Artificial intelligence chipmaker Cerebras Systems' debut earnings report delivered strong top-line growth, but an admission that it is renting its own chips back from a UAE customer to meet demand sent investors to the exit.
Cerebras (NASDAQ: CBRS) posted revenue that almost doubled in the year to March, but warned margins would contract sharply in the current quarter, triggering a sell-off that wiped roughly US$5 billion from the stock in after-hours trade.
- First-quarter GAAP revenue came in at $193.4 million, up 94% year-on-year, with cloud and services revenue climbing 178%.
- Net loss narrowed to $14 million from $23.9 million, or 46 cents per share, a year earlier.
- The company reported a near break-even core non-GAAP operating loss of $3.5 million and a cash position of $3.3 billion in liquidity.
- Core gross margin is forecast to fall to between 36% and 38% in Q2, down from 46.5% in the first.
During the analyst call, it emerged that Cerebras is temporarily renting its own systems back from an existing customer - believed by analysts on the call to be UAE-linked G42 - because it cannot build data centres fast enough to satisfy demand from its OpenAI and AWS partnerships.
Wedbush, which holds an outperform rating on the stock, had flagged ahead of the result that the main bottleneck facing Cerebras stems from TSMC's manufacturing capacity rather than demand, and anticipated the foundry would provide higher-than-expected wafer output in 2026 and 2027.
Cerebras CEO Andrew Feldman said that AI has moved from being a novelty to being useful and productive, and that fast AI is more valuable than slow AI because it is more productive.
Feldman told analysts the company expects to see meaningful impact from the AWS partnership in 2027.
IPO pop to hangover
Cerebras priced its IPO at $185 in May, saw its stock open at $350, and has since fallen 28%, closing on Tuesday at $226.72.
The listing raised $5.55 billion, making it the largest U.S. technology IPO since Uber's debut in 2019.
Global AI spending on inference - running models rather than training them - officially surpassed training expenditure in 2026, the market Cerebras's wafer-scale architecture is built to serve.
Independent benchmarks show the company's CS-3 system running inference up to 21 times faster than Nvidia's flagship Blackwell B200 on large language model workloads, at 32% lower total cost of ownership.
Nvidia acquired inference chip startup Groq for $20 billion in 2026, integrating its deterministic scheduling technology into the upcoming Rubin platform.
UAE concentration
Roughly 86% of Cerebras's $510 million in 2025 revenue came from two UAE-affiliated entities - Mohamed bin Zayed University of Artificial Intelligence at 62%, and G42 at 24%.
MBZUAI accounted for nearly 78% of accounts receivable at the end of 2025.
OpenAI is simultaneously Cerebras's largest prospective revenue source under a $20 billion multi-year compute deal, the provider of a $1 billion working capital loan, and a prospective shareholder through warrants attached to the revenue agreement.
"There are only two hardware vendors that currently serve OpenAI models and we're one of them - an empirical validation that the big models work just fine on us," Feldman said.
GPT-5.4 on Cerebras is currently available to OpenAI engineers and select customers, with GPT-5.5 integration described as the next phase of the rollout.
