Cerebras Stock Gains on First Core Profit but Gives Back Earnings Surge
Market Analysis by Itai Smidt covering: Nebius Group NV, United States 10-Year, Cerebras Systems Inc. Read Itai Smidt's Market Analysis
Cerebras Systems traded at $103.68 on Friday, up $5.93 or 6.07% from Thursday’s close of $97.75, after a session that opened with a post-earnings surge and then gave back most of it as the semiconductor sector sold off. The stock printed $115.51 in after-hours trading Thursday, up 18.17%, and reached $116.70 in Friday’s premarket, a 19.39% gain. It opened the regular session at $108.70, traded at $113 by 9:38 a.m. ET, up 16%, and touched an intraday high of $117.11 before the chip-sector reversal dragged it back toward $104.
The rally erased three weeks of damage in a few hours, then handed most of it back. At the $117.11 high, the stock was 50% above its November 5 all-time low of $77.84. At $103.68 it sits 33% above that low. The fade from $117.11 to $103.68 cost $13.43 per share, or 11.5%, inside a single session.
Thursday set the base for the move. CBRS fell 3.29%, or $3.33, to close at $97.75 from a prior close of $101.08, trading between $93.87 and $104.05 on volume of 14.99 million shares against an average of 9.09 million. The stock had slipped 0.35% to $100.73 earlier in the session before the report. At Thursday’s close, Cerebras carried a market value of $23.22 billion on 237.56 million shares outstanding, with a public float of 98.91 million shares.
The long view frames how far the stock has fallen. Cerebras listed on Nasdaq in May at an IPO price of $185 and hit an all-time high of $386.34 on May 14. Thursday’s $97.75 close was 47.2% below the IPO price and 74.7% below the peak. Over the past month, the stock lost 31.74%, though it gained 11.07% over the past week as it bounced off the November 5 low. Beta of 5.68 makes CBRS one of the most volatile large-cap names in the semiconductor group.
At $103.68, the market value stands at $24.63 billion. The rally added $1.41 billion in market capitalization by late morning, down from the $4 billion gain the stock held at the opening high.
The setup heading into the weekend is a collision between two forces. On one side sits the best quarter in the company’s short public history: a third-straight beat on its own guidance, its first positive adjusted net income, a fourth-quarter revenue guide that cleared consensus by $49 million at the midpoint, a Microsoft cloud deal, and a reiterated plan to triple revenue in 2027. On the other sits a semiconductor sector in liquidation after Nvidia’s chief executive resigned, a 10-year Treasury yield at 5.16%, and a stock that still carries heavy execution risk tied to its OpenAI contract.
Q3 Results: Core Revenue of $230.9 Million and the First Core Profit
Cerebras delivered its third consecutive quarter above guidance, and the margin of the beat widened.
Core revenue reached a record $230.9 million in the third quarter, up 70% from $135.7 million a year earlier and up 10% sequentially. That cleared the company’s own guidance range of $214 million to $216 million by $16 million at the midpoint, or 7%, and topped the $214.7 million consensus by the same amount. GAAP revenue rose 55% to a record $210.8 million.
The $20.1 million gap between core and GAAP revenue comes from accounting for customer warrants. GAAP books $36.4 million of customer-warrant amortization as a reduction of revenue, partly offset by $16.4 million of data-center pass-through revenue that management excludes from core results. Those warrants are tied primarily to OpenAI, which holds warrants to purchase 33.4 million Cerebras shares subject to its purchases of compute capacity.
Profitability crossed a line. Core net income reached $29.82 million, compared with a loss of $10.86 million a year earlier and an expected loss of $38.4 million. That marks the first positive adjusted net income for Cerebras as a public company. Adjusted earnings came in at $0.12 per share against an expected loss of $0.14. On a GAAP basis, the net loss widened to $161.2 million, or $0.67 per share, from $22.2 million, or $0.41 per share, a year earlier, driven by stock-based compensation and warrant accounting.
Core gross margin reached 41.0%, 200 basis points above the guidance midpoint and up 40 basis points sequentially. Management had guided the third quarter as the trough for margins, after core gross margin fell from 46.5% in the first quarter to 40.6% in the second because rented-back systems cost more to operate. The third quarter confirmed the trough and held above it.
The mix inside that margin matters. Hardware core gross margin fell to 31.0% from 38.8% on the final shipments at legacy pricing. Cloud margins carried the quarter, which shows that the higher-value inference business is improving while the hardware line resets.
Operating costs rose. Core operating expenses increased 7% sequentially to $127.7 million, with research and development at $96.8 million. Core operating loss held flat at $33.1 million, a negative 14.3% margin, a sharp improvement from the negative 25% to negative 23% guidance range and from negative 16% in the second quarter.
The quarter answered the question the market was asking after the August report. In the second quarter, Cerebras beat on core revenue but reported a GAAP loss of $2.98 per share against a $0.18 estimate, and the stock fell 15.7% after hours. This time the company beat on every core metric and delivered positive adjusted net income ahead of schedule.
Cloud Revenue Up 272%: The Inference Engine Takes Over
The single most important number in the report is cloud revenue, and it confirms that Cerebras has become a cloud company with a hardware business rather than the reverse.
Core cloud and other services revenue rose 272% year over year to $144.9 million. GAAP cloud revenue grew 265%. Cloud now accounts for 63% of core revenue, up from a minority share a year ago. Core hardware revenue fell to $86.0 million from $96.8 million a year earlier, though it rose 5% sequentially.
That mix shift explains why the margin story is improving. Inference, the process of running trained AI models to generate responses and perform tasks, has become the company’s primary growth driver. Cerebras sells speed: its wafer-scale processors run large language models faster than GPU-based systems, and that speed commands premium pricing for latency-sensitive applications like coding assistants and agentic workflows.
The customer base is broadening. Cerebras signed cloud capacity agreements with AI coding companies including Cognition and Lovable earlier this year, and its inference platform serves customers including Block, Figma, AlphaSense and GSK. It has also opened a security segment with CrowdStrike, which uses Cerebras inference for inline security scanning of enterprise traffic. Management said its pipeline for high-speed AI inference increased by more than 10 times in the quarter.
The company delivered what it described as the first production-scale heterogeneous disaggregated inference offering, a configuration that splits the work of running AI models across different types of hardware. That product broadens the addressable market beyond customers who want Cerebras systems alone.
Capacity is the constraint. Cerebras now has 700 megawatts of data center capacity under contract. Chief Financial Officer Bob Komin said fourth-quarter cloud revenue will be flat sequentially because new capacity arrives too late to materially affect the period. Demand is outrunning the company’s ability to deliver compute, which makes data center construction the gating factor for growth.
That gating factor is also why the company is shifting from rented capacity to owned capacity. Rented-back systems compressed margins in the second and third quarters. As Cerebras-owned systems come online, management expects margins to rise toward a long-term target of 60% or higher. The fourth-quarter margin guide of 47% to 49% is the first evidence of that transition.
The cloud model also changes how investors should value the company. Cloud revenue is recurring and contracted, while hardware revenue is lumpy and depends on purchase cycles. A business that is 63% cloud, growing at 272%, with a backlog measured in tens of billions of dollars, deserves a different multiple than a pure chip vendor. The challenge is proving that the backlog converts to revenue on schedule.
Guidance Raised: Q4 at $310 Million to $330 Million and FY26 at $950 Million
The guidance raise was larger than the beat, and that is what drove the after-hours surge.
Cerebras guided fourth-quarter core revenue to $310 million to $330 million, against a $271 million consensus. At the $320 million midpoint, that is $49 million above expectations, an 18% beat. Sequentially, the guide implies 39% growth from the third quarter’s $230.9 million. Core gross margin is guided to 47% to 49%, against a 42.1% consensus and up from 41% in the third quarter. Core operating margin is guided to negative 1% to positive 1%, effectively breakeven.
The full-year raise is equally sharp. Cerebras lifted 2026 core revenue guidance to $940 million to $960 million from $880 million to $890 million, against an $887.04 million consensus. That marks the second raise of the year: in May the company guided $855 million to $865 million, and in August it raised the range to $880 million to $890 million. The new $950 million midpoint sits 7% above the August midpoint and 10% above the May midpoint.
Full-year core gross margin guidance rose to 42% to 44% from 41% to 43%. Full-year core operating margin guidance improved to negative 10% to negative 8% from negative 19% to negative 17%, a nine-point improvement in a single quarter.
The fourth-quarter guide carries a mix shift. With cloud revenue flat sequentially due to capacity timing, the $79 million to $99 million sequential revenue increase comes from hardware. That means hardware shipments at current pricing, rather than the legacy-priced shipments that weighed on the third quarter, will drive the fourth quarter. Hardware at current pricing carries higher margins, which explains the jump in the gross margin guide.
The 2027 target is the headline. Chief Executive Andrew Feldman said the company expects revenue to triple in 2027 as next-generation hardware enters production and international facilities come online. Tripling the 2026 midpoint implies $2.85 billion in 2027 revenue, slightly below the $2.92 billion consensus. Management expects core gross margin to progress toward its 60% long-term target in 2027 and expects significant progress toward its 40% long-term operating margin target during the year.
That small gap between management’s implied number and consensus is one reason the stock faded. A $2.85 billion target against $2.92 billion consensus gives no upside surprise for 2027. The quarter beat and the fourth-quarter guide were strong, but the long-term framework simply confirmed what the market already modeled.
The guidance pattern supports credibility. Cerebras has beaten its own guidance for three straight quarters and raised its full-year outlook twice. A management team that guides conservatively and then beats earns a premium over time. The fourth-quarter guide sets a high bar, and delivering it would mark the company’s fourth consecutive beat.
Microsoft Becomes the Third Hyperscaler
The Microsoft announcement addressed the biggest structural concern hanging over Cerebras: customer concentration.
Feldman said Cerebras signed an important cloud deal with Microsoft, which along with previously announced agreements with Amazon Web Services and Oracle brings the company to three hyperscalers using its platform as part of their inference strategy. That makes three hyperscaler wins in the past six months. Feldman said Microsoft signed in October and that revenue begins to show up in the first quarter of 2027. Cerebras expects to disclose more details on the agreement in the first half of 2027.
The concentration problem has been the stock’s defining risk. In 2025, Mohamed bin Zayed University of Artificial Intelligence accounted for 62% of revenue and G42 for 24%. In 2026, the OpenAI agreement became the dominant contract, with more than half of the company’s remaining performance obligations tied to it. A single customer delay can reshape the entire revenue trajectory.
Three hyperscalers change that math. AWS, Oracle and Microsoft each operate global cloud platforms with the scale to resell Cerebras inference to thousands of enterprise customers. The AWS partnership will make Cerebras’ disaggregated inference offering generally available through Amazon Bedrock in the first quarter of 2027. Each hyperscaler deal diversifies revenue away from OpenAI and creates a distribution channel that Cerebras could not build alone.
The deal also validates the technology. Microsoft is one of OpenAI’s largest backers and runs Nvidia hardware at enormous scale. Its decision to add Cerebras inference to its platform signals that wafer-scale processors offer something GPU clusters do not, at least for latency-sensitive workloads. Feldman told analysts that competitors are struggling to keep up with customer demand and dealing with growing order backlogs.
The timing matters for the stock. The Microsoft deal lands in the same week Nvidia’s chief executive resigned. Investors debating whether the leadership crisis at the dominant AI chipmaker creates openings for alternatives now have a concrete example of a hyperscaler diversifying its inference stack.
The limits are clear. Cerebras disclosed no financial terms for the Microsoft deal, and analysts on the call could not get deal metrics. Revenue begins only in the first quarter of 2027, so it contributes nothing to the 2026 numbers. Investors are being asked to value a contract they cannot size.
For the forecast, the Microsoft deal shifts the risk profile more than the near-term numbers. It reduces dependence on OpenAI, broadens the customer base and supports the 2027 tripling target. The first quarterly revenue contribution in early 2027 will be the first test of how large the deal is.
OpenAI: The $20 Billion Contract and Its Delays
OpenAI remains the largest single variable in the Cerebras story, and the earnings call revealed both its scale and its friction.
Cerebras’ multi-year agreement with OpenAI is valued at $20 billion and includes a 750-megawatt capacity commitment. More than half of the company’s remaining performance obligations belong to OpenAI. Those obligations totaled $26.1 billion at the end of the third quarter, up from $25.4 billion in the second quarter.
The contract has run into physical constraints. Feldman said delivery delays under the OpenAI agreement stemmed from industry-wide shortages of land and electricity for data centers rather than from chip manufacturing. Construction delays related to land, power and building shells pushed initial capacity deliveries for OpenAI behind schedule. Management factored potential delay credits into its remaining performance obligations and projected a return to the OpenAI delivery schedule in the second half of 2027.
That timeline shift drew caution from analysts. One major brokerage had expected the first tranche of OpenAI capacity to be fully online by the beginning of next year. The call indicated that most of the new capacity will come online from mid-2027 instead. Feldman also said Cerebras is evaluating financing options to meet ramping demand, which raises the prospect of capital raises.
The delays mirror problems across the AI infrastructure sector. Nebius fell 17.72% on Thursday after delays at third-party data center sites pushed capacity into 2027 and forced the company to trim its 2026 revenue guidance. CoreWeave has also cited delays in data center electrical infrastructure. Land and power, not chips, have become the binding constraint on AI deployment.
The delay credits deserve attention. If Cerebras owes OpenAI credits for late delivery, some of the $26.1 billion in obligations will convert to revenue at lower effective prices. Management has accounted for that in its numbers, but the exact size of the credits is not disclosed.
The upside case for OpenAI is the take-or-pay structure. Cerebras’ contract obligates OpenAI to pay for committed capacity, which protects revenue once capacity is delivered. Core cloud and services revenue, which reflects the OpenAI deal, more than tripled in the third quarter, so the contract is already generating material revenue even before the full 750 megawatts come online.
The OpenAI relationship also carries an equity link. OpenAI’s warrants to buy 33.4 million Cerebras shares vest as it purchases compute capacity. That aligns the two companies but also creates dilution as the contract scales.
For the forecast, OpenAI is the swing factor for 2027. Delivery on the second-half 2027 schedule supports the tripling target. Further slippage would push revenue into 2028 and pressure the stock.