130 parts · 1,202 companies
Prices as of2026-10-25
Search
1.04

Who makes startup accelerators for AI data centers

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.

This is the long tail of accelerator design: companies building chips with fundamentally different architectures from merchant GPUs (Merchant GPUs / accelerators) or hyperscaler ASICs (Hyperscaler custom ASICs), usually pitched on cost, latency, or efficiency advantages for a specific workload like inference serving. They still depend on the same downstream chain as everyone else — advanced-node foundry capacity and advanced packaging — so their unit economics are exposed to the same choke points as the merchant vendors they're trying to displace, without the volume to negotiate as favorable terms. Almost all of these companies are private, so public-market exposure is indirect: through the foundries and packaging houses that fab and package their chips, or through eventual IPOs. The investor question is less "who wins" than "which workload segment, if any, actually needs a different architecture than a merchant GPU provides" — a smaller and harder-to-answer question than it looks, since qualification with a customer's software stack is as much the barrier to adoption as raw chip performance, and most AI software has been built and tuned against the incumbent architecture first.

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.

Appears at:Enterprise / small coloLarge colo / hyperscale buildingAI campusGW-class campus· cooling:any — Deployment tends to concentrate with whichever customer or cluster has qualified a given architecture, rather than spreading evenly; presence at any tier depends on that customer's buying decision.

Companies on this part (11)

CompanyRoleOwnershipTickerPrice1dConfidenceNote
Graphcore · unit of SoftBankmanufacturersubsidiary9984.T*inferred
Cerebrasmanufacturerprivateinferred
d-Matrixmanufacturerprivateinferred
Etchedmanufacturerprivateinferred
Furiosamanufacturerprivateinferred
Groqmanufacturerprivateinferred
MatXmanufacturerprivateinferred
Positronmanufacturerprivateinferred
Rebellionsmanufacturerprivateinferred
SambaNovamanufacturerprivateinferred
Tenstorrentmanufacturerprivateinferred

Who competes here

The same table read the other way: every company above competes for startup accelerators sockets. Confidence tags matter — an inferred placement is a lead, not a confirmed rivalry.

Nearby parts

From the blog

Latest feed

Questions about startup accelerators

What are startup accelerators?
Non-GPU accelerator architectures from mostly-private startups, built as an alternative to merchant GPUs for specific AI training or inference workloads.
Which companies make startup accelerators for AI data centers?
This catalog lists 11 companies on Startup accelerators, 1 of them with listed equity: Graphcore, Cerebras, d-Matrix, Etched, Furiosa, Groq, MatX, Positron, Rebellions, SambaNova, and 1 more.
How does AI change demand for startup accelerators?
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.
How reliable are these startup accelerators supplier placements?
The 11 placements on this part break down as 11 inferred. "disclosed" means the company or a filing says so; "reported" comes from trade press or a reference-design list; "inferred" is a taxonomy lead with no citation yet.

Sources: placements without a source URL come from the taxonomy seed and are tagged accordingly. How this data is built.· Markdown · JSON