Cerebras Systems pulled in $510 million in revenue for 2025, a 76% leap from the prior year, according to the AI chipmaker’s S-1 registration statement filed with the Securities and Exchange Commission in April 2026. That number covers the full calendar year, not a single quarter, but it marks a milestone for a company that barely registered on most investors’ radars two years ago. Separately, multiple reports have described a $20 billion computing agreement between Cerebras and OpenAI, a deal that, if confirmed, would reshape the competitive landscape in AI hardware. The catch: Cerebras’s own SEC filings do not explicitly confirm a contract of that size.
What the SEC filings actually show
The $510 million figure is about as solid as startup financials get before an IPO. It appears in audited statements within the S-1, backed by independent auditor sign-offs and management certifications. Revenue for 2024 came in around $290 million, meaning Cerebras nearly doubled its top line in a single year while selling systems built around its signature Wafer-Scale Engine, a processor that occupies an entire silicon wafer rather than being sliced into the smaller chips most companies ship.
That architecture gives Cerebras a genuine technical differentiator. A single WSE-3 chip packs 4 trillion transistors and 44 gigabytes of on-chip memory, allowing it to run large AI models without the complex networking that stitches together thousands of Nvidia GPUs in a typical data center. For customers running massive training jobs or high-throughput inference, the pitch is simpler infrastructure and faster time to results.
The S-1’s management discussion attributes the revenue jump to higher system shipments, expanded service contracts, and growing adoption of Cerebras’s cloud-access offerings. Repeat purchases from existing customers suggest that early deployments led to follow-on orders as workloads scaled. Cerebras also filed an amended S-1 in May 2026, refining language around customer concentration, backlog, and revenue recognition timing, though the core 2025 revenue number held steady.
Both filings are accessible through Cerebras’s EDGAR index page, where investors can review material agreements, risk disclosures, and exhibit lists. Notably, no contract exhibit in either filing identifies a single customer commitment anywhere near $20 billion.
The $20 billion OpenAI question
Reports of a blockbuster computing deal between Cerebras and OpenAI have circulated widely, with outlets including The Information and Bloomberg describing a long-term commitment by OpenAI to deploy Cerebras hardware at unprecedented scale. The reported figure, $20 billion, would dwarf any known chip-supply agreement outside of the hyperscaler world and would instantly make OpenAI Cerebras’s most important customer by a wide margin.
Yet the S-1 and its amendment do not name OpenAI as a counterparty to a deal of that magnitude. In a typical IPO filing, a contract worth multiples of a company’s annual revenue would normally surface in several places: as a filed material agreement, as a named driver of backlog, and as a concentration risk in the financial statement footnotes. Cerebras does disclose dependence on a small number of large customers and flags the risk that losing any one of them could materially harm the business. But the filings stop short of putting a dollar figure on any single relationship or naming OpenAI directly.
That gap does not necessarily mean the deal is fiction. Large technology partnerships often take the form of framework agreements, memoranda of understanding, or phased commitments that only become binding as specific purchase orders drop. A $20 billion headline number could reflect a multi-year capacity target rather than a signed, enforceable contract. It is also possible that the agreement’s structure does not yet trigger the SEC’s exhibit-filing requirements, or that details are being withheld pending final negotiation.
Still, the silence in the filings matters. Cerebras’s own risk factors warn that revenue concentration among a handful of buyers creates vulnerability, and if the OpenAI relationship is as large as reported, that vulnerability is more acute than the S-1 explicitly acknowledges. SEC rules require that registration statements not be misleading in light of all publicly available information, so any significant gap between what executives say publicly and what the filings disclose could invite regulatory scrutiny.
Where Cerebras fits in the AI chip race
Even without the OpenAI deal confirmed, $510 million in annual revenue puts Cerebras in rare company among AI chip startups. For comparison, Nvidia’s data center segment generated over $115 billion in its fiscal year ending January 2026, and AMD’s data center GPU revenue topped $12 billion. Cerebras is not competing at that scale, but its growth rate outpaces both incumbents on a percentage basis, and its wafer-scale approach has attracted customers that traditional GPU clusters cannot serve efficiently.
The competitive picture is shifting fast. Nvidia’s Blackwell architecture has raised the performance ceiling for GPU-based training, while startups like Groq have targeted inference workloads with their own custom silicon. Cerebras’s edge lies in workloads where a single massive chip can eliminate the inter-node communication bottlenecks that slow down distributed GPU clusters. If OpenAI is betting on Cerebras for a significant share of its inference infrastructure, it would signal that even the most GPU-dependent AI lab sees value in architectural diversity.
Cerebras’s IPO timing adds another dimension. The company initially filed its S-1 in late 2024 but delayed the offering amid market volatility and regulatory reviews, including scrutiny of its relationship with Abu Dhabi-based G42, a major customer and investor. The April 2026 refiling and May amendment suggest the company is pushing toward a public listing, though no pricing date has been announced. A confirmed OpenAI partnership of meaningful scale would likely boost the IPO valuation significantly, which may explain why both companies have an incentive to keep details under wraps until the timing is right.
What investors should watch next
The gap between Cerebras’s verified financials and the reported OpenAI mega-deal creates a two-track investment thesis. The conservative read focuses on what the SEC filings confirm: a company growing at 76% annually, with real customers, audited revenue, and a differentiated product in a market where demand for AI compute shows no sign of slowing. On that basis alone, Cerebras looks like a credible challenger in AI hardware, albeit one with meaningful customer concentration risk.
The aggressive read assumes that a substantial portion of the $20 billion OpenAI commitment will eventually convert into contracted, recognized revenue. That scenario implies years of hypergrowth and a potential step-change in Cerebras’s market position, but it also magnifies every risk the S-1 already flags: dependence on one buyer, execution pressure on large-scale deployments, supply chain complexity, and the possibility that regulatory or export-control actions could disrupt the relationship.
The next concrete signals will likely come from one of three places: an updated SEC filing that names OpenAI or quantifies a major contract, a public statement from OpenAI confirming the partnership’s scope, or Cerebras’s IPO prospectus, which would need to disclose material customer relationships in detail before shares begin trading. Until one of those surfaces, the $20 billion figure remains a market expectation, not a confirmed fact, and the $510 million in 2025 revenue stands as the firmest ground available for judging where Cerebras actually is today.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
