Cerebras Systems – Anchor Client

Credibility is the real win for Cerebras Systems.

• Even if OpenAI never deploys a single chip of the $10bn deal it has signed with Cerebras Systems, the deal gives Cerebras a huge credibility boost that should enable to sign more clients which solves the single biggest concern I have had with this company.
• OpenAI has agreed to deploy 750MW of Cerebras Compute chips that are very good at running inference at very high speed.
• By comparison, the equivalent capacity of Blackwell from Nvidia would have cost around $26.3bn meaning a significant saving as long as Cerebras can match Nvidia when it comes to tokens produced per $ invested.
• This is something that Nvidia prides itself on and so I suspect that this deal is more about OpenAI expanding its supplier base in order to ensure competitive tenders in the future and the best value for money.
• However, how OpenAI will pay for the silicon is not clear and given the deal it was able to get with AMD, which is a much stronger and viable supplier than Cerebras, I suspect that something similar may have occurred.
• This would be an option grant from Cerebras to OpenAI with a strike price of $0.01 per share, exercisable as the chips are deployed and when Cerebras shares hit a certain valuation in the market.
• This elegant solution would mean that Cerebras shareholders (via dilution) and marginal buyers of Cerebras in the market are the ones that actually end up paying for the silicon.
• The fact that Cerebras Systems is likely to go public this year and that the silicon is expected to be deployed in multiple stages between now and 2028, further supports the hypothesis that an AMD-like arrangement has been reached.
• While this is suboptimal for Cerebras shareholders, what they really win is credibility that will be of enormous help when it comes to selling the silicon to other customers who will be paying in cash.
• Cerebras Systems filed and then delayed an IPO in 2024, and at the time, I had significant concerns about its public market readiness.
• This was because while Cerebras had demonstrated excellent growth, almost all of it has come from one customer and one shareholder, G42, which is an AI company based in Abu Dhabi that is is owned by the state via its Mubadala Investment Company.
• Revenue concentration was about as high as I have ever seen it, with 83% of 2023 and 87% of H1 2024 revenues coming from G42, which created substantial single-source risk for revenues.
• Furthermore, without a well-known and used development platform, making one’s AI run on Cerebras Systems involves a lot of work, and is best suited to completely vertical algorithms as would be created by governments and military organisations (like G42).
• The addition of OpenAI brings Cerebras into the spotlight and gives its platform a degree of credibility that it has lacked to date, and I think it should win a lot more customers as a result of this deal.
• Hence, even though Cerebras’ owners may be paying for the silicon chips that are going into OpenAI, I suspect that they will make it back and much more through increased revenues and market share from customers that would otherwise not have signed up.
• In this environment where everything is in short supply, Cerebras is pushing against an open door, meaning that this deal is just what it needs at just the right time.
• However, I am pretty sure that this will be conditional on a successful IPO and a subsequent strong rise in the share price.
• Fortunately, this deal delivers the company the ammunition to deliver just that.
• Cerebras Systems will be a company I pay much more attention to going forward.