Recently-IPO'd companies have the most concentrated risks. Cerebras's are unusually identifiable — and unusually high-stakes.
↓ scroll to read
✦ The bottom line
86% of revenue from UAE customers. Export controls limit upside there. Failure to deliver OpenAI capacity triggers $1B loan recall. CUDA ecosystem is the biggest non-customer obstacle.
↓ the brief below
Receivables from one UAE customer · year-end 2025
77.9
%
MBZUAI alone — an Abu Dhabi-based AI university — owed Cerebras 78% of all outstanding receivables as of Dec 31, 2025. (In 2024, G42 was 91% of receivables.) Concentration of receivables is even more pointed than concentration of revenue.
Source · 424B4 · Risk Factors — A substantial portion of our revenue / Accounts Receivable · FY2025 · Filed May 14, 2026
From the prospectus · export controls on UAE shipments
[We are subject to licensing requirements that] apply to the export of [our] systems for export to G42 and MBZUAI in the United Arab Emirates, but the applicable licenses require that we and [other compliance terms] ...
↳ Cerebras can ship to its biggest customers only with US-government licenses. If the policy environment changes — a new administration, a regional conflict, a sanctions update — those licenses can be revoked. The risk isn't theoretical.
Source · 424B4 · Risk Factors — Export Control / Geopolitical · FY2025 · Filed May 14, 2026
✦ Teach me
The CUDA moat (from the other side)
Every AI researcher learned to program AI on CUDA — Nvidia's software framework. Switching to a different ecosystem isn't just about hardware; it's about retraining the people, rewriting the code, and re-validating the results.
Cerebras's pitch is that PyTorch (which most AI code is actually written in) is what matters, and their compiler handles the rest. That's a strong technical argument. But "strong technical argument" and "5 million developers actually switching" are different things.
Wall Street calls this
CUDA lock-in
If Cerebras can't crack the developer-ecosystem problem, their hardware advantage doesn't matter. Watch for *third-party benchmarks and developer adoption metrics,* not just contract announcements.
From the prospectus · the OpenAI capacity trigger
Under the MRA with OpenAI, unless otherwise agreed to, we are required to deliver capacity tranches across specified numbers of data centers with minimum capacity thresholds upon certain time-based milestones. If we fail to deliver such capacity on the stated timelines, or if we experience a certain level of failure with respect to our service levels, OpenAI has the right to terminate a portion or all of the agreement.
↳ The same OpenAI contract that's Cerebras's biggest commercial validation is also a binding capacity commitment. Miss the milestones and OpenAI can pull the rug — including recalling the $1B working-capital loan. Execution risk has never been more concentrated for this company.
Source · 424B4 · Risk Factors — MRA with OpenAI · FY2025 · Filed May 14, 2026
⚠
Watch
Real risks: 86% UAE concentration, export controls, OpenAI delivery milestones, CUDA-ecosystem dominance. Stock is new. Risk profile is high. Conviction will come (or not) over the next 2-3 quarterly reports.