Cerebras Systems raised its 2026 outlook for core revenue to $940 million to $960 million, from $880 million to $890 million in August, in its third-quarter earnings release filed with the SEC on 12 November. At the $950 million midpoint the increase is $65 million, or 7.3%, by our arithmetic.
Update, 12 November, after hours. Cerebras (CBRS) closed the regular session at $97.75, down 3.3% from the prior close of $101.08, then traded at $115.40 in extended hours (extended-hours, Yahoo Finance data, 22:53Z), up 18.1% on the regular close. CNBC reported the stock rose about 18% in extended trading. With 243.0 million shares outstanding across Class A, B and N as of 5 November (10-Q cover), the $17.65 move is worth about $4.3 billion in market value (our arithmetic). After-hours prices are thin and can reverse at the open.
The company also reported $26.1 billion of remaining performance obligations, its contracted backlog. That is about 27 times the new 2026 midpoint, by our arithmetic.
Business Pill · Gross Margin
A short explainer on gross margin. It teaches the general idea only and does not discuss Cerebras.
The key insight: As we read the release, the raise is small next to the backlog: the 2026 core revenue outlook rose $65 million at the midpoint, while remaining performance obligations of $26.1 billion are about 27 times that midpoint, by our arithmetic.
More on this. Who holds the customer warrants: the 10-Q shows OpenAI’s warrant is about 75% of the $1.09 billion asset, by our arithmetic, but only $3.2 million of the $36.4 million third-quarter reduction in revenue; the rest relates to warrants issued to G42 and an Amazon entity: OpenAI Warrants Are 75% of Cerebras’ $1.09B Revenue Offset.
More on the Microsoft deal. The 10-Q says the Microsoft agreement is for cloud inference services, carries preferential commercial terms and has milestones to meet before service starts; MBZUAI, OpenAI and G42 accounted for 79% of third-quarter revenue, by our arithmetic: Cerebras Adds Microsoft as 3 Clients Make 79% of Its Revenue.
GAAP revenue for the quarter ended 30 September was a record $210.8 million, up 55% from $135.7 million a year earlier. Cloud and other services revenue rose 265% to $141.9 million. Hardware revenue was $68.9 million, against $96.8 million a year earlier, down 29% by our arithmetic.
Cloud and other services made up 67% of GAAP revenue in the quarter, against 29% a year earlier, by our arithmetic. Core revenue, a non-GAAP measure, was $230.9 million, up 70%.
The GAAP gross margin was 20.0%, against 40.6% a year earlier; the core gross margin was 41.0%. The GAAP operating margin was minus 106%, the core operating margin minus 14%. The net loss was $161.2 million, against $22.2 million a year earlier; core net income was $29.8 million.
In August Cerebras had guided third-quarter core revenue to about $214 million to $216 million. The $230.9 million it reported is $14.9 million above the top of that range, by our arithmetic. Chief financial officer Bob Komin said: “We exceeded our guidance across the board for the third quarter in a row and delivered sequential improvements on all core financial metrics.”
Cerebras’ $26.1 billion of remaining performance obligations at 30 September against its raised 2026 core revenue outlook, nine-month core revenue and third-quarter core revenue. Core revenue is non-GAAP; the midpoint and the 27x are our arithmetic.
The Raised Outlook
For the fourth quarter Cerebras guides core revenue of about $310 million to $330 million, a core gross margin of 47% to 49% and a core operating margin of minus 1% to plus 1%. Analysts polled by LSEG were looking for $271 million of fourth-quarter revenue and a 42.1% gross margin, CNBC reported.
For the full year it now guides a core gross margin of 42% to 44%, up from 41% to 43% in August, and a core operating margin of minus 10% to minus 8%, up from minus 19% to minus 17%. The fourth-quarter midpoint of $320 million is 38.6% above third-quarter core revenue, by our arithmetic.
Chief executive Andrew Feldman said in the release: “That progress energizes us to do even more as we push the limits of innovation and expect to triple revenue in 2027.”
The Backlog, by Our Arithmetic
Remaining performance obligations were $26.1 billion at 30 September, up from $25.4 billion at 30 June, an increase of $0.7 billion. Against the $950 million midpoint of the 2026 outlook, the backlog is about 27.5 times a full year of core revenue.
The release gives no schedule for when that backlog turns into revenue. Cerebras said it has secured 700 gross megawatts of data center capacity for delivery by the end of 2027, up from more than 600 megawatts in August.
Cash, cash equivalents, restricted cash and short-term investments totalled $7.8 billion, against $8.6 billion at 30 June, with $850 million of debt capacity. Purchases of property and equipment were $1.05 billion in the first nine months, and operating cash flow was minus $388.9 million.
Customers and Warrants
The release names Oracle, which is integrating Cerebras inference into Oracle Cloud Infrastructure, available in 2027, and says Cerebras signed “an important cloud deal with a top three hyperscaler” without naming it. Its risk language cites “its dependence on a limited number of significant customers” and its Master Relationship Agreement with OpenAI.
Feldman told CNBC in an interview that executives would announce a new relationship with Microsoft, one of the world’s top cloud providers, on the analyst call, and that Microsoft signed the deal in October. “We believe revenue begins to show up in Q1 ’27,” he said. He also said Cerebras is evaluating its financing options to meet demand.
Customer warrants on the balance sheet were $140.2 million current and $951.8 million non-current, $1.09 billion in total, against $152.4 million at the end of 2025, by our arithmetic. The release defines the amortization of customer warrants as “equity granted to customers and recorded as contra-revenue”.
That amortization reduced third-quarter revenue by $36.4 million, 17.3% of GAAP revenue, by our arithmetic. It is the largest adjustment between GAAP and core revenue; the other is $16.4 million of pass-through data center revenue that core revenue excludes.
How the Shares Moved
Cerebras shares closed the regular session at $97.75, down 3.3% from $101.08. After the release they traded at $113.85 at 21:44 UTC, 16.5% above the close, by our arithmetic from Yahoo Finance data.
Last week Cerebras disputed short seller Culper, saying under 0.5% of its wafers had been retired. Its CS-4 system, which the release says “will soon be in full production”, was the subject of our earlier piece. Cerebras was also among the companies that warned investors about a cloud-access bill.
The Structural Read
As we read the margins, much of the gap between GAAP and core is non-cash: $36.4 million of customer-warrant amortization was recorded as a reduction in revenue in the quarter, 17.3% of GAAP revenue, by our arithmetic.
As we read the mix, cloud and other services made up 67% of GAAP revenue in the quarter against 29% a year earlier, while hardware revenue fell 29%, by our arithmetic.
Bob Komin, chief financial officer, Cerebras earnings release, 12 November 2026
“We exceeded our guidance across the board for the third quarter in a row and delivered sequential improvements on all core financial metrics”
Three Implications
FOR INVESTORS Cerebras guides fourth-quarter core revenue of about $310 million to $330 million and a core operating margin of minus 1% to plus 1%.
FOR CUSTOMERS The release says Cerebras inference is to be available in Oracle Cloud Infrastructure in 2027 and names a cloud deal with an unnamed top-three hyperscaler.
FOR OUR COVERAGE We will report customer concentration and the backlog schedule from the 10-Q when it is filed.
The framework says: “Inference is a revenue engine: continuous, distributed across millions of applications and agents, incurring marginal cost with every query.”
As we read the release, Cerebras now earns most of its revenue that way: cloud and other services were 67% of GAAP revenue in the quarter, by our arithmetic. The framework predates these results and does not discuss Cerebras.
What Is Not Established
The release does not name the top-three hyperscaler; it says more will be shared next year. Feldman named the new Microsoft relationship to CNBC; we had not seen its terms when we wrote. It gives no revenue share by customer. The 10-Q for the quarter, which usually carries customer concentration and the backlog schedule, had not been filed when we checked EDGAR at 21:40 UTC. Core figures are the company’s own non-GAAP measures. The after-hours price is a single reading.
This Week In The AI Supercycle
The Great Convergence of Capital, Power, and Geopolitics
This week’s issue: for two years the AI build-out was priced as a supply story, and last week as a credit story. This week it was priced as a permission story, and the answer arrived as a cost, not a ban.
Cerebras raised its 2026 core revenue outlook to $940 million to $960 million, $65 million higher at the midpoint, after third-quarter core revenue of $230.9 million came in above its own range. Its $26.1 billion backlog is about 27 times the new midpoint, by our arithmetic.
95,000+ executives read Business Engineer for the AI strategy frameworks cited by ChatGPT, Claude, and Perplexity.
A note on sourcing. Results, outlook, backlog, balance sheet and quotes come from Cerebras’ third-quarter earnings release (8-K Exhibit 99.1, accepted by the SEC at 21:03 UTC on 12 November), the prior outlook from its August release; Reuters first reported the forecast raise in a headline we read. Prices are Yahoo Finance chart data. Midpoints, ratios and percentages marked as ours are our arithmetic.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
