Cerebras Stock Drops 11 Percent Despite 92 Percent Revenue Growth

Cerebras Systems (CBRS) posted $249.7 million in first-quarter revenue on Monday, a 92% jump from the same period a year ago. The stock fell 11% in after-hours trading anyway. The culprit was a two-to-three percentage point decline in projected gross margins for next quarter, from 73% to a range of 70% to 72%. Wall Street, it turns out, does not reward fast-growing AI chipmakers for slowing profitability, even when the top line nearly doubles.

The earnings call on June 23 marked the first time Cerebras reported results as a public company. It went public on March 4 at $36 per share and closed Monday's regular session at $83, a 128% gain in less than four months. The after-hours sell-off trimmed that run but did not erase it. The real question is whether the market's reaction reveals something deeper about the AI hardware boom and the capital structures propping it up.

The Numbers Behind the Headlines

Cerebras reported a GAAP net loss of $94.1 million for Q1 2025, or $0.38 per diluted share. That compares to net income of $17.5 million in the year-ago quarter. The swing from profit to loss, even as revenue surged, reflects the cost of scaling a hardware company that competes directly with Nvidia.

Non-GAAP net income, which strips out stock-based compensation and other adjustments, was $28.7 million. Adjusted EBITDA came in at $38.3 million. Gross profit reached $192.7 million, up from just $30 million a year earlier, putting gross margin at roughly 77%. Operating expenses totaled $155 million.

The company also disclosed $1.3 billion in remaining performance obligations, contracted revenue that has not yet been recognized. CEO Andrew Feldman called the quarter evidence of "Cerebras' successful transition from a startup to a profitable growth company." The GAAP loss complicates that narrative somewhat, but the backlog suggests demand is real.

For Q2 of fiscal year 2026, Cerebras guided revenue between $275 million and $300 million, implying year-over-year growth of 120% to 140%. The projected non-GAAP gross margin of 70% to 72%, however, spooked investors who had priced in margin expansion, not contraction.

The G42 Concentration Problem

Buried in the filing was a number that deserves more attention than the margin guidance: $168.5 million. That is how much revenue came from G42, the Abu Dhabi-based artificial intelligence company. It represents 67.5% of Cerebras' total quarterly revenue from a single customer.

Customer concentration at that level is unusual for a public company and introduces a specific kind of fragility. If G42 delays an order, renegotiates terms, or pivots to a competitor, Cerebras loses two-thirds of its revenue overnight. The company's U.S. revenue was $106.1 million, up 250% from $30.3 million a year ago. International revenue, dominated by G42, was $143.6 million.

This dependency also raises geopolitical questions. The United States has tightened export controls on advanced AI chips multiple times since 2022. Abu Dhabi sits in a gray zone. G42 restructured its China-facing business in 2024 under pressure from the U.S. government, and Microsoft invested $1.5 billion in the company. But regulatory winds shift. A single executive order could disrupt the relationship that provides the majority of Cerebras' revenue.

Feldman and his team are clearly aware of this risk. The OpenAI inference infrastructure deal, announced earlier this year, was a major catalyst for the stock's rally. Diversifying the customer base beyond G42 is not optional. It is existential.

Nvidia's Shadow and the Wafer-Scale Bet

Cerebras' core product is the Wafer-Scale Engine (WSE), a chip that occupies an entire silicon wafer rather than the small die sizes used by conventional processors. The WSE packs far more transistors and memory bandwidth per unit than Nvidia's GPU-based systems. In theory, this architecture offers advantages for training large language models and, increasingly, for inference workloads.

In practice, Nvidia controls an estimated 80% to 90% of the AI chip market. McKinsey projects that market will reach $300 billion by 2030. Even at Cerebras' current growth rate, the company would capture only a small fraction of that total.

Nvidia's dominance rests on more than hardware. Its CUDA software ecosystem, built over nearly two decades, creates switching costs that make it painful for developers and enterprises to move to alternative platforms. Cerebras, AMD, Intel, and a growing list of startups all face the same problem: the chips may be competitive, but the software moat around Nvidia is deep.

Cerebras raised $720 million in private funding before its IPO. The company's market capitalization, even after Monday's decline, reflects enormous optimism about its ability to chip away at Nvidia's lead. Whether that optimism is justified depends less on transistor counts and more on whether enterprises will tolerate the friction of adopting a new hardware stack.

Capital Misallocation and the AI Investment Cycle

The Cerebras story is a microcosm of a broader pattern in AI-related capital markets. Revenue is growing fast, losses are mounting faster, and valuations assume a future that has not arrived. This is not unique to Cerebras. It describes much of the AI hardware and infrastructure sector in 2025.

The total capital flowing into AI infrastructure, from chip fabrication to data center construction to power generation, now runs into hundreds of billions of dollars annually. Governments are subsidizing much of it. The U.S. CHIPS Act allocated $52.7 billion. The European Chips Act committed 43 billion euros. Japan, South Korea, and Taiwan have their own programs. Central banks, by holding interest rates below what a free market would set for much of the past 15 years, enabled the cheap capital that funds these bets.

From an Austrian economics perspective, this looks like a textbook case of malinvestment. Artificially low interest rates distort the capital structure, pulling resources into long-term, capital-intensive projects that may not generate returns sufficient to justify the investment. When rates normalize or demand disappoints, the correction can be abrupt. The 11% after-hours drop on a margin guidance miss, not a revenue miss, hints at how sensitive these valuations are to small changes in expectations.

Bitcoin stands apart from this cycle. It requires no earnings calls, no margin guidance, and no customer concentration disclosures. Its supply schedule is fixed at 21 million coins, enforced by code rather than corporate strategy. The energy spent mining Bitcoin is not a speculative bet on future demand from a single Abu Dhabi client. It is the cost of securing a permissionless monetary network that operates independent of any government's industrial policy. In a world where capital allocation is increasingly shaped by state subsidies and central bank rate manipulation, Bitcoin's indifference to those forces is not a bug. It is the core feature.

The Inference Pivot and Its Limits

Cerebras is positioning itself for the shift from AI training to AI inference. Training a large language model is a one-time, massively parallel computation. Inference, running that trained model to generate responses, is an ongoing workload that scales with user demand. As AI applications proliferate, inference compute requirements could dwarf training requirements.

The OpenAI deal is Cerebras' clearest signal that this pivot is underway. OpenAI processes billions of inference requests daily through ChatGPT and its API. If Cerebras can prove that its WSE architecture handles inference workloads more efficiently than Nvidia's H100 or B200 GPUs, the addressable market expands dramatically.

But efficiency claims need real-world validation at scale. Nvidia is not standing still. Its Blackwell architecture, shipping in volume, targets inference specifically. Amazon, Google, and Microsoft are all building custom inference chips for their own cloud platforms, further fragmenting the market. Cerebras' $1.3 billion backlog is encouraging, but backlog is not revenue, and not all contracts survive contact with shifting priorities.

What to Watch

Three developments will determine whether Cerebras' post-IPO trajectory holds or reverses.

First, the G42 revenue share. If it remains above 60% through the next two quarters, the market will increasingly price in concentration risk regardless of top-line growth. Cerebras needs at least two or three more customers contributing $50 million or more per quarter to credibly claim diversification.

Second, the OpenAI relationship. Details on volume, pricing, and performance benchmarks from the inference deal will matter more than the contract announcement itself. If OpenAI publicly validates WSE performance against Nvidia alternatives, Cerebras gains credibility that no earnings report can provide. If the deal quietly stalls or scales slowly, the narrative weakens.

Third, the margin trajectory. The Q2 guidance of 70% to 72% non-GAAP gross margin signals that scaling production may compress profitability before it improves. If Q3 margins do not recover toward 75% or higher, investors will question whether Cerebras can achieve the operating leverage that justifies its valuation. In a market that punishes margin misses with double-digit sell-offs, the company has little room for error.

---

Source: CoinDesk