Cerebras drops 19% below IPO price as AI chip margins tighten

Cerebras shares plunged 19% in their second day of trading after the AI chipmaker's first quarterly report revealed tighter profit margins, dragging the stock below its May IPO price.

Cerebras shares fell 19% to trade below their initial public offering price after the AI chipmaker's debut earnings report showed profit margins would be tighter than expected, deepening a selloff that has erased roughly $1.5 billion in market value since the company went public in May.

"Cerebras is competing in the most capital-intensive segment of the AI supply chain, and the market is now pricing in the cost of that competition," said Stacy Rasgon, senior semiconductor analyst at Bernstein.

The stock closed at $38.42, below its $42 IPO price, after the company reported gross margins that missed analyst estimates. Cerebras, which makes wafer-scale AI processors that compete with Nvidia's H100 and B200 graphics processing units, said it expects margins to remain under pressure as it invests in production capacity and customer acquisition. The initial earnings reaction sent shares down 8% before the decline accelerated to 19% in the following session.

The selloff adds to a broader June rout in AI-linked stocks. The "Magnificent Seven" plus Broadcom and Oracle have lost roughly $2.7 trillion in market value this month as investors reassess the cost of the AI build-out. Cerebras, which went public at a valuation of about $8 billion, now trades at a market cap of roughly $6.5 billion.

The company's wafer-scale engine (WSE-3) — a single silicon wafer-sized chip that eliminates the need for multiple interconnected GPUs — has attracted customers in government and enterprise AI training. But the technology requires custom manufacturing processes at Taiwan Semiconductor Manufacturing Co., adding cost that competitors using standard GPU architectures can avoid. Cerebras's revenue grew more than 200% year-over-year in its first quarterly report, but spending on sales, marketing, and R&D outpaced that growth, accelerating the company's cash burn rate as it funds development for its next-generation WSE-4 processor, expected in 2027.

The broader AI chip market remains dominated by Nvidia, which controls an estimated 80% of the data center AI accelerator market. Nvidia's H100 GPU, built on TSMC's 4nm process node (which packs more transistors per square millimeter, improving performance per watt), delivers 990 TFLOPS of FP16 compute — roughly 3x the raw throughput of Cerebras's WSE-3 on paper, though Cerebras argues its architecture offers superior efficiency for specific large-language model workloads.

For investors, Cerebras's post-IPO slide offers a cautionary tale about the gap between AI enthusiasm and the economics of chip manufacturing. Nvidia trades at 35x forward earnings, while Cerebras — still unprofitable — commands a price-to-sales multiple that analysts say leaves little room for execution missteps. SK Hynix, the memory chipmaker seeking a $29 billion US listing on the Nasdaq, faces similar scrutiny as investors demand clearer paths to profitability from AI supply chain companies.

This article is for informational purposes only and does not constitute investment advice.