Who makes rack-scale systems for AI data centers
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.
The rack-scale system is the top-level assembly layer: it takes the GPU baseboards (4.03), switch trays (4.04), power shelves (4.12) and liquid-cooling hardware from layer 5 and integrates them into a single cabinet wired for one scale-up fabric domain (layer 3). Nvidia and AMD publish the reference architecture and mechanical/electrical spec; the ODMs and OEMs listed against this part build to that spec, badge the result, and compete mainly on integration quality, lead time and support rather than on the underlying design. AI training and inference clusters push this part into existence at all: a conventional enterprise server rack has no equivalent, because the point of a rack-scale system is to turn dozens of GPUs into one coherent domain, something only large models require. For investors, the interesting split is between the reference-design owners (Nvidia, AMD), who capture the architecture value, and the ODM/OEM tier, which competes on manufacturing scale and qualification speed with each new rack generation. Each new generation resets the qualification race, so watch which ODMs get design wins early and which fall behind on ramp.
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 on this part (23)
| Company | Role | Ownership | Ticker | Price | 1d | Confidence | Note |
|---|---|---|---|---|---|---|---|
| AMD | designer | public | AMD | 477 USD | -1.7% | inferred | AMD design arm; manufacturing → Sanmina SANM; 2025 |
| ASRock Rack | oem | public | 3515.TW | 219 TWD | +0.5% | inferred | |
| ASUS | oem | public | 2357.TW | 957 TWD | -1.5% | inferred | |
| Cisco | oem | public | CSCO | 112 USD | -0.1% | inferred | |
| Compal | odm | public | 2324.TW | 40.10 TWD | +1.1% | inferred | |
| Dell | oem | public | DELL | 472 USD | +0.0% | inferred | |
| Eviden/Atos | oem | public | ATO.PA | 28.52 EUR | 0.0% | inferred | |
| Foxconn (Hon Hai) | odm | public | 2317.TW | 253 TWD | +0.2% | inferred | Ingrasys |
| Fujitsu | oem | public | 6702.T | 3938 JPY | +4.7% | inferred | |
| Gigabyte | oem | public | 2376.TW | 348 TWD | +0.4% | inferred | |
| H3C · unit of Unisplendour | oem | subsidiary | 000938.SZ* | inferred | |||
| HPE | oem | public | HPE | 54.41 USD | -3.7% | inferred | |
| Inspur | oem | public | 000977.SZ | 79.10 CNY | +1.1% | inferred | |
| Inventec | odm | public | 2356.TW | 65.00 TWD | -0.9% | inferred | |
| Lenovo | oem | public | 0992.HK | 30.84 HKD | +0.4% | inferred | |
| MiTAC (Tyan) | odm | public | 3706.TW | 91.30 TWD | -0.3% | inferred | |
| NEC | oem | public | 6701.T | 4955 JPY | +5.2% | inferred | |
| Nvidia | designer | public | NVDA | 228 USD | +3.8% | inferred | |
| Pegatron | odm | public | 4938.TW | 88.80 TWD | -0.7% | inferred | |
| Quanta Computer | odm | public | 2382.TW | 332 TWD | -0.1% | inferred | |
| Supermicro | oem | public | SMCI | 38.46 USD | +0.3% | inferred | |
| Wiwynn · unit of Wistron | odm | public | 6669.TW | 7020 TWD | +3.5% | inferred | |
| xFusion | oem | private | inferred |
Who competes here
The same table read the other way: every company above competes for rack-scale systems sockets. Confidence tags matter — an inferred placement is a lead, not a confirmed rivalry.
Nearby parts
- Next in layer: EMS, system integration & rack-level test
- Layer: Layer 4 — Servers, racks & integration
Watch
Independent videos — no affiliation; nothing loads from YouTube until you click.
Latest feed
- Analysis: Nvidia beats, Taiwan wins — and hyperscalers race to make AI pay — DigiTimes, 2026-08-28Nvidia reported massive revenue growth, driven by the global expansion of AI infrastructure and increased purchases from hyperscalers.
- OpenAI Details Jalapeño Inference Chip Performance and Multigenerational Roadmap — HPCwire, 2026-08-27OpenAI announced the performance of its custom inference chip, Jalapeño, demonstrating improvements in both throughput and latency.
- EuroHPC Extends EuroCC and CASTIEL Projects to Strengthen European HPC Network — HPCwire, 2026-08-27The EuroHPC JU extended its funding for the EuroCC and CASTIEL projects to bolster Europe's High Performance Computing infrastructure.
- AWS and NVIDIA to Deliver 2M Additional GPUs and Next-Gen Infrastructure for Agentic and Physical AI — HPCwire, 2026-08-27Amazon Web Services and NVIDIA announced a major expansion of their collaboration to deploy two million additional GPUs and advanced infrastructure for AI workloads.
- Nvidia expects to sell $20 billion of Vera Rubin systems in Q3 as shipments begin — figure would account for 20% of its data center revenue mix, marks fastest ramp in company history — Tom's Hardware, 2026-08-27Nvidia anticipates that sales of Vera Rubin systems will represent a rapid and significant increase, contributing 20% of its data center revenue mix in the third fiscal quarter.
- AWS to deploy 2 million additional Nvidia GPUs — Data Center Dynamics, 2026-08-27Amazon Web Services is planning to deploy two million extra Nvidia GPUs into their infrastructure.
- Nvidia revenue doubles over last year, as company forecasts record growth — Data Center Dynamics, 2026-08-27Nvidia's revenue is expected to increase significantly, driven by high profitability per unit of data center power.
- Hot Chips 2026: Nvidia presents Groq 3 LPX architecture and unveils its first third-party inference benchmark — LP30-based rack already in production, company says — Tom's Hardware, 2026-08-26Nvidia's VP of hardware presented the architecture of the Groq 3 LPX rack and released a third-party benchmark for its performance.
- Hot Chips 2026: Nvidia touts benefits of its DSX MaxLPS site power management approach — tech allows for more compute from fixed data center power budgets — Tom's Hardware, 2026-08-26Nvidia presented the Rubin GPU and highlighted that power remains a major constraint on data center capacity, demonstrating compute efficiency within a fixed 100MW power budget.
- AMD and Anthropic Announce Strategic Partnership to Deploy Up to 2 Gigawatts of AMD Instinct MI450 Series GPUs — newsroom:amd, 2026-07-22AMD and Anthropic are partnering to deploy up to 2 gigawatts of AMD Instinct MI450 series GPUs.
Sources: placements without a source URL come from the taxonomy seed and are tagged accordingly. How this data is built.· Markdown · JSON