Cerebras Systems raised its full-year 2026 core revenue guidance to $940m-$960m when it reported third-quarter results after the close on 12 November, up from the $880m-$890m it guided only three months earlier. That is a raise of roughly 8% at the midpoint, delivered inside a single quarter, and it came alongside Q3 core revenue of $230.9m — up about 70% year on year and about 7% ahead of the company's own guidance midpoint. The shares closed on Friday 13 November at $108.72, up 11.22% from $97.75, on 36.8m shares against a 9.22m average: about four times normal turnover.
For anyone asking whether Cerebras is a good investment in 2026, the reported numbers support a narrow, specific case and leave an equally specific gap. What they support: demand is running measurably ahead of the company's own forecast, remaining performance obligations stood at $26.1bn at the end of September, and the company confirmed a cloud agreement with Microsoft on the call. What would have to go right: Cerebras is still loss-making at the operating line, with full-year core operating margin guided at -10% to -8%, and that $26.1bn book is roughly 27 times the midpoint of this year's core revenue guide — so the binding constraint is not winning demand but physically delivering against it. Both halves of that are set out below. Neither is a recommendation.
What Cerebras actually reported
The first thing to get straight is which revenue line is being quoted, because Cerebras reports two and they differ by $20m in the quarter. Core revenue — the company's own non-GAAP measure — was $230.9m in Q3, up about 70% year on year. GAAP revenue was $210.8m, up about 55%. That distinction decides what the quarter looks like. Core revenue beat the company's own guidance midpoint of about $215m by roughly $16m, or 7%, a third straight quarter above guidance. Analyst consensus of about $214.7m, however, was set on the GAAP line, and GAAP revenue came in below it. The quarter beat guidance and narrowly missed the Street.
Within that, the cloud business did the heavy lifting. Core cloud revenue was $144.9m, up about 272% year on year, which means the inference-and-training-as-a-service side has gone from a minority of the business to the majority of it in about four quarters. Core gross margin came in at 41.0% for the quarter. On a GAAP basis the company lost $161.2m, or about $0.67 per share — a reminder that the growth is being bought with real cash and real stock-based compensation, not generated by an already-profitable platform.
The guidance is where the quarter earned its reaction, and here too the labels matter:
- Full-year 2026 core revenue: raised to $940m-$960m, from $880m-$890m in August.
- Full-year 2026 core gross margin: raised to 42%-44%, from 41%-43%.
- Full-year 2026 core operating margin: raised to -10% to -8%, from -19% to -17%.
- Q4 core revenue: $310m-$330m.
- Q4 core gross margin: 47%-49%, with core operating margin of roughly -1% to +1% — in management's framing, approximately breakeven.
That last line is the one worth sitting with. The full-year operating margin band of -10% to -8% is an average that includes three quarters already banked at much worse levels. The exit rate the company is guiding to is a business at roughly breakeven on core operating profit, with gross margin stepping up six to eight points quarter on quarter. Management attributed the Q4 gross margin step-up to hardware being sold at higher current prices. Alongside the numbers, the company said data-centre capacity under contract has grown to about 700 megawatts, and that it expects revenue to roughly triple in 2027.
Why the market reacted the way it did
An 11% move on four times normal volume is not a reaction to an 8% guidance raise alone. Three things landed at once.
First, the operating margin revision was larger than the revenue revision. Moving the full-year core operating margin band from -19%/-17% to -10%/-8% is roughly nine points of improvement against an 8% revenue raise. That is the signature of operating leverage arriving earlier than the market had modelled, not of a company simply selling more at the same economics.
Second, the Microsoft agreement. Terms were not disclosed and the company said it would share more next year, so the market was pricing the existence of the relationship rather than its size. For a company whose backlog has been heavily concentrated in a small number of very large counterparties, the identity of a new hyperscale customer carries information well beyond the contract value: it is third-party validation that the architecture is being designed into more than one cloud roadmap.
Third, the backlog disclosure. RPO of $26.1bn at 30 September compares with $24.6bn at the end of December 2025 as disclosed in the company's S-1 filings. That is a genuine increase, but a modest one over nine months, and it reframes the story. Cerebras's problem is not order intake. At roughly 27 times the current year's core revenue guide, the backlog is so large relative to delivered revenue that the valuation question becomes a conversion question: how fast can contracted demand be turned into recognised revenue, and at what margin.
Worth noting on the tape itself: some write-ups carried gains of 16%, 18% or 19%. Those reflect the after-hours move on 12 November and Friday's session high of $120.44, not Friday's close. The close-based move was 11.22%.
The Openbook read
Running the quarter through Openbook's five factors produces an unusually lopsided profile — which is itself the most useful thing to say about the stock.
Growth is the standout and sits at the top of the factor set. Core revenue up about 70% year on year, cloud revenue up about 272%, a guidance raise of roughly 8% at the midpoint delivered mid-year, and a stated expectation of revenue roughly tripling in 2027. On any growth screen this is close to the ceiling of what a listed company can print. Our screener shows how few names of this size carry a comparable growth signature.
Profitability is where the score is weak and will stay weak for now, but the direction of travel is the point. A 41.0% core gross margin guided to 47%-49% in Q4, and a core operating margin moving from roughly -19%/-17% to approximately breakeven on an exit-rate basis, is a profitability score improving from a low base faster than almost anything else in the sector. GAAP remains deeply negative — a $161.2m quarterly loss — and the gap between core and GAAP is not noise. Anyone scoring this on GAAP profitability gets a very different and equally valid answer.
Momentum is strong but has become expensive to hold. The close-based 11.22% gain on four times average volume is the profile of a stock where positioning is crowded. Momentum scores reward exactly this, and reward/risk penalises exactly this.
Solvency deserves more weight than it usually gets. This is a 2026 IPO funding a capital-heavy ramp: 700MW of contracted data-centre capacity does not build itself out of operating cash flow when the business is still around breakeven on core operating profit. Cerebras is converting a $26.1bn backlog on a balance sheet capitalised for a far smaller business, so the funding path is a live variable in the thesis, not a footnote.
Reward/risk is consequently the factor that decides the shape of the whole picture, and it is the least flattering. The reward case is arithmetically clear: a backlog at 27 times current revenue, margins inflecting, and a third hyperscale relationship. The risk case is equally clear and sits almost entirely on execution and concentration. A backlog that large relative to revenue means years of contracted work whose timing the company does not fully control; capacity constraints, supply of high-bandwidth memory and wafer-level packaging, and customer concentration all sit between the order book and the income statement. Reward/risk here is not a cautious footnote on a growth story — it is the story, because the growth is already contracted and the open question is delivery.
The read-across
The most useful comparison is not to the other AI accelerator names but to what happened in networking in the same week. Cisco reported the same evening, raised its full-year guidance, raised its hyperscaler AI infrastructure revenue expectation to $9bn, and its shares fell about 6% — because its non-GAAP product gross margin dropped 350 basis points to 63.7%, with management pointing to hardware mix and elevated memory costs. That is the same cost input that sits in Cerebras's bill of materials. Cerebras guided gross margin sharply up for Q4 while a hardware peer was guiding margin down on memory prices, and management's explanation was higher hardware pricing. Whether Cerebras can keep passing memory costs through is a question the Q4 print will answer, and it is the single most testable claim in this guidance raise.
For the wider AI compute complex, the quarter supports the demand thesis without resolving the returns thesis. Hyperscale capex is still being committed, and a loss-making challenger winning a Microsoft agreement tells you buyers are actively seeking alternatives to the incumbent. That sits against an unusually unsettled backdrop at the top of the sector: Nvidia's 8-K reports that the board accepted Jen-Hsun Huang's resignation as president, chief executive and director effective 11 November, with CFO Colette Kress appointed interim chief executive while a formal search runs. A less certain incumbent is part of why a credible second source of large-scale compute is being valued the way it is.
For peers the read is narrower than the headline suggests. A company delivering $230.9m of quarterly core revenue against a $26.1bn book is not yet taking revenue share from anyone. The read-across that holds is on inputs: 700MW of contracted capacity and a stated 2027 tripling point at the same constrained supply chain — memory, advanced packaging and power — that every other name in the flow is bidding for.
What to watch next
- Nvidia's quarterly results, due this week, with Kress presenting as interim chief executive. These will reset the sector's demand baseline and are the nearest-term risk to the whole AI compute complex, Cerebras included.
- The Q4 print and whether core gross margin actually reaches 47%-49%. This is the falsifiable part of the raise. If memory costs compress it the way they compressed Cisco's product margin, the operating leverage story slips a year.
- Whether Q4 lands at or above breakeven on core operating profit. Guidance is -1% to +1%. The first quarter of positive core operating profit would be a genuine regime change for the Profitability score.
- Disclosure on the Microsoft agreement. Management said more would come next year. Contract size, duration and whether it is capacity-committed or consumption-based all change the quality of the backlog.
- RPO conversion in the next two prints. RPO moved from $24.6bn in December 2025 to $26.1bn in September 2026. Watch the ratio of revenue recognised to backlog added — that rate, not the headline backlog, is what the 2027 tripling depends on.
- Any funding event. Building against 700MW of contracted capacity from a near-breakeven base is the most likely trigger for a capital raise, and the terms would feed straight into the Solvency score.
The guidance raise is real and the margin revision is bigger than the revenue revision. The open question is conversion, and this quarter did not settle it.
Discussion
Log in to join the discussion