CBRS Q1 Revenue Exceeds Expectations, But Q2 Gross Margin Guidance Drops Sharply

Author: David, Tide Research

Market Overview: Cerebras (CBRS) released its first quarterly report since its IPO, reporting Q1 core revenue of $191 million, a 92% year-over-year increase, surpassing market expectations. However, guidance for Q2 core gross margin dropped sharply from 46.5% to 36%-38%, causing the stock to fall over 10% after hours. This company, which uses an entire wafer to manufacture chips and is betting heavily on the AI inference market, holds over $20 billion in contracts with OpenAI and a collaboration framework with AWS, with full-year revenue guidance set at $855–865 million. The growth figures are strong, but so too are the valuation debates.

1. Revenue exceeded expectations, and guidance surpassed expectations even further. Q1 core revenue was $191.3 million (up 92% year-over-year), higher than the consensus estimate of approximately $181 million. Full-year core revenue guidance is set at $855–865 million (up 69% year-over-year), above the market expectation of $828 million. Under GAAP, cloud and services revenue reached $82.8 million, a 178% year-over-year increase, making it the fastest-growing segment.
2. The sharp reduction in gross margin guidance was the biggest negative of the quarter. Q1 core gross margin was 47%, up nearly 5 percentage points year-over-year. However, Q2 guidance was lowered to 36%-38%, a decline of approximately 10 percentage points from Q1; the full-year guidance is set at 38%-41%. Management attributed the decline to insufficient data center capacity: the company is temporarily leasing back systems from existing customers who have already purchased hardware to deploy additional capacity, worsening short-term costs. The stock fell more than 10% after hours.
3. Customer concentration has improved but remains far from resolved. In fiscal year 2025, 86% of revenue came from two affiliated entities in the UAE (MBZUAI at 62% and G42 at 24%). OpenAI began contributing revenue in February 2026, and revenue from the AWS partnership is not expected to appear in financials until 2027. True revenue diversification will not be validated until 2027.
4. Valuation is priced through 2028. At a post-market price of approximately $200, CBRS trades at about 90 times trailing twelve-month revenue; even using the midpoint of the full-year guidance of $860 million, the forward P/S ratio remains above 50x. The median target price among 10 covered analysts is $300 (range: $250–$340), implying that OpenAI’s contracts exceeding $20 billion and AWS deployment commitments will be fulfilled on time and at scale.
5. Short-term catalysts and headwinds coexist. Catalysts: Accelerated deployment of 750 MW of computing power by OpenAI, implementation of AWS inference solutions, and new data center capacity coming online in the second half of the year. Headwinds: Lock-up period includes unusual early release clauses (triggered when market capitalization exceeds $40 billion, a threshold the current market cap has nearly reached), unclear path to gross margin recovery, and OpenAI’s own lack of profitability alongside reductions in certain computing commitments.

Q1 The most easily overlooked aspect in financial reports is the change in revenue structure.

Under the core metric, hardware revenue was $111.6 million, accounting for 58% of total revenue; cloud and service revenue was $79.8 million, accounting for 42%. A year ago during the same period, this ratio was approximately 70:30. Cloud service revenue increased by 167% year-over-year, nearly triple the growth rate of hardware revenue.

Management clarified this trend more explicitly during the earnings call:

Hardware revenue will decline temporarily over the next few quarters as the company redirects more hardware capacity to its own cloud infrastructure to fulfill inference computing contracts with OpenAI and AWS, rather than selling directly to customers. Cerebras is transitioning from a company that sells chips to a company that sells compute power.

This transition also directly explains why the gross margin declined sharply in Q2.

During the conference call, an analyst pressed for details on capacity deployment, and management disclosed:

The company’s current bottleneck is not in TSMC’s chip supply, but in the physical space of data centers. To deliver computing power to OpenAI as quickly as possible, Cerebras is temporarily leasing back previously sold hardware systems from G42—its former largest customer and a minority investor.

Renting third-party facilities to deploy our own systems will temporarily worsen the cost structure, which is the primary reason the gross margin guidance has been lowered from 47% to 36%-38%. Management expects cost pressures to ease starting in the second half of the year when the new data centers come online.

The financial structure of OpenAI’s contract is also worth unpacking. On the surface, it appears to be a multi-year procurement of computing power exceeding $20 billion, but beneath it lie three layered relationships: OpenAI provided Cerebras with a $1 billion operating capital loan (reflected on its Q1 balance sheet as $621 million in current loans and $362 million in non-current loans), while simultaneously receiving warrants in Cerebras.

In other words, OpenAI serves as Cerebras’s largest customer, creditor, and potential shareholder simultaneously. According to the risk disclosures in the S-1, if Cerebras fails to deliver the agreed-upon capacity, OpenAI has the right to terminate the contract and trigger loan repayment.

AWS’s collaboration framework employs a “split inference” architecture: AWS’s Trainium 3 chip handles prompt input (prefill stage), while Cerebras’s CS-3 system is dedicated exclusively to high-speed output generation (decode stage). This design allows Cerebras to avoid承担 the full inference pipeline, focusing only on the stage where it has the greatest speed advantage. However, management declined to disclose the specific scale of the AWS collaboration during the Q&A session, stating that revenue contributions will not be reflected in financial results until 2027.

The common characteristic of these two large orders is that they involve massive contract volumes, but have lengthy fulfillment paths and are highly dependent on Cerebras's data center construction progress.

The annual revenue guidance of $855 million to $865 million implies an average of approximately $220 million per quarter for the final three quarters, with accelerating growth each quarter. Management stated that “quarter-over-quarter year-over-year growth will increase throughout 2026, with more revenue concentrated in the second half of the year.”

On June 8, the day the IPO quiet period ended, nine underwriters simultaneously initiated coverage, all assigning buy or outperform ratings. CBRS rose 18.3% on that day. Such synchronized bullish sentiment—like opening the floodgates—is not uncommon in U.S. IPOs (given underwriters’ inherent financial incentives), but their collective bets point to the same core thesis.

Proposition one: The battlefield for AI computing power is shifting from training to inference, and the rules of competition in inference scenarios differ from those in training.

Morgan Stanley analyst Joseph Moore issued a Buy rating with a $250 price target in his first coverage report on June 8. His core argument is that training scenarios are determined by total computational throughput, where NVIDIA’s GPU clusters hold an absolute dominant position; in inference scenarios, the key metrics are response speed and latency, as models must handle millions of user requests per second, where even minor differences in speed directly impact service costs and user experience. Cerebras’s wafer-scale chip has a structural advantage in inference latency due to its on-chip SRAM capacity far exceeding that of conventional GPUs, eliminating the need for frequent data transfers to external memory. Moore stated that Cerebras is “the only company to have commercially deployed wafer-scale processors,” giving it a first-mover advantage over NVIDIA.

Citi analyst Atif Malik set the highest target price of $340 among covered analysts. Mizuho added a technical detail in its June 8 report: the WSE-3 chip integrates 44 GB of SRAM, several times that of Google’s latest TPU and Groq’s LPU—a hardware advantage that cannot be bridged by architectural optimizations in the short term.

Proposition two: Two large orders have moved Cerebras from a "technology story" to an "revenue story."

The OpenAI contract exceeds $20 billion, covering 750 MW of inference capacity over a multi-year delivery period. Amortized over five years, this single contract alone contributes approximately $4 billion in annual revenue—nearly five times the midpoint of the full-year 2026 revenue guidance. Although management has declined to disclose specific figures for the AWS partnership, the framework has been confirmed: Cerebras’s inference capabilities will be made available to enterprise customers worldwide via Amazon Bedrock.

Q1 financial data provided early validation. OpenAI began deploying Cerebras systems in February, causing cloud service revenue to surge from under $30 million year-over-year to nearly $80 million within a single quarter. Management stated that "year-over-year growth will accelerate each quarter in 2026, with more revenue concentrated in the second half," and issued a full-year guidance of $855–865 million, above the consensus estimate of $828 million.

Proposition three: The coverage density after the quiet period ends is itself a signal.

The median target price among 10 analysts is $300, with a low of $250 (Morgan Stanley) and a high of $340 (Citigroup). Based on the after-hours price of $200, the median target implies approximately 50% upside potential. Wedbush ($270), Needham ($300), Barclays ($280), TD Cowen ($275), and Craig-Hallum (Buy) all initiated coverage within the same week.

The underlying assumption of the bullish thesis can be summarized in one sentence:

If AI inference becomes a larger compute market than training (multiple institutions predict that inference spending will exceed training by 2027), and Cerebras’s speed advantage is real and sustainable, then capturing just 3% to 5% of NVIDIA’s 80%+ market share would be sufficient to support its current valuation.

Three bullish theses, each met with a bearish counterargument.

Counterargument 1: The moat provided by inference speed advantages may be narrower than assumed.

Cerebras's speed advantage is built on its on-chip SRAM capacity, but NVIDIA is not standing still. NVIDIA’s B300 chip, released in March, significantly increases HBM bandwidth, and Groq’s LPU architecture is also rapidly evolving for inference scenarios.

From another perspective: Cerebras’s customers are currently heavily concentrated among just two companies—OpenAI and AWS. Meanwhile, OpenAI is also one of NVIDIA’s largest GPU buyers, and AWS’s in-house Trainium chips are increasingly being deployed across more inference workloads. The fact that Cerebras’s major clients are simultaneously investing in alternative solutions means its speed premium will continue to face pricing pressure.

Counterargument two: The gross profit margin decline may not be just "temporary."

The management attributes the reduction in Q2 gross margin from 47% to 36%-38% to temporary leasing costs caused by insufficient data center capacity. However, this explanation assumes that costs will improve after new data centers come online in the second half of the year.

Given that revenue scale is expected to surge in the second half of the year (management has explicitly stated that revenue will be back-end loaded), and that ramping up capacity at the new data center itself requires time and capital investment, this recovery path is not straightforward.

A deeper issue is the impact of the business model transition itself on gross margins. Cerebras’ shift from selling hardware to selling cloud computing power means assuming the costs of building, operating, and depreciating data centers. As depreciation expenses from its self-built data centers are recognized, it remains uncertain whether the gross margin for cloud services can remain above 50%. The profit margin ceiling for this business model has yet to be tested.

Counterargument three: Customer concentration is a "rebranded, not resolved" issue.

In 2024, G42 accounted for 85% of Cerebras’s revenue. In 2025, G42’s share dropped to 24%, while MBZUAI (Mohamed bin Zayed University of Artificial Intelligence) surged from zero to 62%. The S-1 filing explicitly labels both as “related parties.” Together, these two UAE-related entities still account for 86% of revenue. The diversification of revenue sources is more a matter of name changes than actual dispersion.

Finally, CBRS's IPO lock-up period includes an unusual clause:

If the company’s market capitalization remains above $40 billion, insider shares may be unlocked early. At the after-hours price of $200, the current market cap is approximately $45 billion, nearing the trigger threshold. Regarding short positions, as of May 29, the short interest stood at 17.15% of the float, which is relatively high. If insider shares are released early, adding to existing short-selling pressure, the stock could face concentrated selling pressure.