The Honeymoon Lasted One Day

Cerebras opened Friday, May 16 at $297.55 — down roughly 4% from Thursday's $311.07 close, according to ts2.tech. That Thursday close itself was already a retreat from an opening-bell high of $350, which briefly valued the company at $106.75 billion.

The stock hit $106 billion before most people had finished their morning coffee, then spent the rest of the week coming back to earth.

By Friday, the fully diluted valuation settled closer to $90 billion. Still enormous. Still up 61% from the $185 IPO price. But Wall Street had questions.

The Numbers Everyone Is Glossing Over

Cerebras reported 2025 revenue of approximately $510 million. That sounds impressive until you see the $145.9 million operating loss sitting right next to it, per ts2.tech.

The company is NOT profitable. It is burning money while simultaneously carrying a $24.6 billion backlog — a number that sounds like a guaranteed future, but backlogs are promises, not cash.

D.A. Davidson analyst Gil Luria put a fair-value target at roughly $115 per share — a number that implies the stock, even after Friday's pullback, is trading at more than 2.5 times what a serious analyst thinks it's worth right now. Nicholas Smith at Renaissance Capital flagged the valuation as "quite high even out to 2028," according to Reuters.

Barron's flagged the price-to-sales ratio as far loftier than Nvidia's. Nvidia — the undisputed king of AI chips — trades at a lower multiple than a company that hasn't turned a profit yet.

The Customer Concentration Problem Is Real — But Changing

Cerebras' customer concentration risk looks far worse in the headlines than it does in the actual business.

As recently as 2024, G42 — an Abu Dhabi-backed AI firm — accounted for 85-87% of Cerebras revenue, according to Noah Bean writing in Medium. That's why the 2024 IPO attempt collapsed under CFIUS national security scrutiny. One UAE-linked customer, one government review, company nearly dead.

Fast forward to the April 2026 S-1 filing: G42's share of revenue has collapsed to 24%. OpenAI signed a $10-20 billion compute agreement in December 2025, complete with a $1 billion working capital loan backed by warrants that could convert to significant Cerebras equity. AWS is deploying Cerebras hardware into its own data centers, with a commercial launch planned for H2 2026.

Current SEC filings still show 86% of revenue concentrated in two UAE-linked customers — but that's a lagging number. The business underneath it has fundamentally changed. Most coverage has not made this distinction.

OpenAI and AWS Are NOT Charity

Skeptics will say the OpenAI deal is hype. They're wrong.

OpenAI runs inference at a scale where even small improvements in tokens-per-second translate directly into margin and product quality. They are not buying Cerebras chips because CEO Andrew Feldman gave a good pitch. They ran the numbers.

AWS is even harder to impress. Amazon does NOT integrate third-party silicon into its own infrastructure without clearing a brutal evaluation on reliability, software compatibility, and supply chain predictability. When AWS offers a service, it becomes accessible to every company already on AWS — instantly, without new procurement. That's distribution Cerebras couldn't buy.

As Noah Bean noted in Medium: the single-patron story is "largely over, and it was over before the IPO filing made it official."

The Architecture Bet Is Massive — And Still Unproven at Scale

Cerebras' core claim is that one giant wafer-scale chip can outperform clusters of Nvidia GPUs for AI inference. CEO Andrew Feldman told Reuters: "In Silicon Valley we understand just how big AI will be."

Strong words. The technology is real — TechCrunch confirmed the company burned nearly $200 million and $8 million per month in 2019 just to solve packaging problems that nobody in the semiconductor industry had ever cracked. They invented custom machines to solve problems that didn't have solutions yet.

But surviving the R&D phase is different from winning the market. Nvidia and AMD are NOT sitting still. And Cerebras chips are optimized for inference — generating AI responses — not training. That's a narrower market than the full AI compute stack.

What This Means for Regular People

If you bought at the IPO price of $185, you're up 61% in two days. Congratulations.

If you bought at the Thursday open of $350, you're already underwater.

If you're watching from the sidelines: this is a company with real technology, real customers, and real losses, trading at a valuation that assumes everything goes right — OpenAI deploys at full scale, AWS launches on time, Nvidia doesn't crush them, and that $24.6 billion backlog converts to actual revenue.

That might all happen. It also might not. The stock is priced for perfection on a company that, three years ago, was weeks away from running out of money.

Buy accordingly.