Orders for Cerebras Systems' IPO have reportedly been covered more than 20 times two days before pricing. Reuters and Bloomberg reported on May 10 that the company now plans to raise its range from $115-$125 to $150-$160 and increase the deal size from 28 million to 30 million shares.
At the top of the new range, Cerebras could raise $4.8 billion, about $1.3 billion more than the original plan. Formal pricing is expected on May 13, with CBRS set to list on Nasdaq.
Why investors want this chip company
Cerebras is not trying to copy Nvidia's GPU model. Its pitch is wafer-scale computing: instead of cutting a silicon wafer into many smaller chips, Cerebras turns the whole wafer into one processor.
The flagship WSE-3 integrates 44GB of SRAM on chip, has 900,000 cores, offers single-cycle memory access and sits inside a CS-3 system weighing about 1.8 tons.
For inference workloads that normally move data across GPU clusters, putting memory and compute on one enormous chip can remove a costly layer of inter-chip communication. For some low-latency inference tasks, that architecture can be dramatically faster than GPUs.
That explains why OpenAI signed a $20 billion exclusive compute agreement with Cerebras last year, and why AWS is bringing the technology into Bedrock.
The financials look different from many AI chip stories
Many AI chip IPOs are built around large losses and a future-market narrative. Cerebras' filing is unusually concrete.
| Metric | 2024 | 2025 |
|---|---|---|
| Revenue | ~$165M | $290.3M |
| Growth | - | +76% YoY |
| Net income | -$485M | +$87.9M |
The company moved from a $485 million loss to an $87.9 million profit in a year, while revenue rose 76%. Its two anchor customers, OpenAI and AWS, are purchasing production compute rather than running small proofs of concept.
What 20x demand says
A book covered 20 times before pricing is rare. Recent hot technology listings have often landed closer to three to five times covered, while major AI-linked offerings have typically been lower than Cerebras' reported level.
The underwriting group is also a signal: Morgan Stanley, Citi, Barclays and UBS are joint leads, a lineup more often seen in very large strategic offerings than in ordinary semiconductor deals.
If Cerebras prices at $160 and raises $4.8 billion, its market value could open near the $40 billion range. A sharp first-day move could push intraday value materially higher.
Nvidia rival, or more proof of AI compute demand?
In the short run, the market may read Cerebras' demand as good news for the whole AI compute supply chain. Nvidia, AMD, Cerebras and Groq are all serving a market that still looks capacity constrained.
Longer term, the map is splitting. Nvidia remains dominant in training because of CUDA and its ecosystem. Inference is more open, with Cerebras, Groq and AWS Trainium each pushing for specific workloads. Edge inference remains a separate arena for Qualcomm, Apple and other device players.
OpenAI's $20 billion Cerebras agreement was explicitly focused on inference first. That is the market Cerebras is betting on, and it is also where Nvidia's moat is thinner than in training.
Sources: Cerebras to raise IPO price range to $150-$160 as demand surges (CNBC), Cerebras to Boost Price of IPO to as High as $160 (Bloomberg), CocoLoop, AI chipmaker Cerebras to increase IPO price target (SiliconANGLE)