devices, sitting between facility power and the rack's internal busbar or whips.
AI delta: As power becomes the binding constraint on how many racks a facility can support, rack-level and outlet-level metering on the PDU becomes more operationally important than in a lower-density, less power-constrained enterprise deployment.
Companies: Cyber Power [disclosed], Delta Electronics [disclosed], Eaton [inferred], Legrand [inferred], nVent [inferred], Schneider Electric [inferred], Vertiv [inferred], Panduit [inferred], Rittal [inferred]

### 4.17 Air-cooled heatsinks & vapor chambers
https://gridtochip.site/parts/servers-racks-and-integration/air-cooled-heatsinks-and-vapor-chambers
The finned aluminum or copper heatsinks and vapor chambers that spread and dissipate heat from chips that rely on airflow (4.13) rather than a liquid cold plate.
AI delta: As the highest-power chips in a rack move to liquid cooling, heatsinks and vapor chambers narrow to the components liquid cooling doesn't reach, shrinking in scope even as they get squeezed into more crowded, liquid-plumbed chassis.
Companies: Auras [inferred], AVC [inferred], Chaun-Choung CCI [inferred], Delta Electronics [inferred], Foxconn (Hon Hai) [inferred], Jentech [inferred], Boyd [inferred], Cooler Master [inferred], Taisol [inferred]

## Layer 5 — Liquid cooling (IT-side: chip to CDU)

### 5.01 Direct-to-chip cold plates
https://gridtochip.site/parts/liquid-cooling-it-side-chip-to-cdu/direct-to-chip-cold-plates
A metal plate with internal microchannels, bonded or clamped directly onto a GPU, CPU, or switch package, that carries coolant across the die to remove heat by conduction rather than airflow.
AI delta: Rising per-die power density pushed liquid cold plates from a niche option to the default thermal interface on AI accelerator packages, direct-to-chip cooling being the entry point for the two-phase and microfluidic variants under development for the next density step.
Companies: Auras [inferred], AVC [inferred], Delta Electronics [inferred], Flex [inferred], Fositek [reported], Foxconn (Hon Hai) [inferred], Jentech [inferred], Kaori [inferred], Lotes [reported], Nidec [inferred], Vertiv [inferred], Wiwynn [inferred], Boyd [disclosed], Cooler Master [inferred], CoolIT [disclosed]
- 5.01a Two-phase and microfluidic: Accelsius [inferred], Chilldyne [inferred], Corintis [inferred], Iceotope [inferred], Mikros [inferred], Zutacore [inferred]

### 5.02 Rack manifolds (CDMs)
https://gridtochip.site/parts/liquid-cooling-it-side-chip-to-cdu/rack-manifolds-cdms
The manifold (sometimes called a CDM) that distributes coolant from the CDU supply line to each cold plate in a rack and collects the return flow, usually mounted vertically at the rear or side of the rack.
Companies: Auras [inferred], AVC [inferred], Delta Electronics [inferred], Foxconn (Hon Hai) [inferred], Kaori [inferred], nVent [disclosed], Parker Hannifin [inferred], Vertiv [inferred], Boyd [disclosed], CoolIT [disclosed], Stäubli [inferred]

### 5.03 Quick disconnects (UQD / UQDB)
https://gridtochip.site/parts/liquid-cooling-it-side-chip-to-cdu/quick-disconnects-uqd-uqdb
Blind-mate connectors (UQD or UQDB) that let a compute tray or cold plate be connected to or disconnected from the coolant manifold without draining or spilling the loop.
AI delta: Liquid cooling moving from optional to default on AI accelerator trays scaled the number of blind-mate quick disconnects per rack with GPU count.
Companies: CPC/Colder [inferred], Eaton [inferred], Fositek [reported], Parker Hannifin [inferred], Danfoss [disclosed], Stäubli [inferred], Swagelok [disclosed]

### 5.04 Hoses, tubing, fittings, valves
https://gridtochip.site/parts/liquid-cooling-it-side-chip-to-cdu/hoses-tubing-fittings-valves
The flexible hoses, rigid tubing, compression and push-fit fittings, and shutoff or balancing valves that carry coolant between manifolds, quick disconnects, and the CDU.
Companies: Belimo [inferred], Continental [inferred], Eaton [inferred], Emerson [inferred], Gates [inferred], IMI [inferred], Parker Hannifin [inferred], Danfoss [inferred], Swagelok [inferred]

### 5.05 Coolant distribution units (in-rack, in-row, 1–2.5 MW+ facility units)
https://gridtochip.site/parts/liquid-cooling-it-side-chip-to-cdu/coolant-distribution-units-in-rack-in-row
The unit that separates the IT-side (technology) coolant loop from the facility water loop, exchanging heat between them and pumping and filtering the technology-side flow — sized from in-rack and in-row units up to multi-megawatt facility-class units.
AI delta: The shift of AI accelerator racks from air to liquid cooling pushed CDU capacity requirements up sharply and pulled facility-class (1–2.5 MW-plus) units into more deployments alongside smaller in-rack and in-row units.
Companies: Airedale [inferred], Auras [inferred], Daikin [inferred], Delta Electronics [inferred], Envicool [inferred], Foxconn (Hon Hai) [inferred], Johnson Controls [inferred], Kaori [inferred], Mitsubishi Electric [reported], Motivair [inferred], Munters [inferred], Nidec [inferred], nVent [inferred], Supermicro [inferred], Trane [inferred], Vertiv [inferred], Airsys [inferred], Boyd [inferred], Chilldyne [inferred], CoolIT [inferred], Huawei [inferred], LiquidStack [inferred], Rittal [inferred], Stulz [inferred]

### 5.06 CDU internals: plate heat exchangers, pumps, filters
https://gridtochip.site/parts/liquid-cooling-it-side-chip-to-cdu/cdu-internals-plate-heat-exchangers-pumps-filters
The internal components of a CDU — plate heat exchangers, pumps, and filters — that do the actual heat exchange, circulation, and particulate control inside the unit.
Companies: Alfa Laval [disclosed], Kaori [disclosed], SWEP [inferred], Danfoss [disclosed], Kelvion [disclosed]
- 5.06a Pumps: Iwaki [disclosed], KSB [disclosed], Xylem [disclosed], Grundfos [inferred], Wilo [inferred]
- 5.06b Filtration: Eaton [inferred], Pall [inferred], Parker Hannifin [inferred]

### 5.07 Rear-door heat exchangers
https://gridtochip.site/parts/liquid-cooling-it-side-chip-to-cdu/rear-door-heat-exchangers
A heat exchanger mounted on the rear door of a server rack that cools exhaust air with chilled water or refrigerant before it leaves the rack, without cooling the chips directly.
Companies: Legrand [disclosed], Motivair [inferred], nVent [inferred], Vertiv [inferred], Coolcentric [disclosed], Huawei [inferred], Rittal [inferred], Stulz [inferred]

### 5.08 Immersion cooling systems
https://gridtochip.site/parts/liquid-cooling-it-side-chip-to-cdu/immersion-cooling-systems
Systems that submerge servers or trays directly in a dielectric fluid, either in open tanks (single-phase) or sealed chambers where the fluid boils and recondenses (two-phase), to cool the whole tray rather than individual dies.
Companies: Asperitas [disclosed], Firmus (Sustainable Metal Cloud) [inferred], GRC [disclosed], Hypertec [disclosed], Iceotope [disclosed], LiquidStack [inferred], Midas [disclosed], Submer [disclosed], TMGcore [inferred]

### 5.09 Fluids & coolant treatment
https://gridtochip.site/parts/liquid-cooling-it-side-chip-to-cdu/fluids-and-coolant-treatment
The coolant fluids that circulate through direct-to-chip and immersion loops, plus the chemical treatment (inhibitors, biocides) that keeps them stable over the life of the loop.
AI delta: The AI-driven shift of racks to direct-to-chip and immersion cooling created new, coolant-specific demand for glycol-water and dielectric fluid chemistries just as a major two-phase and dielectric fluid producer exits part of its line for PFAS-related reasons at the end of 2025.
- 5.09a Single-phase dielectric: Castrol [inferred], ExxonMobil [inferred], Lubrizol [inferred], Shell [inferred], Cargill [inferred], Engineered Fluids [inferred]
- 5.09b Two-phase: 3M [inferred], Chemours [inferred]
- 5.09c Direct-to-chip coolant (PG25) and inhibitors: BASF [inferred], Castrol [inferred], ChemTreat [inferred], Dow [inferred], Ecolab (Nalco) [inferred], Kurita [inferred], LyondellBasell [inferred], Shell [inferred]

### 5.10 Leak detection, flow & pressure sensing
https://gridtochip.site/parts/liquid-cooling-it-side-chip-to-cdu/leak-detection-flow-and-pressure-sensing
Sensors and systems that detect coolant leaks and monitor flow rate and pressure throughout the liquid-cooling loop, feeding alerts to rack and row controls.
AI delta: Liquid cooling becoming the default configuration for AI racks raised the number of leak, flow, and pressure monitoring points needed per rack, tracking the growth in quick disconnects and manifold branches.
Companies: Emerson [disclosed], Honeywell [inferred], Keyence [disclosed], nVent [inferred], Schneider Electric [inferred], Sensirion [disclosed], Vertiv [inferred], Endress+Hauser [inferred], IFM [inferred], TTK [inferred]

### 5.11 Rack/row cooling controls & software
https://gridtochip.site/parts/liquid-cooling-it-side-chip-to-cdu/rack-row-cooling-controls-and-software
Software and controllers that manage coolant setpoints, pump speed, and valve positions at the rack and row level, and that tie into facility-wide data center infrastructure management (DCIM) systems.
Companies: Motivair [inferred], nVent [inferred], Schneider Electric [inferred], Vertiv [inferred], CoolIT [inferred]

## Layer 6 — Facility cooling & HVAC

### 6.01 Chillers & compressors
https://gridtochip.site/parts/facility-cooling-and-hvac/chillers-and-compressors
Chillers use a vapor-compression or absorption cycle, built around a compressor, to produce chilled water for the facility's cooling loop.
AI delta: AI training and inference loads keep chiller plants at sustained near-full utilization instead of the intermittent duty conventional HVAC chillers are sized for, favoring larger, efficiency-optimized plants and pulling chiller lead times into the same order queue as other heavy mechanical and electrical equipment.
Companies: Carrier [disclosed], Daikin [disclosed], Gree [inferred], Johnson Controls [disclosed], LG [inferred], Midea [inferred], Mitsubishi Electric [inferred], Schneider Electric [inferred], Trane [disclosed], Vertiv [disclosed], Petra [inferred], Smardt [disclosed], Stulz [disclosed]
- 6.01a Compressors: Copeland [disclosed], Daikin [inferred], Mitsubishi Heavy Industries [inferred], Bitzer [disclosed], Danfoss [inferred]

### 6.02 Dry coolers, adiabatic/evaporative units, cooling towers, fans & motors
https://gridtochip.site/parts/facility-cooling-and-hvac/dry-coolers-adiabatic-evaporative-units-cooling-towers-fans-and
Dry coolers, adiabatic/evaporative units, and cooling towers reject heat picked up by the facility water loop to outside air, driven by fans and motors.
AI delta: Round-the-clock AI load raises the number of full-load hours this equipment runs and makes water-free or low-water dry and adiabatic designs more valuable at sites where water rights or drought risk constrain wet cooling towers.
Companies: Alfa Laval [inferred], Daikin [inferred], LU-VE [inferred], Modine [inferred], Munters [inferred], SPX Technologies [disclosed], Vertiv [inferred], Baltimore Aircoil [disclosed], Evapco [disclosed], Güntner [disclosed], Kelvion [inferred], Thermofin [inferred]
- 6.02a Fans and motors: ABB [disclosed], Hudson [inferred], Nidec [disclosed], Regal Rexnord [inferred], Siemens [disclosed], WEG [inferred], ebm-papst [disclosed], Multi-Wing [inferred], Ziehl-Abegg [disclosed]

### 6.03 CRAH / CRAC / fan walls / in-row units
https://gridtochip.site/parts/facility-cooling-and-hvac/crah-crac-fan-walls-in-row-units
CRAH, CRAC, fan-wall, and in-row units circulate conditioned air across IT equipment in air-cooled halls, using either facility chilled water (CRAH) or an internal refrigerant circuit (CRAC).
AI delta: Liquid cooling absorbs most of the chip heat load in AI racks, so CRAH and fan-wall demand per rack falls even as total site CRAH capacity keeps growing to serve the residual air-cooled load and act as backup.
Companies: Airedale [inferred], Daikin [inferred], Envicool [inferred], Johnson Controls [inferred], Kingspan [inferred], Mitsubishi Electric [inferred], Munters [inferred], Schneider Electric [inferred], Swegon [inferred], Systemair [inferred], Trane [disclosed], Vertiv [disclosed], Huawei [inferred], Nortek Data Center Cooling [disclosed], Rittal [disclosed], Stulz [inferred]

### 6.04 Air handling, economizers, filtration
https://gridtochip.site/parts/facility-cooling-and-hvac/air-handling-economizers-filtration
Air-handling units, outside-air economizers, and filtration equipment that condition and clean the air supplied to air-cooled data center halls.
AI delta: As liquid cooling takes over the chip heat load, air handling and economizer capacity is increasingly sized to the building envelope and residual equipment rather than to full IT load, shrinking its role relative to the facility water loop.
Companies: Daikin [disclosed], Donaldson [inferred], Johnson Controls [disclosed], Munters [inferred], Parker Hannifin [inferred], Systemair [disclosed], Trane [disclosed], Camfil [disclosed], Freudenberg [inferred], Mann+Hummel [inferred]

### 6.05 Facility water loop hardware (pumps, valves, piping, heat exchangers, tanks)
https://gridtochip.site/parts/facility-cooling-and-hvac/facility-water-loop-hardware-pumps-valves
The pumps, valves, piping, plate heat exchangers, and thermal-storage tanks that move and buffer water between chillers, heat-rejection equipment, and the CRAH or CDU interfaces on the IT side.
AI delta: Liquid-cooled AI racks add a second, often higher-temperature facility water loop to carry chip heat out to the CDU interface, increasing the pumps, piping, and heat-exchanger capacity a site needs beyond what an air-cooled building of the same size would require.
- 6.05a Pumps: KSB [inferred], Pentair [disclosed], Xylem [disclosed], Armstrong Fluid [disclosed], Grundfos [inferred], Wilo [inferred]
- 6.05b Valves and actuators: Belimo [inferred], Crane [disclosed], Emerson [disclosed], Honeywell [inferred], Johnson Controls [inferred], Mueller [inferred], Siemens [disclosed], Watts [inferred], Danfoss [inferred]
- 6.05c Piping: Georg Fischer [disclosed], Mueller Industries [inferred], Aquatherm [inferred], Charlotte Pipe [disclosed], Victaulic [disclosed]
- 6.05d Plate HX: Alfa Laval [disclosed], SWEP [inferred], Danfoss [inferred], Kelvion [disclosed]
- 6.05e Tanks and thermal storage: Caldwell [inferred], CST [inferred]

### 6.06 Water treatment, reuse, consumption
https://gridtochip.site/parts/facility-cooling-and-hvac/water-treatment-reuse-consumption
Chemical treatment, filtration, and reuse systems that condition make-up water and manage the water a facility's cooling system consumes or discharges.
AI delta: AI campuses' scale and continuous operation have made cooling-water consumption a public and regulatory issue in a way conventional data centers rarely triggered, pushing sites toward reclaimed or reuse water, treatment that extends closed-loop cycles, or water-free heat rejection.
Companies: ChemTreat [disclosed], DuPont [disclosed], Ecolab (Nalco) [inferred], Kurita [inferred], Pentair [inferred], Veolia [inferred], Watts [inferred], Xylem [disclosed], Gradiant [inferred]

### 6.07 Heat reuse & district heating interfaces
https://gridtochip.site/parts/facility-cooling-and-hvac/heat-reuse-and-district-heating-interfaces
Heat exchangers and interconnection equipment that export waste heat from a data center's cooling loop into a district heating network or other end use instead of rejecting it to atmosphere.
AI delta: Liquid-cooled AI racks run their facility loop hotter than legacy air-cooled halls, producing waste heat closer to the temperature district heating networks need and lowering the heat-pump work required to reuse it.
Companies: Alfa Laval [inferred], Carrier [inferred], Everllence [inferred], Fortum [disclosed], Siemens Energy [inferred], Trane [inferred], Danfoss [disclosed], Kelvion [disclosed], Qarnot [disclosed], Vattenfall [inferred]

### 6.08 Building management, controls, environmental sensors
https://gridtochip.site/parts/facility-cooling-and-hvac/building-management-controls-environmental-sensors
The building management system, PLCs, and environmental sensors that monitor and automatically control chillers, air handlers, pumps, and dampers across the facility.
AI delta: Continuous, high-density AI loads leave less thermal margin for error than conventional buildings, which has pulled a new category of AI-driven optimization software into the control layer to hold tighter setpoints without tripping the cooling plant.
Companies: ABB [inferred], Carrier [disclosed], Delta Electronics [disclosed], Distech [inferred], Honeywell [inferred], Johnson Controls [disclosed], Schneider Electric [inferred], Siemens [disclosed], Trane [disclosed], Vertiv [inferred]
- 6.08a PLCs: Rockwell [disclosed], Schneider Electric [inferred], Siemens [disclosed], Beckhoff [disclosed], Phoenix Contact [inferred]
- 6.08b Sensors: Belimo [inferred], Sensirion [disclosed], Vaisala [disclosed], Dwyer [inferred]
- 6.08c AI-driven cooling control: EkkoSense [inferred], Phaidra [disclosed], Vigilent [disclosed]

### 6.09 Fire detection & suppression
https://gridtochip.site/parts/facility-cooling-and-hvac/fire-detection-and-suppression
Smoke and heat detection systems and fire suppression equipment — agents, piping, and discharge nozzles — that protect the data hall and electrical rooms.
AI delta: Higher power density and the added electrical and coolant-leak surface of liquid cooling widen what detection and suppression systems need to cover per rack, while the discontinuation of 3M's Novec 1230 clean agent forces new designs onto alternative agents.
- 6.09a Detection: Hochiki [inferred], Honeywell [inferred], Johnson Controls [disclosed], Siemens [disclosed], Bosch [disclosed], Kidde Global Solutions [inferred], Securiton [inferred]
- 6.09b Suppression: Johnson Controls [disclosed], Siemens [disclosed], Fike [disclosed], Kidde Fire Systems [inferred], Marioff HI-FOG [disclosed], Minimax Viking [disclosed]
- 6.09c Agents: Chemours [inferred], inert gas from Air Liquide [inferred], 3M Novec 1230 [inferred]

## Layer 7 — Electrical: power distribution (grid to rack)

### 7.01 Large power transformers (345–765 kV) & substations
https://gridtochip.site/parts/electrical-power-distribution-grid-to-rack/large-power-transformers-345-765-kv-and-substations
Grid-scale transformers and switchyards that step transmission voltage down to what a data center campus can use, plus the substation civil works around them.
AI delta: lead times of 2–4 years; the single biggest schedule risk for GW campuses.
Supply status: constrained
Companies: CG Power [inferred], Efacec [inferred], GE Vernova [inferred], Hitachi Energy [disclosed], Hyosung Heavy [inferred], Hyundai Electric [inferred], Končar [inferred], LS Electric [inferred], Mitsubishi Electric [inferred], Prolec GE [inferred], Siemens Energy [disclosed], TBEA [inferred], WEG [inferred], Delta Star [inferred], Pennsylvania Transformer [disclosed], SGB-SMIT [inferred], Toshiba [inferred], Virginia Transformer [inferred]

### 7.02 HV switchgear, GIS, breakers, protection relays
https://gridtochip.site/parts/electrical-power-distribution-grid-to-rack/hv-switchgear-gis-breakers-protection-relays
The high-voltage switching, protection, and fault-isolation gear between the substation transformers and the campus's medium-voltage distribution, including gas-insulated (GIS) designs for compact sites.
AI delta: AI campuses ordering multiple large transformers at once are also multiplying HV switchgear and breaker orders, pulling on the same protection-relay and GIS supply chains already stretched by grid buildout.
Companies: ABB [inferred], Eaton [inferred], GE Vernova [inferred], Hitachi Energy [disclosed], Hyosung Heavy [inferred], Hyundai Electric [inferred], Mitsubishi Electric [inferred], Nari [inferred], Pinggao [inferred], Schneider Electric [inferred], Siemens Energy [inferred], Sieyuan [inferred]
- 7.02a Relays: ABB [inferred], GE Vernova [inferred], Schneider Electric [inferred], Siemens [inferred], Schweitzer Engineering Laboratories (SEL) [inferred]
- 7.02b SF6-free: ABB [inferred], GE Vernova [disclosed], Hitachi Energy [disclosed], Siemens Energy [disclosed]

### 7.03 MV switchgear, E-houses, power skids
https://gridtochip.site/parts/electrical-power-distribution-grid-to-rack/mv-switchgear-e-houses-power-skids
Medium-voltage switchgear packaged into pre-fabricated electrical houses (E-houses) or skids that distribute power across a campus and can be factory-built and shipped as modules.
AI delta: E-houses and power skids let a campus add pre-fabricated MV distribution capacity in parallel with site construction rather than waiting on stick-built electrical rooms, and AI campus buildout is now testing how fast factories can turn them around.
Companies: ABB [inferred], Eaton [inferred], GE Vernova [inferred], Hitachi Energy [inferred], Mitsubishi Electric [inferred], Nari [inferred], Powell [disclosed], Schneider Electric [inferred], Siemens [inferred], Vertiv [inferred], IEM [inferred], PCX [inferred]

### 7.04 MV/LV transformers (pad-mount, dry-type, cast-resin)
https://gridtochip.site/parts/electrical-power-distribution-grid-to-rack/mv-lv-transformers-pad-mount-dry-type
Step-down transformers that take medium-voltage distribution down to the low voltage used inside a building, available in pad-mount (outdoor, oil-filled) and dry-type or cast-resin (indoor) designs.
AI delta: Each new hall or power block needs its own set of these transformers, so campus buildout multiplies unit orders even though any individual transformer is smaller and faster to build than the units in 7.01.
Companies: ABB [inferred], Eaton [inferred], Fuji Electric [inferred], Hammond Power [disclosed], Hitachi Energy [inferred], Hyundai Electric [inferred], LS Electric [inferred], Prolec GE [inferred], Schneider Electric [inferred], Siemens [inferred], WEG [inferred], ERMCO [inferred], Howard Industries [inferred], Jinpan [inferred], MGM Transformer [disclosed], Olsun [disclosed], TMEIC [inferred], Virginia Transformer [inferred]

### 7.05 LV switchgear, switchboards, panelboards
https://gridtochip.site/parts/electrical-power-distribution-grid-to-rack/lv-switchgear-switchboards-panelboards
Low-voltage electrical distribution equipment — switchgear, switchboards, and panelboards — that receives power from MV/LV transformers and distributes it onward to busway, PDUs, and building loads.
AI delta: Larger, denser power blocks push individual LV switchboards toward higher ampacity ratings and tighter integration with UPS and generator controls than a conventional commercial building would need.
Companies: ABB [inferred], Chint [inferred], Eaton [inferred], Hyundai Electric [inferred], Legrand [inferred], Mitsubishi Electric [inferred], Powell [inferred], Schneider Electric [inferred], Siemens [inferred], IEM [inferred]

### 7.06 Busway / busbar trunking
https://gridtochip.site/parts/electrical-power-distribution-grid-to-rack/busway-busbar-trunking
Rigid, modular busbar systems that carry power along a data hall's overhead or perimeter path with tap-off points, used as an alternative to running individual cables to each row.
AI delta: Higher rack power draw favors busway over cable runs because busway supports higher-ampacity tap-offs and is easier to reconfigure as rack layouts and power densities change between generations.
Companies: ABB [inferred], Chint [inferred], Delta Electronics [inferred], Eaton [inferred], LS Electric [inferred], Schneider Electric [inferred], Siemens [inferred], Starline [disclosed], Vertiv [inferred], Graziadio [inferred]

### 7.07 Floor PDUs, RPPs, static transfer switches, ATS
https://gridtochip.site/parts/electrical-power-distribution-grid-to-rack/floor-pdus-rpps-static-transfer-switches-ats
Floor-mounted power distribution units, remote power panels, and the static and automatic transfer switches that route power between utility, UPS, and generator sources down to the rack level.
AI delta: Higher rack power draw is pushing floor PDU and RPP ratings up and shortening the distance these units sit from the rack, since longer low-voltage runs at high current become less efficient.
Companies: ABB [inferred], Caterpillar [inferred], Cummins [inferred], Delta Electronics [disclosed], Eaton [inferred], Generac [inferred], Legrand [inferred], Russelectric [inferred], Schneider Electric [inferred], Vertiv [inferred], Piller [inferred], Rehlko (Kohler Energy) [inferred]

### 7.08 Circuit protection, fuses, surge, grounding
https://gridtochip.site/parts/electrical-power-distribution-grid-to-rack/circuit-protection-fuses-surge-grounding
Fuses, circuit breakers, surge protection devices, and grounding systems that protect electrical equipment and personnel throughout the power chain, from switchgear down to the rack.
AI delta: Higher fault currents and ampacities at AI-density power levels push circuit protection toward higher-rated, more specialized fuses and breakers than a conventional commercial building needs.
Companies: ABB [inferred], Bel Fuse [inferred], Eaton [inferred], Littelfuse [inferred], Mersen [disclosed], nVent [inferred], Schneider Electric [inferred], Siemens [inferred], Harger [disclosed], Phoenix Contact [inferred]

### 7.09 LV cable, wire, tray, raceway
https://gridtochip.site/parts/electrical-power-distribution-grid-to-rack/lv-cable-wire-tray-raceway
Low-voltage power cable and wire, and the tray and raceway systems that route and support it between switchgear, PDUs, and racks.
AI delta: A single AI data hall requires a large volume of heavy-gauge low-voltage cable and tray to carry the higher currents that dense racks draw, and that demand now competes with grid buildout and electrification for the same copper supply chain.
Companies: Belden [inferred], Cerro/Marmon [inferred], Furukawa [inferred], Hengtong [inferred], LS Cable [inferred], Nexans [inferred], Prysmian [disclosed], Sumitomo Electric [inferred], Leoni [inferred], Okonite [inferred], Southwire [inferred]
- 7.09a Tray and supports: Atkore [disclosed], Eaton [inferred], Legrand [inferred], nVent [inferred], Chatsworth [disclosed], Hilti [inferred], Panduit [disclosed]

### 7.10 UPS (static, modular, rotary/DRUPS) & UPS power semis
https://gridtochip.site/parts/electrical-power-distribution-grid-to-rack/ups-static-modular-rotary-drups-and-ups-power-semis
Uninterruptible power supplies that bridge the gap between a utility outage and generator start, built as static (battery-based), modular, or rotary/diesel-rotary (DRUPS) designs, plus the power semiconductors inside them.
AI delta: GPU cluster power draw swings sharply and quickly with training and inference workload changes, and UPS systems are increasingly being asked to absorb those swings for grid stability rather than just bridge outages.
Companies: ABB [disclosed], Cyber Power [disclosed], Delta Electronics [inferred], Eaton [inferred], Fuji Electric [disclosed], Kehua [inferred], Legrand [inferred], Mitsubishi Electric [inferred], Schneider Electric [inferred], Vertiv [disclosed], Centiel [disclosed], Hitec [inferred], Huawei [inferred], Piller [disclosed], Rehlko (Kohler Energy) [inferred], Riello [disclosed], Socomec [inferred], Toshiba [disclosed]
- 7.10a IGBT and SiC modules: Fuji Electric [disclosed], Infineon [disclosed], Mitsubishi Electric [inferred], onsemi [inferred], Wolfspeed [disclosed], Semikron Danfoss [disclosed]

### 7.11 Energy storage: UPS batteries & BESS
https://gridtochip.site/parts/electrical-power-distribution-grid-to-rack/energy-storage-ups-batteries-and-bess
Battery systems that store energy for UPS ride-through, plus larger grid-scale battery energy storage systems (BESS) that a campus can use for peak shaving, load smoothing, or backup at a scale beyond a single UPS.
AI delta: AI campuses are pairing grid-scale BESS with their interconnection to smooth GPU load swings and to draw down stored energy during peak periods, a use case largely new to data centers rather than inherited from conventional IT backup power.
- 7.11a Li-ion for UPS: BYD [inferred], CATL [inferred], EnerSys [inferred], LG Energy Solution [inferred], Panasonic [inferred], Saft [inferred], Samsung SDI [inferred]
- 7.11b NiZn: ZincFive [disclosed]
- 7.11c VRLA: EnerSys [inferred], Narada [inferred], C&D [disclosed]
- 7.11d Sodium-ion: CATL [inferred], Natron [inferred]
- 7.11e Grid-scale BESS: BYD [inferred], Eos [disclosed], Fluence [disclosed], Jupiter Power [inferred], Sungrow [disclosed], Tesla [inferred], Wärtsilä [disclosed], Hithium [disclosed], Power [inferred], Powin [inferred]
- 7.11f PCS and inverters: Dynapower [inferred], SMA [inferred], Sungrow [inferred], EPC [inferred], Power [inferred]

### 7.12 Power smoothing / supercapacitors (GPU load-swing mitigation)
https://gridtochip.site/parts/electrical-power-distribution-grid-to-rack/power-smoothing-supercapacitors-gpu-load-swing-mitigation
Supercapacitor-based and firmware-coordinated systems that absorb the fast, large power swings synchronized GPU training and inference jobs create, protecting upstream electrical equipment and the grid from those transients.
AI delta: Synchronized GPU clusters can swing power draw by a large share of full load almost instantaneously as a training job pauses, checkpoints, or resumes, a response speed that batteries and generators alone are not built for.
Companies: Delta Electronics [inferred], Eaton [inferred], Vertiv [inferred], Skeleton Technologies [disclosed], UCAP Power [inferred]

### 7.13 Backup generation: gensets, fuel systems, HVO
https://gridtochip.site/parts/electrical-power-distribution-grid-to-rack/backup-generation-gensets-fuel-systems-hvo
Diesel or gas-fired backup generators, their fuel storage and delivery systems, and renewable diesel (HVO) as a lower-emission fuel alternative, providing power when utility and UPS ride-through are not enough.
AI delta: AI campus generator orders are large enough, and concentrated enough in time, that they compete for the same large-engine manufacturing capacity that marine, rail, and industrial customers use, lengthening lead times across all of those markets.
Companies: Caterpillar [inferred], Cummins [inferred], Everllence [inferred], Generac [disclosed], Mitsubishi Heavy Industries [disclosed], Perkins [inferred], Rolls-Royce [disclosed], Volvo Group (Penta) [inferred], Wärtsilä [inferred], Rehlko (Kohler Energy) [disclosed]
- 7.13a Day tanks and fuel systems: Preferred Utilities [inferred], Pryco [disclosed]
- 7.13b Aftertreatment: Johnson Matthey [inferred], Miratech [disclosed]
- 7.13c HVO: Diamond Green Diesel [inferred], Neste [disclosed]
- 7.13d Rental fleets: Herc [inferred], Sunbelt Rentals Holdings [disclosed], United Rentals [inferred], Aggreko [disclosed]

### 7.14 Paralleling switchgear & generator controls
https://gridtochip.site/parts/electrical-power-distribution-grid-to-rack/paralleling-switchgear-and-generator-controls
Switchgear and control systems that synchronize and combine the output of multiple generators so they can operate as one larger backed-up power source, and manage the transition between utility, UPS, and generator power.
AI delta: A gw_class campus's generator fleet can require far more paralleled units than a conventional building, making synchronization and load-sharing controls a proportionally harder engineering problem.
Companies: ASCO [inferred], Caterpillar [inferred], Cummins [inferred], Eaton [inferred], Russelectric [disclosed], Woodward [disclosed], Basler [disclosed], ComAp [disclosed], Deep Sea [disclosed], Rehlko (Kohler Energy) [inferred]

### 7.15 Metering, power quality, monitoring
https://gridtochip.site/parts/electrical-power-distribution-grid-to-rack/metering-power-quality-monitoring
Meters and power-quality monitoring equipment installed throughout the electrical chain to measure consumption, detect anomalies, and feed the data facility operators use to manage load and bill tenants.
AI delta: Fast GPU load swings make power-quality monitoring more important than in a conventional data center, since operators need to see transients that could stress upstream equipment (7.12) or trip protection devices (7.08) before they cause a problem.
Companies: ABB [inferred], Eaton [inferred], Fluke [inferred], Legrand [inferred], Schneider Electric [inferred], Siemens [inferred], Vertiv [inferred], Accuenergy [disclosed], Camlin [inferred], Elspec [disclosed], Janitza [disclosed], Packet Power [disclosed], Socomec [inferred]

### 7.16 800 V DC architecture & solid-state transformers (Rubin / Kyber era)
https://gridtochip.site/parts/electrical-power-distribution-grid-to-rack/800-v-dc-architecture-and-solid-state-transformers-rubin-kyber-e
An emerging DC power distribution architecture, using 800 V DC in place of the AC busbar and power-shelf approach current racks use, paired with solid-state transformers (SSTs) that could replace multiple stages of conventional AC transformation.
AI delta: Rack power draw is approaching levels where converting AC to DC multiple times between the grid and the chip becomes a meaningful efficiency and space loss, which is the design pressure pushing 800 V DC distribution and solid-state transformers onto the roadmap for the generations after current designs.
Companies: ABB [inferred], Coherent [inferred], Delta Electronics [inferred], Eaton [inferred], Flex [inferred], Hitachi Energy [inferred], Infineon [inferred], Lite-On [disclosed], Mitsubishi Electric [inferred], Navitas [inferred], onsemi [inferred], Rohm [disclosed], Schneider Electric [inferred], Siemens [inferred], STMicroelectronics [inferred], Texas Instruments [inferred], Vertiv [inferred], Wolfspeed [inferred]
- 7.16a SSTs (emerging): Delta Electronics [inferred], Heliox [inferred], Hitachi Energy [inferred], Ideal Power [inferred], Siemens Energy [inferred]

## Layer 8 — Power generation & energy supply

### 8.01 Gas turbines (onsite / behind-the-meter) & HRSGs
https://gridtochip.site/parts/power-generation-and-energy-supply/gas-turbines-onsite-behind-the-meter-and-hrsgs
Combustion turbines burning natural gas (or dual-fuel), paired with heat-recovery steam generators, used to generate power onsite or behind the meter at a data center campus.
AI delta: new-build turbine slots are sold out years ahead; refurbished and aeroderivative units command premiums.
Supply status: constrained
Companies: Baker Hughes [inferred], Doosan Enerbility [disclosed], GE Vernova [inferred], Kawasaki Heavy [disclosed], Mitsubishi Heavy Industries [inferred], Siemens Energy [inferred], Solar Turbines [inferred], Ansaldo Energia [disclosed], ProEnergy [inferred]
- 8.01a HRSGs: Vogt [disclosed], Cleaver-Brooks [disclosed], Nooter/Eriksen [inferred]

### 8.02 Reciprocating engines & engine microgrids
https://gridtochip.site/parts/power-generation-and-energy-supply/reciprocating-engines-and-engine-microgrids
Diesel, natural-gas, or dual-fuel reciprocating engines assembled into modular microgrids for onsite or bridge power at a data center site.
AI delta: AI campuses facing multi-year grid interconnection queues increasingly deploy reciprocating-engine microgrids as bridge power to energize racks before permanent utility service arrives.
Companies: Caterpillar [inferred], Cummins [inferred], Everllence [disclosed], Generac [inferred], Kawasaki Heavy [inferred], Mitsubishi Heavy Industries [disclosed], Rolls-Royce [inferred], Southern Company [inferred], Wärtsilä [inferred], Enchanted Rock [inferred], INNIO Jenbacher [disclosed], Mainspring [disclosed], VoltaGrid [inferred]

### 8.03 Fuel cells
https://gridtochip.site/parts/power-generation-and-energy-supply/fuel-cells
Fuel cells, chiefly solid-oxide units, that generate electricity electrochemically from natural gas or hydrogen without combustion, deployed as onsite or behind-the-meter power.
AI delta: Hyperscalers facing interconnection queues and emissions targets have signed onsite fuel-cell deals to energize AI campuses faster than grid upgrades would allow.
Companies: Ballard [disclosed], Bloom Energy [disclosed], Ceres [inferred], Doosan Fuel Cell [disclosed], FuelCell Energy [disclosed], Plug Power [disclosed]

### 8.04 Nuclear: existing fleet PPAs & restarts
https://gridtochip.site/parts/power-generation-and-energy-supply/nuclear-existing-fleet-ppas-and-restarts
Power purchase agreements with existing nuclear plant operators, including deals that fund restarts of previously shuttered reactors, to secure firm, carbon-free power for AI load.
AI delta: AI operators are contracting directly for existing nuclear output, including funding restarts of retired reactors, to lock in firm carbon-free power without waiting for new generation.
Companies: AEP [inferred], Ameren [inferred], Constellation [inferred], Dominion [inferred], Duke Energy [inferred], Entergy [inferred], NextEra [inferred], PSEG [inferred], Southern Company [inferred], Talen [inferred], Vistra [inferred], Xcel [inferred], Holtec/Palisades [inferred]

### 8.05 SMRs, advanced reactors, components, fuel
https://gridtochip.site/parts/power-generation-and-energy-supply/smrs-advanced-reactors-components-fuel
Small modular reactors and other advanced reactor designs, plus the components, fuel, and enrichment supply chain that would build and run them.
AI delta: AI operators have become the largest source of offtake commitments and equity capital for SMR and advanced reactor developers, betting on future firm carbon-free supply years ahead of any commercial operation.
- 8.05a Developers: GE Vernova [inferred], Nano Nuclear [disclosed], NuScale [disclosed], Oklo [inferred], Rolls-Royce [inferred], Westinghouse [disclosed], Aalo [inferred], Blue Energy [inferred], Deep Fission [inferred], Holtec SMR-300 [disclosed], Kairos [disclosed], Last Energy [inferred], Natura [inferred], Newcleo [inferred], Radiant [disclosed], TerraPower [disclosed], Terrestrial Energy [inferred], Valar [inferred], X-energy [disclosed]
- 8.05b Components: BWXT [disclosed], Curtiss-Wright [inferred], Doosan Enerbility [inferred], Flowserve [inferred], Graham [inferred], Mirion [inferred], Framatome [inferred]
- 8.05c Fuel and enrichment: Cameco [inferred], Centrus [disclosed], Lightbridge [inferred], Global Laser Enrichment [disclosed], Orano [inferred], Standard Nuclear [inferred], Urenco [inferred]
- 8.05d Uranium: Cameco [inferred], Denison [disclosed], enCore [disclosed], Energy Fuels [inferred], Kazatomprom [inferred], NexGen [disclosed], Uranium Energy [inferred]

### 8.06 Fusion (long-dated PPAs)
https://gridtochip.site/parts/power-generation-and-energy-supply/fusion-long-dated-ppas
Long-dated power purchase agreements with fusion developers, none of which have yet demonstrated commercial electricity generation.
AI delta: AI operators have signed long-dated fusion PPAs as options on future firm power, providing capital and demand signals to developers years before any commercial generation is proven.
Companies: Commonwealth Fusion [disclosed], General Fusion [disclosed], Helion [disclosed], Marvel Fusion [disclosed], Pacific Fusion [disclosed], TAE [disclosed], Tokamak Energy [disclosed], Type One [disclosed], Zap [disclosed]

### 8.07 Utilities, IPPs & power providers signing data-center deals
https://gridtochip.site/parts/power-generation-and-energy-supply/utilities-ipps-and-power-providers-signing-data-center-deals
Utilities, independent power producers, and grid operators that sign the power purchase agreements, interconnection contracts, and generation deals underlying a data center's supply.
AI delta: AI data centers have become some of the largest new individual customers many utilities and IPPs have signed in decades, reshaping interconnection queue priority and large-load rate design.
Companies: AEP [inferred], AES [inferred], Alliant [inferred], Ameren [inferred], Brookfield Renewable [inferred], Capital Power [inferred], CenterPoint [inferred], Clearway [inferred], CMS [inferred], Constellation [inferred], Dominion [inferred], DTE [inferred], Duke Energy [inferred], EDP Renewables [inferred], Emera [inferred], Enel [inferred], Engie [inferred], Entergy [inferred], Evergy [inferred], Exelon [inferred], FirstEnergy [inferred], Fortis [inferred], Iberdrola/Avangrid [inferred], NextEra [inferred], NRG [inferred], PG&E [inferred], Portland General [inferred], PPL [inferred], RWE [inferred], Sempra/Oncor [inferred], Southern Company [inferred], Talen [inferred], TotalEnergies [inferred], TransAlta [inferred], Vistra [inferred], WEC [inferred], Xcel [inferred], Ørsted [inferred], Apex [inferred], BC Hydro [inferred], BPA [inferred], Calpine [inferred], Cypress Creek [inferred], EDF [inferred], Hydro-Québec [inferred], Invenergy [inferred], OPG [inferred], Pattern [inferred], Statkraft [inferred], Tenaska [inferred], TVA [inferred], Vattenfall [inferred]

### 8.08 Renewables & long-duration storage hardware
https://gridtochip.site/parts/power-generation-and-energy-supply/renewables-and-long-duration-storage-hardware
Solar, wind, geothermal, and hydro generation hardware, plus long-duration energy storage, contracted or co-located to supply data-center power.
- 8.08a Solar: Canadian Solar [inferred], First Solar [disclosed], Hanwha Qcells [inferred], JA Solar [inferred], Jinko [disclosed], Longi [disclosed], Trina [disclosed]
- 8.08b Inverters: SMA [disclosed], Sungrow [inferred], Fimer [disclosed], Huawei [inferred], Power [inferred]
- 8.08c Trackers: Array [disclosed], Nextracker [inferred], GameChange [disclosed]
- 8.08d Wind: GE Vernova [disclosed], Goldwind [inferred], Nordex [inferred], Siemens Energy [inferred], Vestas [disclosed], Envision [inferred]
- 8.08e Geothermal: Ormat [inferred], Eavor [disclosed], Fervo [inferred], Quaise [disclosed], Sage Geosystems [disclosed], XGS [disclosed], Zanskar [inferred]
- 8.08f Hydro: Andritz [disclosed], Voith [disclosed]
- 8.08g LDES: Eos [inferred], ESS [disclosed], Antora [disclosed], Energy Dome [inferred], Form Energy [inferred], Hydrostor [disclosed]

### 8.09 Transmission, interconnection & grid equipment
https://gridtochip.site/parts/power-generation-and-energy-supply/transmission-interconnection-and-grid-equipment
High-voltage transmission equipment, interconnection hardware, grid-enhancing technology, and the EPC firms and developers that build and expand the grid feeding a data center site.
AI delta: AI data centers have made transmission capacity and interconnection queues a visible bottleneck, pulling forward demand for grid-enhancing technology and advanced conductors as faster alternatives to new transmission lines.
Companies: AESO [disclosed], CAISO [inferred], ERCOT [inferred], IESO [inferred], ISO-NE [disclosed], MISO [disclosed], NYISO [disclosed], PJM [disclosed], SPP [disclosed]
- 8.09a HVDC: Furukawa [inferred], GE Vernova [inferred], Hitachi Energy [disclosed], LS Cable [inferred], Mitsubishi Electric [inferred], Nexans [disclosed], NKT [disclosed], Prysmian [inferred], Siemens Energy [inferred], Sumitomo Electric [inferred]
- 8.09b STATCOM and synchronous condensers: ABB [inferred], AMSC [disclosed], GE Vernova [inferred], Hitachi Energy [inferred], Siemens Energy [inferred], WEG [inferred], Ingeteam [inferred]
- 8.09c Grid-enhancing tech: Heimdall [disclosed], LineVision [disclosed], Smart Wires [inferred]
- 8.09d Advanced conductors: Nexans [inferred], Prysmian [inferred], CTC Global [inferred], Southwire [inferred], TS Conductor [disclosed]
- 8.09e Poles and structures: Arcosa [inferred], Valmont [inferred], Falcon Steel [inferred], Sabre [inferred]
- 8.09f Transmission EPC: Dycom [inferred], MasTec [inferred], MYR [inferred], Primoris [inferred], Quanta Services [inferred], Black & Veatch [inferred], Burns & McDonnell [inferred], Kiewit [inferred], Power [inferred], Sargent & Lundy [inferred]
- 8.09g Developers: AEP [inferred], NextEra [inferred], Anbaric [inferred], Grid United [inferred], Invenergy [inferred], Pattern [inferred]

### 8.10 Gas supply chain (for onsite generation)
https://gridtochip.site/parts/power-generation-and-energy-supply/gas-supply-chain-for-onsite-generation
Pipelines, producers, compression equipment, LNG and virtual-pipeline logistics, and metering that supply natural gas to onsite generation.
AI delta: Onsite gas generation for AI campuses has added a new source of demand for pipeline capacity, compression equipment, and virtual-pipeline delivery, particularly at sites without existing pipeline access.
- 8.10a Pipelines and midstream: Boardwalk [inferred], Enbridge [inferred], Energy Transfer [inferred], Kinder Morgan [inferred], MPLX [inferred], ONEOK [inferred], Targa [inferred], TC Energy [inferred], Williams [inferred]
- 8.10b Producers: Antero [inferred], ARC [inferred], Chevron [inferred], Comstock [inferred], Coterra [inferred], Diversified [inferred], EQT [inferred], Expand Energy [inferred], ExxonMobil [inferred], Range [inferred], Tourmaline [inferred]
- 8.10c Compression and gas handling: Archrock [disclosed], Baker Hughes [inferred], Caterpillar [inferred], Kodiak Gas [inferred], Siemens Energy [disclosed], USA Compression [inferred], Ariel [inferred]
- 8.10d LNG and virtual pipeline: Chart [disclosed], New Fortress [disclosed], Stabilis [inferred]
- 8.10e Meters and regulators: Emerson [disclosed], Honeywell [inferred], Itron [inferred]

## Layer 9 — Building shell, site & construction

### 9.01 Data center developers, operators & REITs
https://gridtochip.site/parts/building-shell-site-and-construction/data-center-developers-operators-and-reits
The companies that own, develop, and operate the physical data center buildings AI workloads run in, from public REITs and colocation operators to private developers and bitcoin miners that have pivoted to AI hosting.
AI delta: AI campuses require single-site power and land commitments far larger than enterprise colocation, drawing new entrants — repositioned bitcoin miners, PE-backed platforms — into land-and-power development.
Companies: Accton (Edgecore) [inferred], AdaniConneX [inferred], AirTrunk [inferred], Applied Digital [inferred], Cipher [inferred], CleanSpark [inferred], Compass [inferred], Core Scientific [inferred], CyrusOne [inferred], Data4 [inferred], DataBank [inferred], Digital Edge [inferred], Digital Realty [inferred], Equinix [inferred], Fermi America [inferred], Galaxy [inferred], GDS [inferred], Green Mountain [inferred], Hut 8 [inferred], IREN [inferred], Iron Mountain [inferred], Keel Infrastructure [inferred], Keppel DC REIT [inferred], NEXTDC [inferred], Northern Data [inferred], NTT [inferred], QTS [inferred], Reliance/Jio [inferred], Riot [inferred], Sakura Internet [inferred], Scala [inferred], Sify [inferred], SoftBank [inferred], Soluna [inferred], STACK [inferred], Telehouse [inferred], TeraWulf [inferred], VNET [inferred], WhiteFiber/Bit Digital [inferred], Aligned [inferred], atNorth [inferred], Bulk [inferred], CloudHQ [inferred], Cologix [inferred], Colt [inferred], Crusoe [inferred], CtrlS [inferred], DataVolt [inferred], Echelon [inferred], EcoDataCenter [inferred], Flexential [inferred], G42 (Khazna) [inferred], Global Switch [inferred], Humain [inferred], Lancium [inferred], Novva [inferred], Nscale [inferred], Prime [inferred], Princeton Digital [inferred], Sabey [inferred], ST Telemedia GDC [inferred], Stream [inferred], Switch [inferred], T5 [inferred], Vantage [inferred], Verne [inferred], Virtus [inferred], Yondr [inferred], Yotta [inferred]

### 9.02 Hyperscalers, neoclouds & model labs (the tenants and owners)
https://gridtochip.site/parts/building-shell-site-and-construction/hyperscalers-neoclouds-and-model-labs-the-tenants-and-owners
The hyperscalers, neoclouds, and AI labs that occupy data center capacity as tenants or owner-operators, and whose capital spending ultimately drives demand for every layer below this one.
AI delta: AI training and inference workloads are the primary driver of the capital spending that funds every other layer of this taxonomy.
Companies: Alibaba [inferred], Alphabet (Google) [inferred], Amazon (AWS) [inferred], Apple [inferred], Baidu [inferred], KT [inferred], Meta Platforms [inferred], Microsoft [inferred], Naver [inferred], Oracle [inferred], Samsung SDS [inferred], SK Telecom [inferred], Tencent [inferred], ByteDance [inferred], G42 (Khazna) [inferred], Huawei [inferred], Humain [inferred], Stargate UAE [inferred]
- 9.02a Neoclouds: CoreWeave [inferred], DigitalOcean [inferred], IREN [inferred], Nebius [inferred], Cerebras [inferred], Crusoe [inferred], Firmus (Sustainable Metal Cloud) [inferred], Fluidstack [inferred], Groq [inferred], Lambda [inferred], Nscale [inferred], Together [inferred], Vultr [inferred]
- 9.02b Labs: OpenAI [inferred], Anthropic [inferred], Mistral [inferred], xAI [inferred]

### 9.03 Architecture, engineering & MEP design
https://gridtochip.site/parts/building-shell-site-and-construction/architecture-engineering-and-mep-design
Architecture and engineering firms that design a data center's building envelope and its mechanical, electrical, and plumbing (MEP) systems before construction begins.
AI delta: AI campuses require MEP designs for higher-voltage electrical distribution and liquid cooling loops that a decade of enterprise-colocation templates didn't anticipate.
Companies: AECOM [inferred], AtkinsRéalis [inferred], Fluor [inferred], Jacobs [inferred], Stantec [inferred], Sweco [inferred], Tetra Tech [inferred], WSP [inferred], Arup [inferred], Bechtel [inferred], Black & Veatch [inferred], Burns & McDonnell [inferred], Buro Happold [inferred], Corgan [inferred], Cundall [inferred], Deerns [inferred], Exyte [inferred], Gensler [inferred], HDR [inferred], kW Mission Critical [inferred], M Moser [inferred], Mott MacDonald [inferred], NV5 [inferred], Page [inferred], Ramboll [inferred], Salas O'Brien [inferred], Sargent & Lundy [inferred], Sheehan Nagle Hartray [inferred], Syska Hennessy [inferred], Ware Malcomb [inferred]

### 9.04 General contractors & design-build
https://gridtochip.site/parts/building-shell-site-and-construction/general-contractors-and-design-build
General contractors and design-build firms that construct the data center shell and coordinate the trade subcontractors once architecture and MEP design (9.03) are complete.
AI delta: AI campuses are larger, more repetitive across sites, and more schedule-driven than the enterprise colocation builds this industry has historically served.
Companies: Balfour Beatty [inferred], Bouygues [inferred], Eiffage [inferred], Ferrovial [inferred], Fluor [inferred], Fortis [inferred], Hochtief [inferred], Hyundai E&C [inferred], Kajima [inferred], Larsen & Toubro [inferred], Lendlease [inferred], Multiplex [inferred], Obayashi [inferred], Primoris [inferred], Samsung C&T [inferred], Shimizu [inferred], Skanska [inferred], Sterling Infrastructure [inferred], Strabag [inferred], Taisei [inferred], Turner [inferred], Vinci [inferred], Alberici [inferred], Bechtel [inferred], Big-D [inferred], Brasfield & Gorrie [inferred], Clark [inferred], Clayco [inferred], Clune [inferred], Dornan [inferred], DPR [inferred], Gammon [inferred], Gilbane [inferred], Hensel Phelps [inferred], HITT [inferred], Hoffman [inferred], Holder [inferred], JE Dunn [inferred], Jones Engineering [inferred], Kiewit [inferred], Kirby [inferred], Laing O'Rourke [inferred], Layton [inferred], Mace [inferred], McCarthy [inferred], Mercury [inferred], Mortenson [inferred], Pepper [inferred], Rogers-O'Brien [inferred], Ryan [inferred], Sisk [inferred], Structure Tone [inferred], Sundt [inferred], Swinerton [inferred], Takenaka [inferred], Walsh [inferred], Whiting-Turner [inferred], Winthrop [inferred], Yates [inferred], Zachry [inferred]

### 9.05 MEP & electrical contractors (self-perform)
https://gridtochip.site/parts/building-shell-site-and-construction/mep-and-electrical-contractors-self-perform
Self-performing MEP and electrical contractors who install the power distribution, cooling piping, and electrical systems specified by design (9.03) under the general contractor's coordination (9.04).
AI delta: Higher-density AI racks require more electrical capacity and more liquid-cooling piping per square foot than a comparable enterprise build, raising skilled-labor hours per project.
Companies: Comfort Systems [inferred], Cupertino Electric [inferred], EMCOR [inferred], IES Holdings [inferred], Limbach [inferred], Southland [inferred], Sturgeon [inferred], Bergelectric [inferred], Cache Valley Electric [inferred], Dornan [inferred], Dynamic Systems [inferred], Faith Technologies [inferred], Guarantee Electrical [inferred], Harris [inferred], Helix Electric [inferred], Jones Engineering [inferred], Kirby [inferred], Ludvik [inferred], McKinstry [inferred], Mercury [inferred], Miller Electric [inferred], MMR [inferred], Rosendin [inferred], TDIndustries [inferred], Winthrop [inferred]

### 9.06 Structure, steel, precast, envelope, site equipment
https://gridtochip.site/parts/building-shell-site-and-construction/structure-steel-precast-envelope-site-equipment
The raw structural and envelope inputs to a data center building — steel, precast concrete, cement and aggregates, wall and roof systems, raised floor, doors, and the heavy site equipment used to build it.
AI delta: AI campuses favor precast and modular structural systems, fabricated off-site and erected faster than site-built alternatives, on top of a large increase in overall construction volume.
- 9.06a Steel: Cleveland-Cliffs [inferred], Commercial Metals [disclosed], Gerdau [disclosed], Nucor [disclosed], Steel Dynamics [inferred], Schuff/DBM [inferred], Zekelman/Atlas Tube [disclosed]
- 9.06b Precast: Smith-Midland [disclosed], Clark [disclosed], Coreslab [disclosed], Gate [inferred], Metromont [disclosed], Tindall [disclosed], Wells [inferred]
- 9.06c Cement and aggregates: Cemex [disclosed], CRH [disclosed], Eagle Materials [disclosed], Heidelberg [disclosed], Holcim [inferred], Knife River [disclosed], Martin Marietta [disclosed], Vulcan [disclosed], Summit/Quikrete [disclosed]
- 9.06d Envelope: BlueScope/Butler [disclosed], Carlisle [disclosed], Holcim [disclosed], Kingspan [inferred], Nucor [disclosed], Owens Corning [disclosed]
- 9.06e Raised floor: Kingspan [inferred], ASM [inferred]
- 9.06f Doors and hardware: Allegion [disclosed], ASSA ABLOY [disclosed], Dormakaba [disclosed]
- 9.06g Site and civil: Granite [inferred], Sterling Infrastructure [inferred], Kiewit [inferred]
- 9.06h Equipment and rental: Caterpillar [inferred], Deere [inferred], Herc [disclosed], Manitowoc [inferred], Oshkosh [inferred], Sunbelt Rentals Holdings [disclosed], Terex [disclosed], United Rentals [disclosed]

### 9.07 Prefab / modular data centers & power/cooling skids
https://gridtochip.site/parts/building-shell-site-and-construction/prefab-modular-data-centers-and-power-cooling-skids
Factory-built modular data center units and prefabricated power and cooling skids that are assembled off-site and installed as complete, pre-integrated blocks.
AI delta: AI campuses' need for fast, repeatable deployment favors factory-built modular power and cooling systems that can be manufactured in parallel with site and shell construction.
Companies: ABB [inferred], Compass [inferred], Delta Electronics [inferred], Eaton [inferred], Nucor [inferred], Powell [inferred], Schneider Electric [inferred], Vertiv [disclosed], BladeRoom [disclosed], Cannon [inferred], Flexnode [inferred], Huawei [inferred], IEM [inferred], Nautilus [inferred], PCX [disclosed], Rittal [inferred]
- 9.07a Containerized "AI factories": Dell [inferred], HPE [inferred], Supermicro [inferred]

### 9.08 Site utilities, water & fiber entrance
https://gridtochip.site/parts/building-shell-site-and-construction/site-utilities-water-and-fiber-entrance
The utility connections a data center needs beyond electrical power — water supply and wastewater handling, on-site water storage and reuse, and the fiber and carrier entrances that connect it to the network.
AI delta: Liquid and evaporative cooling used in AI facilities increases water demand relative to older air-cooled designs, pushing water rights into the same siting-constraint category as power.
- 9.08a Water and wastewater: ADS [disclosed], American Water Works [disclosed], Essential Utilities [inferred], Mueller Water [inferred], Veolia [inferred], Xylem [disclosed], Forterra [inferred], JM Eagle [inferred]
- 9.08b Storage tanks: Caldwell [disclosed], CST [disclosed], Tank Connection [disclosed]
- 9.08c Onsite reuse and wells: Veolia [inferred], Gradiant [inferred]
- 9.08d Dark fiber and carrier entrance: Cogent [disclosed], Lumen [disclosed], Uniti [disclosed], Arelion [disclosed], Colt [inferred], euNetworks [inferred], Everstream [inferred], FiberLight [disclosed], GTT [inferred], Segra [disclosed], Zayo [disclosed]

### 9.09 Physical security & access
https://gridtochip.site/parts/building-shell-site-and-construction/physical-security-and-access
Physical access control, video surveillance, perimeter detection, and guarding systems that secure the facility and its data halls.
AI delta: AI buildout mainly adds volume — more sites needing a full security stack — rather than changing per-site security requirements.
Companies: Allegion [inferred], ASSA ABLOY [inferred], Axis [disclosed], Dormakaba [inferred], Hanwha Vision [disclosed], Honeywell [inferred], Horton [inferred], Johnson Controls [disclosed], Motorola/Avigilon [disclosed], Suprema [inferred], Boon Edam [inferred], Bosch [inferred], Brivo [disclosed], Genetec [inferred], Idemia [disclosed], Verkada [disclosed]
- 9.09a Perimeter intrusion detection: Senstar [disclosed], Magos [disclosed], Southwest Microwave [disclosed]
- 9.09b Guarding: Securitas [disclosed], Allied Universal [disclosed], GardaWorld [inferred]

### 9.10 Testing, commissioning & certification
https://gridtochip.site/parts/building-shell-site-and-construction/testing-commissioning-and-certification
Independent commissioning agents, load banks, and test equipment that verify a completed data center's power and cooling systems perform to specification before it goes live, along with the certification bodies that set the standards they test against.
AI delta: Higher-density, liquid-cooled AI facilities have more cooling-loop and power-delivery interdependencies to verify, requiring longer and more extensive commissioning cycles.
- 9.10a Cx agents: Bureau Veritas [inferred], Intertek [inferred], Hood Patterson & Dewar [inferred], Primary Integration [inferred], Salas O'Brien [inferred]
- 9.10b Load banks: Johnson Controls [inferred], Aggreko [inferred], Avtron [inferred], ComRent [inferred], Mosebach [inferred]
- 9.10c Test equipment: Doble [inferred], Fluke [inferred], Keysight [inferred], Megger [inferred], Omicron [inferred]
- 9.10d Certification and standards: UL Solutions [inferred], ASHRAE [inferred], Open Compute Project (OCP) [inferred], TIA [inferred], Uptime Institute [inferred], USGBC (LEED) [inferred]

## Layer 10 — Materials & upstream inputs

### 10.01 Copper (busbars, cable, cold plates, transformer windings)
https://gridtochip.site/parts/materials-and-upstream-inputs/copper-busbars-cable
Refined copper metal and copper products — wire, cable, sheet and strip — used as the primary electrical conductor and a thermal-transfer material throughout the data center.
AI delta: Denser racks carry more current at lower voltage and add direct-contact liquid cooling, both of which increase the amount of copper needed per rack and per megawatt of IT load.
- 10.01a Miners: Anglo American [inferred], Antofagasta [disclosed], BHP [inferred], First Quantum [disclosed], Freeport [disclosed], Glencore [disclosed], Ivanhoe [inferred], Lundin [inferred], Rio Tinto [inferred], Southern Copper [inferred], Teck [disclosed], Zijin [inferred], Codelco [inferred]
- 10.01b Refining and products: Aurubis [disclosed], Hindalco [inferred], Jiangxi Copper [inferred], Mitsubishi Materials [inferred], Mueller Industries [disclosed], Nexans [inferred], Prysmian [inferred], KME [inferred], Southwire [disclosed], Wieland [disclosed]

### 10.02 Aluminum (heat sinks, cable, structures)
https://gridtochip.site/parts/materials-and-upstream-inputs/aluminum-heat-sinks-cable
Refined aluminum metal and mill products — sheet, extrusion, casting — used for heat sinks, cable conductors, and structural and enclosure parts across the data center.
AI delta: AI racks still use aluminum for chassis, rails and structure, but the highest-power components shift heat removal to liquid cold plates, so aluminum's role moves from primary thermal path toward structural and secondary-cooling uses as density rises.
Companies: Alcoa [disclosed], Arconic [inferred], Century [inferred], Chalco [inferred], Constellium [disclosed], Kaiser [disclosed], Norsk Hydro [inferred], Novelis [disclosed], Rio Tinto [inferred], EGA [inferred]

### 10.03 Grain-oriented electrical steel & amorphous cores (transformers)
https://gridtochip.site/parts/materials-and-upstream-inputs/grain-oriented-electrical-steel-and-amorphous-cores-transformers
Grain-oriented electrical steel and amorphous metal alloy used to build the laminated cores inside power, distribution and instrument transformers.
AI delta: GOES supply is the input constraint behind transformer lead times.
Supply status: tight
Companies: ArcelorMittal [inferred], Baosteel [inferred], Cleveland-Cliffs [disclosed], JFE [inferred], Nippon Steel [disclosed], POSCO [inferred], Shougang [inferred], thyssenkrupp [disclosed], Proterial [disclosed]

### 10.04 Rare earths & permanent magnets (fans, motors, gensets, wind)
https://gridtochip.site/parts/materials-and-upstream-inputs/rare-earths-and-permanent-magnets-fans-motors
Rare-earth elements and the permanent magnets made from them, used in motors and fans throughout the facility, from server and cooling fans to generator and turbine components.
AI delta: Liquid cooling adds pump and CDU motors that air-cooled racks didn't need, and denser fan arrays add motor count, so AI infrastructure increases the number of rare-earth magnets drawn per rack even though data centers remain a small buyer relative to autos and wind.
Companies: Arafura [inferred], Energy Fuels [inferred], Iluka [inferred], JL Mag [inferred], Lynas [disclosed], MP Materials [disclosed], Neo Performance [disclosed], Northern Rare Earth [inferred], Shin-Etsu [inferred], TDK [inferred], USA Rare Earth [disclosed], Niron [inferred], Noveon [inferred], Proterial [inferred], VAC [disclosed]

### 10.05 Specialty glass & glass fiber (optical fiber, T-glass, glass cores)
https://gridtochip.site/parts/materials-and-upstream-inputs/specialty-glass-and-glass-fiber-optical-fiber-t-glass
Purpose-engineered glass and glass fiber — optical fiber, low-loss T-glass fabric, and glass-core material — used in network cabling and inside advanced chip packages.
AI delta: AI clusters run far more optical fiber per rack than general-purpose compute because of the added scale-out and scale-up network fabric, and the largest accelerator packages are moving to glass-core and T-glass substrates to control signal loss as package size grows.
Companies: AGC [disclosed], Corning [inferred], Jushi [inferred], Nippon Electric Glass [disclosed], Nittobo [disclosed], Owens Corning [inferred], Taishan [inferred], Schott [inferred]

### 10.06 Refrigerants & insulating gases
https://gridtochip.site/parts/materials-and-upstream-inputs/refrigerants-and-insulating-gases
Refrigerant chemicals used in cooling equipment and dielectric gases used to insulate and interrupt current inside medium- and high-voltage switchgear.
AI delta: Larger, denser electrical distribution behind AI campuses requires more medium- and high-voltage switchgear, and therefore more insulating gas, while facility-level heat rejection keeps refrigerant-based chillers in the loop even where the racks themselves are liquid-cooled.
Companies: Arkema [disclosed], Chemours [inferred], Daikin [inferred], Honeywell [inferred], Orbia/Koura [inferred], Syensqo [inferred]
- 10.06a SF6 alternatives: 

### 10.07 Industrial & specialty gases (fabs, lasers, suppression)
https://gridtochip.site/parts/materials-and-upstream-inputs/industrial-and-specialty-gases-fabs-lasers
Bulk and specialty industrial gases — nitrogen, oxygen, hydrogen, argon and others — used in chip fabrication, laser manufacturing, and fire suppression systems.
AI delta: AI accelerator, memory and networking-chip demand is driving fab capacity expansion, and each new or expanded fab brings a long-dated on-site industrial-gas supply contract with it.
Companies: Air Liquide [disclosed], Air Products [disclosed], Linde [disclosed], Taiyo Nippon Sanso [disclosed], Messer [inferred], SK Specialty [inferred]

### 10.08 Critical minerals: gallium, germanium, indium, tungsten, beryllium
https://gridtochip.site/parts/materials-and-upstream-inputs/critical-minerals-gallium-germanium-indium-tungsten-beryllium
A group of minor metals — gallium, germanium, indium, tungsten and beryllium — used in semiconductors, optics, thermal interface materials and specialty alloys, with supply concentrated in China and subject to export controls.
AI delta: AI hardware adds a growing draw on this group through compound-semiconductor power devices, optics and thermal-interface formulations, layering additional demand onto a supply base already concentrated and subject to export licensing.
Companies: 5N Plus [disclosed], Almonty [disclosed], Materion [disclosed], Teck [inferred], Umicore [disclosed], Indium [disclosed], Nyrstar [inferred]

## Layer 11 — Software, tools & services touching the hardware (optional)

### 11.01 DCIM, digital twins, capacity planning
https://gridtochip.site/parts/software-tools-and-services-touching-the-hardware-optional/dcim-digital-twins-capacity-planning
Software that models a data center's power, cooling, space, and asset capacity in a live digital twin, giving operators visibility into what a facility can still support.
AI delta: Denser AI racks, mixed liquid and air cooling, and faster fit-out timelines increase the complexity DCIM and digital-twin tools are asked to manage.
Companies: Autodesk [inferred], Bentley [inferred], Cadence [disclosed], Carrier [inferred], Device42 [disclosed], Nvidia [inferred], Schneider Electric [inferred], Siemens [inferred], Vertiv [inferred], EkkoSense [inferred], Hyperview [disclosed], RF Code [disclosed], Sunbird [disclosed]

### 11.02 Cluster management & GPU orchestration (out of scope for a parts site; listed for completeness)
https://gridtochip.site/parts/software-tools-and-services-touching-the-hardware-optional/cluster-management-and-gpu-orchestration-out-of-scope-for-a-part
Software that schedules AI workloads onto GPU clusters and manages the compute fleet, sitting outside the physical parts this site otherwise tracks.
AI delta: Larger, more expensive GPU clusters raise the cost of scheduling and utilization mistakes, increasing demand for orchestration software built specifically for AI training and inference.
Companies: Dell [inferred], HPE [inferred], Nvidia [disclosed], Supermicro [inferred], Rafay [disclosed], SchedMD/Slurm [disclosed]

### 11.03 Site selection, power procurement, market intelligence, brokerage
https://gridtochip.site/parts/software-tools-and-services-touching-the-hardware-optional/site-selection-power-procurement-market-intelligence-brokerage
Firms and platforms that help developers pick sites, secure power, and read market conditions before and during a data center build.
AI delta: Power availability, not land or building cost, is increasingly the binding constraint on where AI data centers can be sited, raising the value of power-market intelligence in site selection.
Companies: CBRE [inferred], Colliers [inferred], Cushman & Wakefield [inferred], JLL [inferred], Newmark [inferred], Baxtel [inferred], datacenterHawk [inferred], DC Byte [inferred], Enverus [inferred], Grid Strategies [inferred], LevelTen [inferred], SemiAnalysis [inferred], Wood Mackenzie [inferred]

### 11.04 Capital & financing (context)
https://gridtochip.site/parts/software-tools-and-services-touching-the-hardware-optional/capital-and-financing-context
The private equity firms, credit funds, banks, and specialist lenders that finance data center construction and equipment purchases, provided here as context rather than as a hardware part.
AI delta: The scale of AI data center and GPU capital expenditure has pulled in project finance, private credit, and asset-backed lending structures at a scale prior data center cycles did not require.
Companies: Apollo [inferred], Ares [inferred], Bank of America [inferred], BlackRock [inferred], Blackstone [inferred], Blue Owl [inferred], Brookfield [inferred], Carlyle [inferred], Deutsche Bank [inferred], DigitalBridge [inferred], Goldman Sachs [inferred], JPMorgan [inferred], KKR [inferred], Macquarie [inferred], MGX [inferred], Morgan Stanley [inferred], MUFG [inferred], PIMCO [inferred], SMBC [inferred], TPG [inferred], Stonepeak [inferred]
- 11.04a GPU-backed lending: Blackstone [inferred], Brookfield [inferred], Coatue [inferred], Magnetar [inferred]

### 11.05 Test & measurement instruments
https://gridtochip.site/parts/software-tools-and-services-touching-the-hardware-optional/test-and-measurement-instruments
Test and measurement instruments — oscilloscopes, network analyzers, power analyzers, and related gear — used to validate silicon, cabling, and power systems through build and commissioning.
AI delta: Higher-speed interconnect and higher-density power systems in AI racks require tighter test tolerances and more specialized instruments than prior generations of data center hardware.
Companies: Anritsu [disclosed], Chroma [disclosed], Emerson [disclosed], Fortive [inferred], Hioki [disclosed], Keysight [inferred], Teradyne [inferred], Viavi [disclosed], Yokogawa [inferred], EXFO [disclosed], Rohde & Schwarz [disclosed]

### 11.06 Robotics, inspection & operations hardware
https://gridtochip.site/parts/software-tools-and-services-touching-the-hardware-optional/robotics-inspection-and-operations-hardware
Robots and inspection hardware used for physical operations tasks inside and around data center facilities, such as monitoring and inspection rounds.
AI delta: Denser, higher-value AI racks raise the cost of an undetected thermal or power fault, adding to the case for continuous robotic inspection alongside fixed sensors.
Companies: Boston Dynamics [disclosed], Ghost Robotics [inferred], Teledyne [inferred], Levatas [inferred]