Cerebras (CBRS.US) Beats Expectations on Both Results and Guidance; Morgan Stanley: Strong Hardware Sales Offset Cloud Business Deployment Delays, Maintains US$178 Price Target

Morgan Stanley released a research note after AI chipmaker Cerebras Systems (CBRS.US) reported its latest quarterly results, stating that the company's core financial metrics comprehensively exceeded market expectations, with strong hardware sales driving fourth-quarter revenue and profitability guidance significantly better than expected. However, as the core business supporting the company's long-term growth prospects, large customers' cloud compute deployment progress has fallen behind contractual agreements, bringing some uncertainty to future revenue growth. Nevertheless, Morgan Stanley maintained its "Overweight" rating and US$178 price target on the stock, implying approximately 82% upside from its November 12 closing price of US$97.75. The results showed that Cerebras' third-quarter core revenue grew 70.1% year-over-year to US$230.9 million, up 10% quarter-over-quarter, above the market expectation of US$214.7 million and also exceeding the company's prior guidance range of US$214 million to US$216 million. On a GAAP basis, third-quarter revenue grew approximately 55% year-over-year to US$210.8 million. On profitability, Cerebras' third-quarter adjusted gross margin was 41%, above the market expectation of 39.2%; adjusted operating margin was -14.3%, significantly better than Morgan Stanley's prior estimate of -24.4%; adjusted earnings per share was US$0.12, versus the market's prior expectation of a loss of US$0.15 per share and Morgan Stanley's estimate of a loss of US$0.21 per share. By segment, Cerebras' third-quarter core cloud computing and other services revenue grew 272.3% year-over-year to US$144.9 million, up 13.5% quarter-over-quarter, but below the market expectation of US$153.4 million. Core hardware revenue declined 11.2% year-over-year to US$86 million, up 4.6% quarter-over-quarter, significantly above the market expectation of US$61.2 million. On guidance, Cerebras expects fourth-quarter core revenue of between US$310 million and US$330 million, with a midpoint of US$320 million, up 38.6% quarter-over-quarter and 86.7% year-over-year, well above the market expectation of US$274.9 million. Adjusted gross margin is expected to reach 47% to 49%, with a midpoint of 48%, above the market expectation of 41.2%; adjusted operating margin is expected to be between -1% and 1%, with operating breakeven potentially achievable. Meanwhile, Cerebras raised its full-year 2026 core revenue guidance to US$940 million to US$960 million, with core gross margin expected at 42% to 44% and core operating margin expected at -10% to -8%. Management still expects 2027 revenue to reach three times that of 2026. Morgan Stanley analyst Joseph Moore noted that Cerebras' reported financial data was strong, with new hardware shipments driving fourth-quarter revenue guidance well above expectations. As early hardware sales agreements with lower pricing gradually expire and new cloud business secures more favorable pricing, the company's gross margin improvement is also faster than previously expected. However, the analyst said that compared with a few months ago, Morgan Stanley has become more cautious on the core investment thesis supporting Cerebras' long-term growth, mainly due to delays in cloud compute deployment progress by the company's largest customer. Cerebras disclosed that due to constraints in data center site selection, power supply, and building facility delivery, the first batch of cloud compute deployment under the company's largest contract has fallen behind the original schedule. Although chip and hardware supply is sufficient, data center infrastructure is not yet ready, leading the company to expect fourth-quarter cloud business revenue to be roughly flat versus the third quarter. Morgan Stanley had previously expected the first batch of compute under the relevant contract to be fully operational in early 2027, but based on the latest management disclosure, most of the additional compute is expected to come online gradually starting mid-2027. Currently, Cerebras has contracted compute infrastructure with total capacity of 700 megawatts (MW) scheduled for delivery by the end of 2027, up from 600 MW disclosed in the previous quarter. The company expects to catch up with the contractually agreed delivery schedule in the second half of 2027. Notably, because the relevant cloud compute deployment involves binding long-term contracts, delivery delays could expose Cerebras to certain economic penalties and even put some contracted capacity at risk of cancellation. As of the end of the third quarter, Cerebras' remaining performance obligations (RPO) reached US$26.1 billion, up from US$25.4 billion at the end of the second quarter. This figure already deducts customer credits expected to arise from delivery delays and does not yet include new cooperation agreements with Oracle Corporation (ORCL.US) and Microsoft Corporation (MSFT.US). Although Cerebras did not disclose the specific amount of the relevant credits, the company said the amount is small relative to total remaining performance obligations and is expected to have limited impact on revenue and gross margin, as the credits will be amortized over the remaining contract term. Based on its own delivery schedule model and related cost assumptions, Morgan Stanley estimates that delivery delays could reduce Cerebras' remaining performance obligations by US$600 million to US$900 million, with the midpoint equivalent to approximately 3% of its total remaining performance obligations. The firm further noted that, to date, no cancellations have occurred under the relevant contracts. However, under the contract terms, if insufficient compute delivery persists for at least 90 days, both parties have the right to cancel the affected portion of capacity. Cerebras expects that capacity at risk of cancellation in 2026 will represent less than 5% of total contracted committed capacity. Morgan Stanley believes that, relative to economic penalties arising from delayed delivery, what is more noteworthy is whether long-term contract revenue can be realized as expected. Although the relevant contracts are structured on a "take-or-pay" basis, actual revenue still depends on whether Cerebras can meet the agreed delivery milestones. If compute deployment continues to lag, it could not only delay revenue recognition but also affect the ultimate scale of contract revenue that can be realized. Based on the slower cloud compute deployment progress, Morgan Stanley lowered its 2027 cloud business revenue forecast for Cerebras but still expects the cloud compute capacity coming into operation in the second half of 2027 to be approximately 500 MW. At the same time, new customers and hardware orders recently secured by Cerebras are becoming an important force supporting short-term performance growth. The company announced a new cloud computing cooperation agreement with Microsoft Corporation, with related revenue expected to begin contributing in the first half of 2027. Following Amazon.com, Inc.'s (AMZN.US) AWS and Oracle Corporation, Microsoft Corporation becomes the third major cloud computing customer Cerebras has secured in the past six months. In addition, new hardware orders including those from Oracle Corporation will begin shipping in the fourth quarter of this year and continue at least through the first half of 2027. Oracle Corporation's cloud infrastructure (OCI) AI inference service based on Cerebras technology is expected to launch in summer 2027. Morgan Stanley believes that partnerships with Microsoft Corporation, Oracle Corporation, and AWS not only reflect major technology companies' recognition of Cerebras' technology but also help the company expand business opportunities and reduce dependence on a single customer. However, hardware business revenue still depends on the delivery timing of individual large orders and the construction progress of customer data centers, making it relatively less predictable. This means that, against the backdrop of delayed core cloud business deployment, Cerebras' short-term performance growth will rely more on volatile hardware sales. On technology progress, Morgan Stanley remains optimistic about Cerebras' competitive advantage in high-speed AI inference. Cerebras adopts a wafer-scale processor architecture that can provide performance and cost advantages different from traditional graphics processing units (GPUs) in high-speed inference scenarios. Traditional GPUs can improve response speed for individual users by reducing the number of simultaneously processed requests, but this also lowers overall hardware processing efficiency, thereby raising unit inference costs. By contrast, Cerebras believes its technology can achieve high-speed inference while maintaining relatively high overall throughput, offering better cost efficiency in application scenarios generating approximately 300 tokens per second per user, and its advantage is expected to widen as inference speed requirements increase further. In addition, the disaggregated AI inference solution jointly developed by Cerebras and AMD (AMD.US) has begun serving workloads in actual production environments. Cerebras said the solution can achieve up to approximately five times the token generation under the same power consumption conditions. On next-generation products, Cerebras' CS4 system has begun shipping, and the next-generation CS5 processor has entered system testing and begun running AI models, with production expected in early summer 2027. Morgan Stanley believes the above technological progress is expected to further enhance Cerebras' performance advantages and unit economics once the compute facilities gradually come into operation in the future. Based on the latest results and business progress, Morgan Stanley raised Cerebras' 2026 core revenue forecast from US$886 million to US$952 million and its 2027 core revenue forecast from US$2.88 billion to US$3 billion, while basically maintaining its 2028 revenue forecast unchanged at an expected US$6.67 billion. On earnings forecasts, the firm raised Cerebras' 2026 adjusted earnings per share forecast from a loss of US$0.46 to a profit of US$0.08, and its 2027 adjusted earnings per share forecast from US$1.03 to US$1.81, with further growth to US$5.54 expected in 2028. Morgan Stanley also raised Cerebras' 2027 core gross margin forecast from 52.1% to 54.4% and its 2028 core gross margin forecast from 58% to 59.3%. The firm expects that, as business scale expands and profitability improves, the company's 2027 adjusted operating margin will reach 19.8%, significantly above its prior estimate of 10.9%. For stock valuation, Morgan Stanley maintained its US$178 price target unchanged, based on Cerebras' 2028 adjusted revenue of approximately US$6.3 billion after deducting customer warrant-related revenue offsets, applying an 8x price-to-sales valuation. The firm believes that new major customers, improved profitability, and AI inference technology progress are sufficient to support Cerebras' current share price, but for the stock to return to previous highs, the market still needs greater confidence in its largest customer's cloud compute deployment progress. In a bull case, if contracted compute capacity can come into operation faster than expected and hardware demand further exceeds expectations, Morgan Stanley believes Cerebras' share price could reach US$330. Conversely, if cloud compute deployment continues to be delayed, equipment utilization is below expectations, and hardware demand is weak, the stock's valuation in a bear case could fall to US$40. Morgan Stanley said that although the latest results and upward earnings forecast revisions support maintaining the "Overweight" rating, it will remain relatively cautious on Cerebras' long-term growth prospects until cloud compute delivery progress catches up with the contract schedule and related revenue begins to be realized as expected.