Cerebras Systems Q2 core revenue doubles to $210m as cloud business quadruples

Cerebras Systems Inc. (NASDAQ: CBRS) reported second-quarter 2026 results on August 12, showing core revenue of $209.9 million, up 103% year over year, and core cloud and other services revenue of $127.7 million, up 287%. The Sunnyvale wafer-scale AI infrastructure company also raised full-year 2026 core revenue guidance to $880 to $890 million and disclosed remaining performance obligations of $25.4 billion at quarter-end. Shares had rallied 11.5% during the regular session to $261.76 but fell 15.7% after hours to $221.01, as GAAP revenue of $180.1 million came in below Street estimates near $194 million and GAAP net loss widened to $450.5 million, driven largely by post-IPO stock-based compensation. The reaction crystallises the central tension for the newly public company: extraordinary top-line growth and a very large forward book set against stretched multiples, a widened headline loss and the market’s continuing effort to price the gap between GAAP results and Cerebras Systems’ preferred core measures.

Why Cerebras Systems posted a GAAP revenue miss even as core revenue crossed $200 million

The GAAP-versus-core divide is central to any honest read of the quarter. On a GAAP basis, total revenue was $180.1 million, up 74% year over year but around $14 million short of consensus. On a core basis, which excludes pass-through data-centre revenue and adds back non-cash amortisation of customer warrant assets that is recorded as contra-revenue, total revenue was $209.9 million, more than doubling from $103.3 million a year earlier and slightly above the company’s own $194 million core revenue guide. The gap between the two numbers is not accounting sleight of hand; it reflects real accounting treatment of the customer warrants Cerebras Systems has issued to strategic buyers, most notably in connection with large capacity commitments. However, it does mean that different desks measured the release against different benchmarks. Sources that framed the print as a GAAP miss reached one conclusion; sources measuring against the core guide reached another. Both are defensible readings and both explain why post-market trading was so volatile.

What the near-quadrupling of cloud revenue signals about inference demand at scale

The cloud line is the number that matters most for the medium-term thesis. GAAP cloud and other services revenue reached $126.0 million, up 281% year over year, while core cloud and other services revenue reached $127.7 million, up 287%. Cloud now represents roughly 70% of GAAP revenue and about 61% of core revenue, up from around 32% of GAAP revenue in the prior-year quarter. The reason management describes speed as an unlocking factor is that fast token generation lets customers deploy inference in latency-sensitive workflows that GPU clusters cannot service cost-effectively, including inline security, real-time agentic loops and coding assistants. Cerebras Systems cited Cognition and Lovable as new cloud capacity customers among AI coding companies, and pointed to Block, Figma, AlphaSense and GSK as examples of production agentic use cases across finance, design, market intelligence and life sciences. The mix shift toward cloud is what gives management confidence to guide 2027 revenue to more than triple current-year levels, though delivery depends on data-centre commissioning cadence rather than on demand.

How the $25.4 billion remaining performance obligation reshapes the 2027 revenue outlook

Cerebras Systems reported remaining performance obligations of $25.4 billion as of June 30, up sharply from the balance implied by prior filings and dominated by the multi-year OpenAI capacity agreement disclosed alongside first-quarter results, under which OpenAI plans to deploy 750 megawatts of Cerebras inference computing capacity. RPO is a contractual, non-cancellable book of business, so it functions as a visibility floor rather than a revenue forecast; the timing of conversion depends on when data-centre capacity comes online, how quickly customers ramp usage against contracted commitments and, for warrant-linked contracts, on how the associated amortisation flows through GAAP revenue. Against 2026 guidance of $880 to $890 million core revenue and management’s stated intention to more than triple revenue in 2027, the RPO is more than sufficient to support the trajectory in principle. The credible risk is execution slippage on the megawatts, not customer demand.

Why the widened GAAP net loss looks worse than the underlying operating trajectory suggests

GAAP net loss of $450.5 million for the quarter is a striking number, and it is why the market focused on the EPS print of negative $2.98 versus consensus near negative $0.17. Roughly $377 million of that loss is stock-based compensation, most of which relates to equity awards that vested or were granted in connection with the May 2026 IPO. Amortisation of customer warrant assets added a further $44 million against gross profit. On a core basis, the operating loss was $33.6 million, an improvement from a $43.9 million core operating loss a year earlier, and core operating margin narrowed from negative 42% to negative 16%, an approximately 2,600-basis-point improvement. Core gross margin expanded to 40.6% from 31.2%, an approximately 940-basis-point improvement. Investors will differ on whether to treat the IPO-vintage stock-based compensation as a one-time distortion or as a recurring cost of running an AI infrastructure business at Cerebras Systems’ scale, but the direction of underlying margin improvement is not in dispute.

What the OpenAI, AMD, AWS and CrowdStrike arrangements collectively imply for competitive positioning

The commercial base disclosed with the quarter is broader than a single OpenAI relationship. Cerebras Systems is the launch partner for OpenAI’s GPT-5.6 Sol frontier model at 750 tokens per second, and the multi-year OpenAI capacity agreement remains the anchor. Alongside it, the company disclosed a disaggregated inference architecture with AMD that management said will deliver Cerebras speed while increasing throughput by up to five times, targeted for production in the fourth quarter of 2026. The AWS relationship was described as deepened, with Cerebras disaggregated inference and the same throughput benefit expected to reach Amazon Bedrock in the first quarter of 2027. Bedrock exposure would substantially broaden the addressable enterprise base by removing a distribution barrier. The CrowdStrike partnership carves out a new segment, using fast inference to enable inline security using large language models across a large share of enterprise traffic, and represents an interesting proof point that inference speed can create categories rather than only accelerate existing workloads. The collective read is that Cerebras Systems is positioning itself as inference-first infrastructure rather than as a hyperscaler-substitute training platform, and the AMD and AWS routes reduce the counterparty concentration risk that a pure OpenAI story would carry.

How the wafer-scale architecture insulates Cerebras Systems from HBM, CoWoS and three-nanometre bottlenecks

Management again emphasised the supply-chain shape of its wafer-scale approach: no high-bandwidth memory, no CoWoS packaging, and no reliance on three-nanometre fabrication, each of which is currently supply-limited across the AI accelerator ecosystem. Contract manufacturing capacity was expanded across Flex, Sanmina and Rocket EMS, and TSMC wafer supply for continued growth has been secured. Manufacturing capacity is expected to scale more than 10 times in 2026. This is not a marketing point; it is the specific reason Cerebras Systems believes it can convert its RPO on the timeline it has set, while competitors dependent on HBM3e allocation or advanced packaging capacity may face structural delivery constraints. The scarcity of these inputs is well documented across the industry, and if the supply environment for HBM and CoWoS remains tight through 2027, the wafer-scale route becomes a comparative advantage rather than merely an architectural preference. However, this advantage narrows if HBM and CoWoS capacity opens up faster than currently expected or if GPU inference economics improve materially.

What the $6.4 billion IPO raise, $850 million revolver and 600 megawatts under contract permit operationally

Cerebras Systems raised $6.4 billion in gross IPO proceeds during the second quarter and closed a revolving credit facility for up to $850 million to accelerate data-centre acquisitions. Total liquidity at quarter-end stood at $8.6 billion in cash, cash equivalents, restricted cash and short-term investments, plus the debt capacity. Data-centre capacity live and under contract for delivery by the end of 2027 exceeded 600 megawatts, with the opportunity pipeline extending into gigawatts. Capital-intensive infrastructure companies are typically judged on the gap between contracted capacity and financing plan, and on that measure Cerebras Systems’ balance sheet appears sized for the near-term build-out, with the working capital loan providing operational elasticity for equipment purchases ahead of customer billings. The company’s cash use pattern in the first half, with $1.3 billion deployed in investing activities including $549 million of property and equipment purchases, gives some sense of the burn associated with capacity commissioning at this scale, though it should moderate as leased and colocation arrangements substitute for owned build.

How the after-hours reaction compares with the pre-earnings run and the IPO peak

Shares of Cerebras Systems entered earnings day trading around $234 and closed the regular session up 11.5% at $261.76, extending a strong run since the May IPO, in which shares had reached an intraday high of $350 before consolidating and then trending lower through mid-summer. The after-hours drop of roughly 15.7% took the stock back to $221.01, near the higher end of the recent trading range but well below the IPO peak. At the pre-drop close, market capitalisation stood at roughly $68 billion; the after-hours print implies a market capitalisation nearer $58 billion. Even at the lower level, price-to-sales on trailing revenue is elevated by any conventional standard, reflecting the market’s willingness to price forward growth into current multiples. The 52-week range of $160.81 to $386.34 conveys the underlying volatility. Business News Today does not treat any single-session move as a verdict on the strategy; the more useful reading is that the shares now embed a specific set of assumptions about the pace of megawatt commissioning, the RPO conversion rate and the trajectory of core operating margin toward breakeven.

What analyst positioning, insider selling and third-party inquiries mean for the investor picture

Sell-side positioning coming into the quarter was broadly constructive, with a majority of covering analysts at Strong Buy or Buy and an average 12-month price target in the high-$290s, above the after-hours price. Insider transactions filed since the IPO have leaned toward selling, with third-party trackers citing approximately $20.7 million of insider sales over the trailing three months and no reported open-market purchases; investors familiar with post-IPO patterns will recognise that early insider selling frequently reflects lock-up mechanics and tax obligations tied to IPO-vintage stock-based compensation rather than a directional signal on the business. Separately, at least one plaintiffs’ law firm publicly announced in July that it was investigating potential securities claims relating to Cerebras Systems disclosures; such investigations are common following high-profile IPOs and do not themselves represent regulatory action, findings of wrongdoing, or the filing of a class action, and Business News Today makes no assessment of the merits of any such inquiry. Cerebras Systems has not disclosed any regulatory enforcement matter or restatement, and the risk factors in its most recent filings include the standard concentration disclosures around OpenAI, Group 42 Holding Ltd, Mohamed bin Zayed University of Artificial Intelligence and AWS.

What should investors track as Cerebras Systems moves from Q2 print to Q3 delivery and 2027 ramp?

• Core revenue of $209.9 million grew 103% year over year and slightly exceeded the company’s own $194 million core guide, while GAAP revenue of $180.1 million fell short of Street expectations near $194 million, creating a mixed print that market participants read very differently by measurement frame.
• Core cloud and other services revenue of $127.7 million grew 287% year over year, taking cloud to roughly 61% of core revenue, and is now the primary growth engine for the business.
• Remaining performance obligations of $25.4 billion provide multi-year visibility, but conversion depends on data-centre commissioning cadence rather than customer demand; delivery risk sits with megawatts, not orders.
• GAAP net loss of $450.5 million was dominated by approximately $377 million of stock-based compensation and $44 million of customer warrant amortisation; core operating loss narrowed to $33.6 million and core gross margin expanded roughly 940 basis points year over year to 40.6%.
• Full-year 2026 core revenue guidance was raised to $880 to $890 million with core gross margin of 41% to 43%; management also stated the intention to more than triple revenue in 2027.
• Data-centre capacity live and under contract for delivery by end-2027 exceeded 600 megawatts with the pipeline extending into gigawatts, supported by $8.6 billion in total liquidity and an $850 million revolving credit facility.
• Manufacturing capacity is expected to scale more than 10 times in 2026 across Flex, Sanmina and Rocket EMS, with TSMC wafer supply secured and the wafer-scale architecture avoiding HBM, CoWoS and three-nanometre bottlenecks.
• Commercial base broadened materially: OpenAI as launch partner for GPT-5.6 Sol at 750 tokens per second, a disaggregated inference architecture with AMD targeted for production in Q4 2026 with up to five times throughput, an Amazon Bedrock integration targeted for Q1 2027, and a new inline security use case with CrowdStrike.
• Shares closed the regular session up 11.5% at $261.76 before falling 15.7% after hours to $221.01, taking implied market capitalisation from roughly $68 billion toward $58 billion; the 52-week range spans $160.81 to $386.34, reflecting the volatility of an early post-IPO name.
• The thesis strengthens with disciplined megawatt commissioning through 2026 and 2027, continued core operating margin improvement toward breakeven and a demonstrable ramp in cloud revenue against RPO; it weakens if supply advantages from wafer-scale narrow, if data-centre delivery slips against contracted timelines, or if the gap between GAAP results and core measures remains a persistent source of market confusion.