Cerebras Systems just revealed why the AI chip business is harder than it looks. The company reported strong revenue growth in its first earnings report since going public, but a dramatic margin warning exposed a fundamental shift in how AI infrastructure companies actually make money. The company is discovering that selling computing power, not chips, requires a completely different financial playbook.

Why Did Cerebras' Profit Margins Suddenly Collapse?

Cerebras reported Q1 core revenue of $191.3 million, up 92% year-over-year and beating analyst expectations of roughly $181 million. The company also raised its full-year revenue guidance to $855-$865 million, suggesting continued momentum. But then came the shock: Q2 gross margin guidance plummeted to 36-38%, down from Q1's 47%. The stock fell more than 10% after-hours on the news.

The reason reveals a critical tension in the AI infrastructure race. Cerebras is shifting from selling chips directly to customers to building its own data centers and selling computing power instead. This transformation requires massive upfront capital investment and operational complexity. To fulfill its massive contract with OpenAI as quickly as possible, Cerebras is temporarily leasing back hardware systems it previously sold to other customers and deploying them in third-party data center facilities. This arrangement dramatically increases short-term costs, which is why margins are being squeezed.

What Changed in Cerebras' Business Model?

The earnings report revealed a striking shift in revenue composition. Hardware sales, which once dominated the business, now account for only 58% of revenue at $111.6 million. Cloud and services revenue jumped to 42% at $79.8 million. A year earlier, the split was roughly 70% hardware and 30% services. More tellingly, cloud services revenue grew 167% year-over-year, nearly three times faster than hardware sales.

This transition from chip seller to computing power provider explains the margin pressure. When you sell a chip, you manufacture it, ship it, and collect payment. When you sell computing power, you must build data centers, maintain them, manage power consumption, and deliver consistent performance over years. The financial structure of Cerebras' largest contract illustrates this complexity. OpenAI's deal, valued at over $20 billion, includes not just computing power purchases but also a $1 billion operating capital loan to Cerebras and warrants for company stock. In essence, OpenAI is simultaneously a customer, creditor, and potential shareholder.

How to Understand Cerebras' Path Forward

• Data Center Bottleneck: The company's current constraint is not chip supply from manufacturing partner TSMC, but rather physical space in data centers. New facilities are expected to come online in the second half of 2026, which management believes will ease cost pressures and improve margins.
• Customer Concentration Risk: In fiscal 2025, 86% of revenue came from just two entities linked to the United Arab Emirates, with MBZUAI accounting for 62% and G42 for 24%. OpenAI revenue is expected to begin contributing in February 2026, and AWS deployments may not show financially until 2027, meaning genuine diversification remains years away.
• Contract Realization Timeline: Full-year revenue guidance of $855-$865 million implies the next three quarters must average roughly $220 million each, with sequential acceleration. Management stated that year-over-year growth will increase each quarter in 2026, with more revenue concentrated in the second half.

The AWS partnership adds another layer of complexity. AWS's Trainium 3 chip handles the initial stage of processing user prompts, while Cerebras' CS-3 system handles the faster output generation stage. This "split inference" architecture allows Cerebras to focus on the part of the pipeline where its speed advantage is greatest. However, management declined to disclose the specific scale of the AWS cooperation during earnings calls, and revenue contribution is not expected until 2027.

What Are Investors Actually Betting On?

Cerebras trades at roughly 50 times forward sales based on full-year guidance, a valuation that assumes flawless execution of massive, long-dated contracts. The median price target from 10 covering analysts is $300, with a range of $250 to $340. This pricing implicitly assumes that OpenAI's 750-megawatt computing power deployment and AWS inference solutions will be delivered on schedule and at projected volumes.

Several risks complicate this outlook. The lock-up period for early investors contains unconventional early release clauses triggered if Cerebras' market cap exceeds $40 billion, a threshold the company is approaching. The path to gross margin recovery remains unclear, dependent on data center construction timelines. Additionally, OpenAI itself is not yet profitable and has already scaled back some computing power commitments, raising questions about whether the company can sustain its massive infrastructure spending.

The broader lesson from Cerebras' earnings is that the AI infrastructure business is fundamentally different from the chip business. Building the physical and financial infrastructure to deliver computing power at scale requires managing capital intensity, operational complexity, and customer concentration risks that chip sales alone never demanded. For investors and industry observers, Cerebras' margin guidance cut is a reminder that in AI infrastructure, the real bottleneck may not be chip design or manufacturing, but the unglamorous work of building, financing, and operating the data centers that power the AI revolution.

" }