Cerebras builds one chip per wafer and keeps weights in on-chip SRAM, which is why it leads public speed charts for inference: about 2,700 tokens per second on GPT-OSS-120B in Artificial Analysis tests. It listed on Nasdaq in May 2026 in the year's largest US IPO, and OpenAI has contracted 750 MW of Cerebras capacity through 2028.
Key Takeaways
- What it is — Inference specialist. A single chip the size of a whole silicon wafer, with model weights held in on-chip SRAM.
- Best for — The fastest tokens per second on open-weight models, for agents and interactive products where latency is the product.
- Watch out — Model catalogue is limited to what Cerebras hosts, and customer concentration with OpenAI is high.
At a glance
| Company | Cerebras Systems (Nasdaq: CBRS) |
|---|---|
| Current chips | Wafer-Scale Engine 3 (WSE-3) in the CS-3 system |
| Architecture | A single chip the size of a whole silicon wafer, with model weights held in on-chip SRAM |
| Memory | 44 GB of on-chip SRAM per CS-3; larger models are spread across several systems |
| Scale-up | Clusters of CS-3 systems; each CS-3 rated at 12.5 FP16 petaflops dense and about 23 kW |
| Software | OpenAI-compatible inference API for hosted open-weight models; training stack for dedicated customers |
| Where to get it | Cerebras Cloud API, dedicated capacity, and on-premises systems |
| Pricing | Per token. GPT-OSS-120B listed at $0.25 per million input and $0.69 per million output tokens (January 2026) |
What changed in 2026
- Jan 2026. OpenAI signs a $10 billion, 750 MW inference deal with Cerebras, scaling through 2028.
- May 2026. Cerebras lists on Nasdaq as CBRS at $185 a share, raising about $6.38 billion gross.
Scores
Scored 1 to 5 on the same five axes as every vendor in the AI chip companies analysis.
| Performance | Scale-up | Software | Access | 2026 momentum | Total / 25 |
|---|---|---|---|---|---|
| 5 | 3 | 3 | 4 | 5 | 20 |
Verdict
The pick when output speed matters more than model choice. Use it as an inference tier next to your GPU or TPU estate, not as a replacement for it.
Alternatives
In the same lane (inference specialist):
- Groq: low-latency, low-cost inference on popular open-weight models through a simple API.
- SambaNova: enterprises and governments that want fast inference on their own premises with many models per rack.
In other lanes:
- Google TPU (hyperscaler custom silicon)
- AWS Trainium (hyperscaler custom silicon)
- AMD Instinct (merchant gpu)
- Tenstorrent (open hardware)
- Microsoft Maia (hyperscaler custom silicon)
Head-to-head comparisons
- Google TPU vs AWS Trainium vs Nvidia GPU: The three ways to train and serve frontier-scale models in 2026, scored on scale-up domain, software, access and price.
- Cerebras vs Groq vs SambaNova: The SRAM-first inference specialists compared on speed, token price, deployment options and company risk.
New to custom AI silicon? Our partner site CTAIO explains how TPUs, pods and GPU clusters differ in TPU vs GPU.
Frequently asked questions
What is Cerebras?
Cerebras builds one chip per wafer and keeps weights in on-chip SRAM, which is why it leads public speed charts for inference: about 2,700 tokens per second on GPT-OSS-120B in Artificial Analysis tests. It listed on Nasdaq in May 2026 in the year's largest US IPO, and OpenAI has contracted 750 MW of Cerebras capacity through 2028.
How do you get access to Cerebras?
Cerebras Cloud API, dedicated capacity, and on-premises systems. Pricing: Per token. GPT-OSS-120B listed at $0.25 per million input and $0.69 per million output tokens (January 2026).
What software does Cerebras run?
OpenAI-compatible inference API for hosted open-weight models; training stack for dedicated customers.
Is Cerebras a good alternative to Nvidia?
The pick when output speed matters more than model choice. Use it as an inference tier next to your GPU or TPU estate, not as a replacement for it. It fits best for: the fastest tokens per second on open-weight models, for agents and interactive products where latency is the product. The main caveat: model catalogue is limited to what Cerebras hosts, and customer concentration with OpenAI is high.
Sources
Facts verified 2026-10-31. Vendor performance figures are as stated by the vendor.
Ready to Find the Right AI Tools?
Browse our data-driven rankings to find the best AI tools for your team.