$385.
That was Cerebras's opening price on Thursday. IPO priced at $185, already far above the initial range of $115-125, and even after being forced to raise it to $150-160, it still wasn't enough — market demand was too hungry.
Double on opening, close at $311, market cap $66 billion. Co-founder CEO Andrew Feldman's net worth jumped to nearly $1.9 billion in a single day.
This isn't a regular IPO. It's the first shot of 2026 tech IPO season, and it sounded a bit absurd.
Who Is Cerebras?
Simply put: the only company besides Nvidia that has redesigned AI accelerators from the chip architecture level.
Nvidia's GPUs are general-purpose — training, inference, graphics, all doable. Cerebras's WSE (Wafer Scale Engine) is a wafer-level giant chip — the entire wafer is one chip, specifically optimized for AI training and inference. The area is dozens of times that of a GPU, and memory bandwidth is another order of magnitude.
The advantage: during large model training, communication overhead drops dramatically. The disadvantage: low manufacturing yield, high cost, narrow application scenarios.
An IPO Stuck in CFIUS for a Year
Cerebras's IPO path wasn't smooth. Filed in 2024, but a large investment from Abu Dhabi's Group 42 triggered CFIUS (Committee on Foreign Investment in the United States) review.
CFIUS review dragged on for a year.
During the review, investors noticed another problem: Group 42 accounted for nearly all of Cerebras's revenue. This means — Cerebras's "commercialization story" is essentially repeat purchases from a single customer. That's a red light in IPO roadshow materials.
Now the review has passed. And investors aren't just ignoring this red light — they've turned it into an opportunity.
What Is the Market Buying?
What does a $66B valuation correspond to in financials? Frankly, Cerebras's revenue scale doesn't match this valuation. The market isn't buying today's numbers — it's buying three expectations:
First, the option value of "a second Nvidia." AI compute demand keeps expanding, and nobody believes Nvidia can独占 this cake forever. Cerebras is the only listed non-Nvidia AI chip标的. Capital wanting to buy the "AI chip diversification" narrative has nowhere else to go.
Second, the certainty of the large model training market. Cerebras's customer base, though concentrated, consists of top model companies. As long as these companies keep expanding compute, Cerebras's orders are guaranteed.
Third, the scarcity of the IPO window. The 2026 tech IPO market has just thawed, and Cerebras is among the first batch. Investors are willing to pay the premium for early listing.
Let's Calm Down
$66 billion isn't a small number. AMD's market cap is about $300 billion, but AMD's product lines cover CPUs, GPUs, FPGAs, data centers — diversified revenue sources. Cerebras has only one product line, with extremely high customer concentration.
If Group 42's procurement pace slows, if Nvidia's next-gen chip continues to widen the gap in training scenarios, if large model companies start designing their own chips — any one of these risks materializing makes $66B expensive.
Of course, after closing at $311, the stock continued rising in after-hours trading. Short-term market sentiment doesn't look at fundamentals — it looks at scarcity and narrative.
My Take
Cerebras's technical route is real, not a PPT. But its commercialization story isn't hard enough yet. The IPO day's surge reflects expectations for "the second AI chip pole," not an already-realized competitive landscape.
The indicators to watch are simple: revenue growth in the next earnings season, number of new customers, and whether large customers beyond Group 42 emerge.
If all three look good, $66B might be just the starting point. If not, the pullback will be fast.
Main sources: