Cerebras went public last week with the stock nearly doubling on day one from the initial buy-ins. With their market cap touching $100 billion, investors have started to question if they could be the next Nvidia and with that, thoughts of buying in. I’ve taken a keen interest in this company due to my position in AMD and while the surface metrics look attractive, I believe there is actually more going on that investors should be aware of before committing.
Here’s what I’ve managed to put together.
(Quick disclosure so it’s clear where I stand: I don’t own Cerebras and I’m not planning to. Not shorting either. The AI play to me seems more ripe for profit-taking at the moment more than anything else. Read on and make your own call.)
CBRS 15-minute, Day 1 + Day 2 of Trading
What exactly does Cerebras do?
Cerebras has 2 core businesses one is the hardware which is chip manufacturing. The other is the cloud business where they lease out compute.
So if you’re a customer, you can either buy the chips or you can save yourself the trouble (and capex) and lease the compute provided by the chips.
On the chips front, Cerebras builds AI chips, but not the way Nvidia does.
Geek moment here (skip if you are not into the technicalities). Nvidia takes a silicon wafer (the round, dinner-plate-sized disc you’ve seen in factory tour videos) and cuts it into hundreds of small rectangles. Each rectangle becomes one GPU. To train a large AI model, you wire thousands of these GPUs together.
Cerebras asked a different question. Why cut? Use the entire wafer as one chip. Their Wafer Scale Engine 3 (WSE-3) packs 4 trillion transistors and 900,000 cores onto a single piece of silicon the size of a dinner plate. It’s by far the largest computer chip ever made. As everything sits on one piece of silicon, data doesn’t have to travel between separate chips. That makes it very fast for AI inference (the part where ChatGPT actually answers your question, as opposed to training the model in the first place).
The Fundamentals & the Bull Case
Cerebras annual revenue (USD millions)= $24.6m to $510m
2022 to 2025, roughly 20x in three years
Roughly 20x revenue growth in three years. That’s the number doing most of the work in the bull case.Margins are also moving in the right direction. Gross margins climbed from around 12% in 2022 to the low 40s in 2025. The company is still loss-making (operating loss of roughly $146 million in 2025), but the direction of travel is real.
On top of that, two major deals:
- OpenAI: a multi-year contract worth more than $20 billion for inference capacity
- AWS: deployment of Cerebras CS-3 systems inside Amazon Bedrock
Total remaining performance obligations (RPO, basically contracted revenue not yet delivered): $24.6 billion, against $510 million of trailing revenue. A 48x backlog-to-revenue ratio is genuinely unusual for a chip company at this stage.
Overall, the bulls aren’t being unreasonable. Their argument, briefly:
- Inference is the bigger market than training. Once a model is trained, every prompt to ChatGPT, Claude, or Gemini is an inference call. That market is growing fast, and Cerebras is purpose-built for it.
- OpenAI doesn’t sign $20 billion contracts lightly. OpenAI launched its first model running on Cerebras chips earlier this year. That’s a live product dependency, not just a signed PDF. Product dependencies are much harder to unwind than contracts.
- The AWS partnership is real distribution. When Amazon agrees to put your hardware inside Bedrock, you get access to every Bedrock customer.
- Hyperscalers want a Nvidia alternative. Microsoft, Amazon, Google, Meta all have strategic reasons to break Nvidia’s pricing power. Cerebras is one of the few credible inference plays they can lean on.
The Bear Case
Here’s what I think most investors might have overlooked.
1. The valuation prices in perfection
At the $185 IPO price, Cerebras went public at roughly 45x to 67x trailing sales, depending on which share count you use.
After the day-one pop to a $100 billion market cap, it’s closer to 95x trailing sales.
Price to sales (trailing) at recent prices. Cerebras at 95x sales.
Nvidia at 30x. Nvidia is profitable. Cerebras lost $146m operating in 2025.
For reference, Nvidia (gold standard, profitable, with 80% gross margins) trades nowhere near that level. Very few profitable chip companies have sustained 90x sales for long. When you pay 95x sales the thesis might be a little less grounded on fundamentals.
2. Customer concentration is the real story
Customer concentration by revenue share. One dependency, traded for another
UAE entities dominated 2024 and 2025. OpenAI dominates the forward backlog.
OpenAI can renegotiate. OpenAI can delay. OpenAI is also reportedly buying chips from AMD and working on its own custom silicon. The contract reportedly includes provisions that let OpenAI seize loan funds and demand repayment if Cerebras misses delivery milestones. That gives the customer significant leverage.
3. Manufacturing risk is unique to wafer-scale
This is the part most retail skips. When you manufacture a chip from an entire wafer, you can’t just discard the defective sections the way Nvidia does. Defect rates on a single 12-inch wafer are inherently higher. Cerebras uses redundant cores and yield optimisation to route around dead spots, but the company has never produced at the volumes the OpenAI contract requires.
If yields drop, or if TSMC prioritises larger customers like Apple and Nvidia during a capacity crunch, Cerebras can miss delivery. Missed delivery triggers the contract clauses mentioned above.
4. Geopolitical risk is unusual for a chip company
This isn’t Cerebras’s first IPO attempt. They tried to go public in 2024 and got halted. The reason? Too much of their revenue came from UAE customers, which regulators flagged as a national security risk.
To get cleared this time, G42 had to restructure into non-voting shares. CFIUS only signed off in October 2025. Cerebras also needs ongoing US export licences to keep selling to the UAE, and those licences can be revoked.
The bigger point: AI chip policy moves with politics. Most semiconductor companies have to deal with this, but few are as directly exposed as Cerebras. A change in administration, a shift in US-China-Middle East relations, or a new export rule hits Cerebras harder than it hits Nvidia or AMD.
5. Lock-up expiry
The standard IPO dynamic that retail tends to forget. Insiders, VCs, and employees are locked up for 180 days. When the lock-up expires, supply hits the market from holders who bought in at an $8 billion valuation. That has historically been a difficult period for newly public, momentum-driven names.
The Bottomline
Cerebras isn’t a bad company. The technology is genuinely impressive. The contracts are real. The growth is real.
My real gripe is less about the fundamentals and more about the state of AI right now. Buying Cerebras at these levels is, in my view, buying directly into the hype that AI is running on. I don’t know if this party still has legs, or what stage of it we’re in. What I do know is that for Cerebras specifically, the risks outweigh the rewards.
I’d rather sit this one out and put the capital somewhere the odds are clearer.
Not financial advice. Do your own due diligence. But if anyone tells you Cerebras is a no-brainer, ask them to walk you through the customer concentration page and explain the rationale.
Join our Telegram to stay updated on the latest investing news: https://t.me/realDrWealth
