Cerebras Systems, the wafer-scale AI chip company, is going public at a valuation of up to $48.8 billion. That is interesting. The company reported $510 million in 2025 revenue, which means it is valued at roughly 100 times sales. Also interesting. But the weirdest number in the S-1 filing might be this one: 86% of Cerebras's 2025 revenue came from two UAE-based entities.
That is a funny thing for a company pitching itself as the next Nvidia. Nvidia sells GPUs to basically everyone. Cerebras, until very recently, sold wafer-scale systems mainly to G42 and the Mohamed bin Zayed University of Artificial Intelligence in the United Arab Emirates. If you are an investor trying to decide whether Cerebras is "the next Nvidia" or "OpenAI's captive bet," the customer concentration math is a useful place to start.
The story gets more interesting when you look at how Cerebras turned a loss into a profit. The company reported GAAP net income of $237.8 million in 2025. That sounds good! But $363.3 million of that came from a one-time, non-cash accounting adjustment - the paper gain from extinguishing a forward contract liability related to G42. Take that away, add back stock-based compensation, and the company's actual non-GAAP net loss was $75.7 million in 2025. Which is worse than the $21.8 million loss in 2024.
So you have a company that was deeply dependent on UAE sovereign money, used an accounting adjustment to show profitability, and is now going public at a 100x revenue multiple. That is the setup. The punchline is that Cerebras is trying to swap one kind of dependency for another.
Enter the $20 billion Master Relationship Agreement with OpenAI. The deal commits OpenAI to buying 750 megawatts of inference compute capacity from Cerebras, with an option to expand to 2 gigawatts. If you squint, this looks like Cerebras is trading sovereign dependency for platform dependency. Instead of 86% of revenue coming from the UAE, the future might see 100% of growth coming from OpenAI.
The financial engineering here is sort of elegant. You start with a technology that is genuinely different - wafer-scale chips that avoid the inter-chip communication bottlenecks of GPU clusters. You use sovereign wealth money to prove it works at scale. You paper over the losses with an accounting adjustment tied to that same sovereign relationship. Then you use that "profitability" and the sovereign-scale deployment to land a monster deal with the most important customer in AI. Then you go public at a valuation that prices in the OpenAI deal as if it's already diversified revenue.
The IPO pricing tells its own story. Cerebras initially filed to sell shares at $115 to $125 each, then raised the range to $150 to $160 after the roadshow showed more than twenty times oversubscribed demand. At the high end, the deal would raise about $4.8 billion. The valuation more than doubled from where it was just a few months ago.
What you are buying, if you buy Cerebras stock, is not really a "Nvidia competitor" in the traditional sense. You are buying a call option on OpenAI's inference demand. You are also buying a story about dependency shifting: from risky sovereign concentration (what if UAE politics change?) to risky platform concentration (what if OpenAI builds its own chips or picks a different supplier?).
The wafer-scale technology might be great. The OpenAI deal might be transformative. But the financial structure is what's actually novel here. It's a machine that turns sovereign financing into platform dependency while using accounting adjustments to smooth the transition from loss-making to public-market-ready. Whether that's genius or just financial engineering dressed up as AI disruption depends on whether you think OpenAI will actually need all that capacity, and whether Cerebras can sell to anyone else.
Either way, it's a pretty interesting way to build a $49 billion company.