Cerebras Systems made a splash last week with an IPO that saw its stock jump 68% on the first day, closing at $311 and pushing the company's market cap to roughly $95 billion. The tech is legitimately impressive—they make a chip the size of a manhole cover that's designed to make AI training go dramatically faster than traditional approaches.
But before you get caught up in the hype, let's talk about the red flags buried in the paperwork.
The UAE problem
Everyone has been talking about how Cerebras cut ties with G42, the controversial UAE-based company that was previously a major customer. That's supposed to be the good news—proof that Cerebras cleaned up its act and diversified away from risky geopolitical exposure.
Except here's what they're not emphasizing: another UAE entity, MBZUAI (Mohamed Bin Zayed University of Artificial Intelligence), accounted for 62% of Cerebras's revenue last year. When you add G42 into the mix, a staggering 86% of the company's revenue came from the United Arab Emirates.
That's not diversification. That's concentration risk on steroids.
If U.S. export policy toward the UAE changes—or if Middle Eastern AI funding dries up—Cerebras's revenue could crater overnight. The fact that the IPO prospectus downplayed this dependency while hyping the G42 breakup tells you everything you need to know about how they want investors to perceive the risk.
The OpenAI deal—and the conflict of interest
Cerebras recently signed a massive multi-billion-dollar contract with OpenAI to provide AI infrastructure over the next several years. On paper, this looks like the validation every hardware startup dreams of—one of the most important AI companies in the world betting big on your technology.
But there's a problem. According to court disclosures from the ongoing Elon Musk lawsuit against OpenAI, OpenAI president Greg Brockman personally owns equity in Cerebras. That means Brockman was negotiating deals on behalf of OpenAI to buy hardware from a company where he has a financial stake.
That's a textbook conflict of interest. It doesn't mean the deal is bad or that Cerebras's technology isn't legitimate, but it does raise serious questions about whether OpenAI's procurement process was truly arm's length. If this turns into a legal or regulatory issue, it could jeopardize the very contract that's propping up Cerebras's valuation.
The financials: not as pretty as they look
Cerebras reported $510 million in revenue for 2025 and $238 million in net profit, which sounds great—until you realize that most of that profit came from a one-time accounting adjustment. Strip out the accounting tricks, and the company actually lost around $146 million on its core operations last year.
At a $95 billion valuation, that's more than 180x trailing revenue and deeply negative on an operating basis. Even if you give them credit for future growth from the OpenAI deal, you're paying an enormous premium for a company that's still burning cash and dependent on a tiny handful of customers.
What about the tech?
The actual product is legitimately cool. Cerebras's Wafer-Scale Engine is a single massive chip that eliminates the need to connect dozens of smaller GPUs together, which theoretically makes AI training faster and more efficient. They've also recently launched their Naxtra sodium-ion battery technology and announced a 60 GWh deal with HyperStrong, signaling they're serious about scaling manufacturing.
The problem isn't whether the technology works—it's whether the business works. Great tech doesn't always translate into great companies, especially when your revenue is concentrated in one region, your largest contract has conflict-of-interest issues, and you're trading at a valuation that assumes flawless execution for years to come.
Should you buy?
At $311, you're paying IPO hype prices. The original institutional investors priced the IPO at $185, and even that was expensive given the revenue concentration and profitability concerns. If you're interested in Cerebras, the smart move is to wait for the initial euphoria to wear off.
There's also a lock-up period expiring around November, when early investors and insiders will be free to sell. Historically, that's when IPO stocks take a hit as people who got in early cash out. If the stock drops back toward the $185-$200 range, it might start to make sense. At current levels, you're just chasing momentum.
Bottom line
Cerebras has impressive technology and some marquee customers, but the risks are real. Revenue concentration in the UAE, potential conflicts of interest with OpenAI, and a valuation that assumes everything goes perfectly are all reasons to be cautious. If you can't explain where the next $50 billion in market cap is supposed to come from, you're speculating, not investing.
