CBRS’ First-Ever Earnings Report After IPO: Revenue Doubles, but Gross Margin Guidance Plummets; Path to Fulfilling OpenAI’s Large-Scale Order Remains Long
TechFlow Selected TechFlow Selected
CBRS’ First-Ever Earnings Report After IPO: Revenue Doubles, but Gross Margin Guidance Plummets; Path to Fulfilling OpenAI’s Large-Scale Order Remains Long
From Selling Chips to Selling Computing Power: The $2 Billion OpenAI Contract Is Huge—but Fulfillment Will Take a Long Time
Navigating Web3 tides with focused insightsAuthor: David, TideFlow Research
TideFlow Summary: Cerebras (CBRS) released its first quarterly report since its IPO. Core revenue for Q1 stood at $191 million, up 92% year-on-year—exceeding market expectations. However, the company slashed its Q2 core gross margin guidance from 46.5% to 36–38%, triggering a post-earnings stock decline of over 10%. This company—which builds chips using an entire silicon wafer and bets heavily on the AI inference market—holds an OpenAI contract worth over $20 billion and a collaboration framework with AWS. Its full-year revenue guidance stands at $855–865 million. Growth metrics are robust; valuation debates are equally intense.
Key Focus Areas
- Revenue beats expectations—and guidance beats even more. Q1 core revenue totaled $191.3 million (up 92% YoY), surpassing consensus estimates of ~$181 million. Full-year core revenue guidance is $855–865 million (up 69% YoY), above the market’s $828 million expectation. Under GAAP, cloud and services revenue reached $82.8 million, up 178% YoY—the fastest-growing segment.
- The sharp drop in gross margin guidance is this quarter’s biggest negative. Q1 core gross margin was 47%, up nearly 5 percentage points YoY. Yet Q2 guidance fell sharply to 36–38%, down roughly 10 percentage points from Q1; full-year guidance stands at 38–41%. Management attributed the decline to insufficient data center capacity: the company is temporarily leasing back systems already sold to existing customers—including G42—to deploy capacity, worsening short-term costs. The stock fell over 10% post-earnings.
- Customer concentration shows signs of improvement—but remains unresolved. In FY2025, 86% of revenue came from two UAE-affiliated entities (MBZUAI accounted for 62%; G42, 24%). OpenAI began contributing revenue in February 2026; AWS collaboration is expected to reflect financially only in 2027. Real revenue diversification won’t be validated until 2027.
- Valuation pricing looks out to 2028. At a post-earnings price of ~$200, CBRS trades at ~90x trailing-twelve-month revenue. Even using the midpoint of full-year guidance ($860 million), forward P/S remains above 50x. Among 10 covering analysts, the median target price is $300 (range: $250–$340), implicitly assuming timely and full execution of the OpenAI $20+ billion contract and AWS deployment.
- Short-term catalysts and headwinds coexist. Catalysts: Accelerated deployment of OpenAI’s 750MW compute capacity; AWS inference solution rollout; new data center capacity coming online in H2. Headwinds: Lock-up provisions include an unconventional early-release clause—triggered if market cap exceeds $40 billion (current market cap is already near that threshold); unclear path to gross margin recovery; OpenAI itself remains unprofitable and has begun scaling back certain compute commitments.
Earnings Reveal Business Model Transformation: From Chip Seller to Compute Provider
The most overlooked detail in the Q1 earnings report is the shift in revenue composition.
Under core accounting, hardware revenue was $111.6 million (58% of total revenue); cloud and services revenue was $79.8 million (42%). A year earlier, that split was roughly 70:30. Cloud services revenue grew 167% YoY—nearly three times faster than hardware revenue.
Management made this trend explicit on the earnings call:
Hardware revenue will decline temporarily over the next few quarters, as the company shifts more hardware capacity into its own cloud infrastructure to fulfill OpenAI and AWS inference compute contracts—not sell directly to customers. Cerebras is transforming from “a chip company” into “a compute company.”
This transition also directly explains the Q2 gross margin plunge.
An analyst probed management on capacity deployment details during the call. Management disclosed:
The current bottleneck lies not in TSMC’s chip supply but in physical data center space. To accelerate compute delivery to OpenAI, Cerebras is “temporarily leasing back” hardware systems it had previously sold to G42 (its largest prior customer—and also a minority equity investor).
Renting third-party facilities to deploy proprietary systems worsens short-term cost structure—this is the primary reason gross margin guidance dropped from 47% to 36–38%. Management expects relief beginning in H2 as new data centers come online.
The financial structure of the OpenAI contract merits closer examination. On the surface, it’s a multi-year compute procurement deal worth over $20 billion—but beneath lie three interlocking relationships: OpenAI extended Cerebras a $1 billion operating capital loan (reflected on Q1’s balance sheet as $621 million in current loans and $362 million in non-current loans) and received warrants in Cerebras.
In other words, OpenAI serves simultaneously as Cerebras’ largest customer, creditor, and potential shareholder. As noted in the S-1 risk factors, should Cerebras fail to deliver capacity per schedule, OpenAI may terminate the contract and trigger loan repayment.
AWS’s collaboration framework adopts a “split inference” architecture: AWS’s Trainium 3 chips handle prompt processing (the prefill stage), while Cerebras’ CS-3 system specializes exclusively in high-speed output generation (the decode stage). This design lets Cerebras avoid bearing the full inference pipeline—focusing only on the step where its speed advantage is greatest. Yet management declined to disclose the scale of the AWS partnership during the Q&A, stating revenue contribution won’t appear on financial statements until 2027.
The common thread across both mega-deals is: massive contract size—but long, uncertain execution timelines, highly dependent on Cerebras’ data center construction progress.
The full-year revenue guidance of $855–865 million implies ~$220 million per quarter for the remaining three quarters—with sequential acceleration required. Management stated, “QoQ YoY growth will increase each quarter in 2026, with heavier revenue weighting in H2.”
Bull Case: Nine Investment Banks Simultaneously Initiate Buy Ratings—What Are They Buying?
On June 8—the day the IPO quiet period ended—nine underwriters simultaneously initiated coverage, all issuing Buy or Overweight ratings. CBRS surged 18.3% that day. Such coordinated “opening-the-floodgates” bullishness isn’t unusual among U.S. IPOs (underwriters naturally have aligned interests), yet their underlying logic converges on the same core thesis.
Thesis One: The AI compute battlefield is shifting from training to inference—and competition rules differ fundamentally.
Morgan Stanley analyst Joseph Moore issued an Overweight rating and $250 target price in his June 8 initiation report. His central argument: Training prioritizes aggregate throughput—where NVIDIA GPU clusters dominate absolutely; inference prioritizes per-request latency and speed, because models must process millions of user requests per second—speed directly impacts service cost and user experience. Cerebras’ wafer-scale chips feature on-die SRAM capacity far exceeding conventional GPUs, minimizing data movement to external memory and delivering structural latency advantages in inference. Moore described Cerebras as “the only company to commercialize wafer-scale processors,” giving it a first-mover advantage over NVIDIA.
Citi analyst Atif Malik set the highest target price among initiations—$340. Mizuho’s June 8 note added a technical detail: The WSE-3 chip integrates 44GB of on-die SRAM—several times larger than Google’s latest TPU or Groq’s LPU—a hardware gap unlikely to close via architectural optimization in the near term.
Thesis Two: These two mega-deals propel Cerebras from a “technology story” to a “revenue story.”
The OpenAI contract exceeds $20 billion, covering 750MW of inference compute delivered over multiple years. Amortized over five years, this single contract contributes ~$4 billion annually—nearly five times the midpoint of Cerebras’ FY2026 revenue guidance. While management declined to disclose AWS’s exact dollar value, the framework is confirmed: Cerebras’ inference capabilities will be offered globally to enterprise customers via Amazon Bedrock.
Q1 results provide early validation. OpenAI began deploying Cerebras systems in February; cloud services revenue jumped from under $30 million YoY to nearly $80 million in one quarter. Management reiterated, “QoQ YoY growth will increase each quarter in 2026, with heavier revenue weighting in H2,” and full-year guidance ($855–865 million) exceeded consensus ($828 million).
Thesis Three: The density of coverage immediately after the quiet period itself signals conviction.
The median target price among 10 analysts is $300—ranging from $250 (Morgan Stanley) to $340 (Citi). At the post-earnings price of $200, the median target implies ~50% upside. Wedbush ($270), Needham ($300), Barclays ($280), TD Cowen ($275), and Craig-Hallum (Buy) all launched coverage within the same week.
The bull case rests on one foundational assumption:
If AI inference becomes a larger compute market than training (multiple institutions forecast inference compute spending will exceed training by 2027), and if Cerebras’ speed advantage proves real and sustainable, then capturing just 3–5% of NVIDIA’s >80% market share would fully justify its current valuation.
Bear Case: Gross Margin, Customer Concentration, and the Fragility of a $50 Billion Valuation
Each bull thesis faces a bear counterargument.
Counterargument One: The inference speed moat may be narrower than assumed.
Cerebras’ speed edge stems from on-die SRAM capacity—but NVIDIA isn’t standing still. Its March-launched B300 chip significantly boosts HBM bandwidth; Groq’s LPU architecture is rapidly iterating for inference workloads.
Viewed differently: Cerebras’ key customers remain concentrated at OpenAI and AWS—yet OpenAI is also one of NVIDIA’s largest GPU buyers, and AWS’s in-house Trainium chips increasingly cover inference use cases. Cerebras’ major clients are simultaneously betting on alternatives—meaning its speed premium will face persistent pricing pressure.
Counterargument Two: Gross margin erosion may not be “temporary.”
Management attributes the Q2 gross margin drop—from 47% to 36–38%—to temporary rental costs arising from insufficient data center capacity. But this explanation assumes costs improve once new data centers go live in H2.
Given H2 revenue is projected to surge (management explicitly flagged backend-loaded revenue), and given new data center ramp-up itself demands time and capital investment, this margin recovery path is far from assured.
A deeper issue lies in the business model transformation itself. As Cerebras shifts from selling hardware to selling cloud-based compute, it must absorb data center construction, operations, and depreciation costs. With depreciation expenses now flowing through the P&L, uncertainty surrounds whether cloud services gross margin can sustainably hold above 50%. This business model’s profit ceiling remains untested.
Counterargument Three: Customer concentration is “renamed—but unsolved.”
In 2024, G42 alone contributed 85% of Cerebras’ revenue. In 2025, G42’s share fell to 24%, but MBZUAI (Mohammed bin Zayed University of Artificial Intelligence) surged from zero to 62%. The S-1 prospectus explicitly identifies both as “affiliates.” Combined, these two UAE affiliates still account for 86% of revenue. Revenue diversification reflects name changes—not substantive dispersion.
Finally, CBRS’s IPO lock-up includes an unconventional clause:
If the company’s market cap sustains above $40 billion, insider shares may unlock early. At the post-earnings price of $200, the current market cap sits near ~$45 billion—already approaching the trigger threshold. Short interest stood at 17.15% of float as of May 29—a relatively high level. An early lock-up release flooding the market with insider shares—combined with existing short pressure—could trigger concentrated selling.
Join TechFlow official community to stay tuned
Telegram:https://t.me/TechFlowDaily
X (Twitter):https://x.com/TechFlowPost
X (Twitter) EN:https://x.com/BlockFlow_News
@chaoxiangooo
