Cerebras Systems CEO Andrew Feldman wants you to know his company has a very good problem. The AI chipmaker is sitting on a $25 billion backlog of orders from some of the biggest names in tech, and it simply cannot build chips fast enough to fill them all.

That backlog, formally known as remaining performance obligations, represents committed revenue that Cerebras has yet to deliver. For a company that just reported Q1 2026 revenue of $193.4 million, a $25 billion pipeline is the kind of number that makes Wall Street do a double take.

The OpenAI anchor and a record-setting IPO

The single largest contributor to that backlog is a multi-year agreement with OpenAI worth more than $20 billion. The deal covers 750 megawatts of inference compute capacity, which in plain English means Cerebras will be powering a massive chunk of the infrastructure OpenAI uses to actually run its AI models after they’ve been trained.

Cerebras went public in May 2026 in what became the largest semiconductor IPO in history. Valuations have swung considerably, ranging from an initial estimate around $23 billion to peaks near $70 billion after trading began. Despite the eye-popping backlog and revenue growth of roughly 92% year-over-year, the market punished Cerebras shares after the company issued cautious guidance on margins.

The supply chain bottleneck

Feldman has been vocal about the problem. In a recent interview on the 20VC podcast, he pointed to insufficient data center infrastructure growth as a key bottleneck. Memory sourcing has also been flagged as a constraint across the sector. Cerebras’ wafer-scale architecture, which places an entire chip on a single silicon wafer rather than cutting it into smaller pieces, demands enormous amounts of high-bandwidth memory to function at peak performance.

Why crypto and digital asset investors should pay attention

The same data center infrastructure powering AI inference also underpins much of the crypto ecosystem. Mining operations, decentralized compute networks like Render and Akash, and the growing category of AI-crypto convergence projects all compete for the same scarce resources: chips, power, and rack space. When companies like Cerebras and OpenAI lock up 750 MW of compute capacity in long-term contracts, that tightens the supply of infrastructure available to everyone else.

There is also the valuation question. Cerebras’ post-IPO valuation reaching as high as $70 billion for a company generating less than $200 million in quarterly revenue reflects a market that is pricing AI infrastructure companies on future potential rather than current fundamentals.

Investors watching the AI hardware race should track whether Cerebras can convert its backlog into actual revenue at healthy margins. The 92% year-over-year growth is promising, but margin guidance spooked the market for a reason.