Thursday morning a Silicon Valley chip company nobody on Main Street had heard of opened at $350 a share on the Nasdaq. By the closing bell, two of its founders were billionaires. The company is Cerebras Systems. The IPO is the biggest pure AI listing Wall Street has seen in years.
Then Friday hit. The stock dropped 10 percent in a single session, dragging the rest of the AI trade with it. Welcome to public markets, where the same number that makes you rich on Thursday makes you sweat on Friday.
Look, this IPO matters for reasons that go past the share price. Cerebras isn’t building yet another GPU. It’s making a single chip the size of an entire silicon wafer, and selling it as the alternative to NVIDIA. That’s a bigger fight than most retail investors realize.
The Numbers Wall Street Wrote Down
Cerebras priced its IPO at $185 a share Wednesday night. CNBC’s debut coverage caught the opening tick: $350. That’s a 90 percent pop before normal humans could place a buy order.
By the end of Thursday’s session the stock had cooled to $311.07. Still up 68 percent on the day. The 30 million share float raised $5.55 billion in fresh capital for the company. The fully diluted valuation crossed $95 billion.
Compare that with Meta’s IPO at $104 billion in 2012, or Alibaba’s at $168 billion in 2014. Cerebras lands in their neighborhood at debut, before it has earned a single dollar of public market trust. That’s how starved the IPO calendar has been for pure AI exposure, and how badly funds wanted in.
The Wafer Scale Engine: Why Cerebras Isn’t Building a GPU
Every other AI chip company on the planet builds GPUs or GPU substitutes the size of a postage stamp. Maybe a bit bigger. NVIDIA’s flagship H200 is about the size of your thumb.
Cerebras throws that whole approach away. Their Wafer Scale Engine 3, the WSE-3, takes the entire 12 inch silicon wafer that other companies cut up into hundreds of small chips, and instead uses the whole disk as one giant processor. The TechMarketer breakdown noted the WSE-3 is 58 times larger than a leading GPU chip. It delivers inference up to 15 times faster on supported models.
The catch is that wafer scale chips are notoriously hard to manufacture. Any defect anywhere on the wafer can kill the whole part. Cerebras has worked around this for almost a decade with redundancy in the architecture. They are now the only company shipping wafer scale parts at any meaningful volume.
If you believe AI workloads keep getting larger and more memory hungry, the wafer scale bet looks smart. If you think workloads stay roughly where they are, GPUs win on flexibility and supply chain.
The OpenAI and AWS Customers Wall Street Cared About
Two customer names did most of the work justifying the $95 billion valuation.
OpenAI signed a cloud deal in January 2026 worth more than $20 billion that runs through 2028. Cerebras supplies the compute, OpenAI runs inference workloads on it. This is the same OpenAI that buys most of its compute from NVIDIA via Microsoft. Adding Cerebras alongside is a hedge against any single supplier bottleneck.
Amazon Web Services signed a binding term sheet in March to deploy Cerebras systems inside AWS data centers. That’s hyperscaler validation. AWS doesn’t put non NVIDIA chips into customer facing infrastructure unless the economics are real.
Add a Department of Energy supercomputer contract, a Mayo Clinic medical AI deal, and roughly 40 other named enterprise customers. The revenue stack is concentrated but real.
Friday: The Pop Gives Back 10 Percent
Day two was less fun. CNBC’s Friday tracker caught the slide. Cerebras opened weak, traded through $290 around midday, and closed near $280. Down 10 percent on the session.
The selloff wasn’t Cerebras specific. Treasury yields jumped to a one year high. Trump returned from Beijing without a US China framework. Oil futures spiked on Strait of Hormuz tension. The whole AI complex sold off: NVIDIA down 4.4 percent, Intel down 6, AMD down 5.7, Micron down 6.6. Cerebras took the biggest hit because it was the newest and most over owned in the trade.
This is also the well known IPO pattern. Allocations get distributed Wednesday night. Lock ups don’t apply to flippers. Day two flippers sell. Volume drops. The float settles into longer term hands over the next four to six weeks.
The Billionaires Created Overnight
CEO Andrew Feldman walked into Thursday morning with options and walked out with $3.2 billion in paper wealth. CTO Sean Lie added $1.7 billion to his net worth. CNBC’s separate piece walked through the cap table.
Feldman is on his fifth startup. He sold his previous company SeaMicro to AMD for $334 million in 2012. Cerebras was bootstrapped from that exit plus a series of vinegar tough funding rounds at unfashionable times. Lie is a UCLA trained chip designer who has been the technical architect of the wafer scale engine since the first prototype.
Two engineers. Two billionaires. Ten years of public skepticism about whether wafer scale would ever work outside a lab. That’s the story most retail traders never see.
Why This IPO Signals a Bigger Pipeline
The IPO calendar has been frozen for tech for almost two years. Rising rates killed valuations. Funds got cautious. Founders waited. Companies that should have been public, like Databricks, Stripe, OpenAI’s spinoffs, all stayed private.
Cerebras opens the window. Motley Fool laid out the next wave: SambaNova, Groq, Tenstorrent, and at least three Chinese chip companies are watching the Cerebras reception to decide whether to file their own S 1s this summer. If Cerebras trades flat to up over the next six weeks, the floodgates open.
For the AI ecosystem, more public AI hardware companies means more price discovery, more competitive pressure on NVIDIA’s margins, and more capital for the second tier players. Net positive.
Where the Risks Actually Live
Three real risks to call out.
First, customer concentration. If OpenAI’s $20 billion contract gets renegotiated lower, or if AWS slows its deployment plans, a meaningful chunk of forward revenue evaporates. Both customers have leverage Cerebras can’t easily replace.
Second, NVIDIA’s response. Jensen Huang’s company does not lose comfortably. Expect aggressive pricing on H200 and Blackwell parts, deeper bundling with networking and software, and selective long term supply contracts with strategic customers. Cerebras still has to win on absolute performance against a defender with infinite balance sheet.
Third, manufacturing yield. Wafer scale chips depend on TSMC’s most advanced nodes. Any geopolitical disruption to Taiwan supply, or any yield hiccup on the next process generation, could starve the product roadmap.
None of these are deal breakers today. All of them are real two to three years out.
Why This Matters
For the AI investment thesis, the Cerebras IPO is the first proof point this year that public markets will pay full price for a credible NVIDIA alternative. That changes how venture funds think about their chip stage portfolios, how strategic buyers think about acquisitions, and how NVIDIA thinks about defending its 80 percent margin.
For retail investors who missed Thursday’s pop, the question is whether to chase. History on first day IPO doublers is mixed. Some hold and grind higher (Snowflake). Some give back the move and never recover (Robinhood). The fundamentals here look stronger than most. The valuation is steep.
For the broader IPO calendar, this is the unfreezing signal. Watch the next four to six weeks. If Cerebras holds above $250, expect three to five more AI infrastructure filings before September.
USABlaze Takeaway
My read: this was a real IPO, not a meme one. The technology works at production scale. The customer logos are blue chip. The founders aren’t selling. The valuation is rich but defensible if the OpenAI and AWS deals extend.
That doesn’t mean buy it Monday morning at any price. It does mean Cerebras has earned a seat at the AI table, alongside NVIDIA, Broadcom, and a couple of others. The category just got more interesting, more competitive, and more investable.
The 10 percent Friday slide is a footnote, not a verdict. Wait two weeks. Let the float settle. Then make a call.
Sources: CNBC, CNBC (day 2), CNBC (billionaires), Motley Fool, TheTechMarketer.
By The USABlaze Editorial Desk