AI chip startup Cerebras Systems is targeting a valuation of up to $26.6 billion in what could become one of the largest tech IPOs of the year, testing investor appetite for semiconductor plays beyond Nvidia's dominant position.
The Sunnyvale, California-based company is seeking to price shares between $115 and $125, according to regulatory filings, in a deal that would raise roughly $2.5 billion and value the eight-year-old startup at more than established chipmakers like Advanced Micro Devices commanded just five years ago.
Cerebras has carved out a niche in the AI infrastructure market with its wafer-scale engine technology—massive chips the size of dinner plates that dwarf traditional GPUs. The company's CS-3 system features 4 trillion transistors on a single wafer, compared to tens of billions in conventional chips, enabling faster training of large language models.
The pitch to investors: while Nvidia dominates AI training and inference with its GPU architecture, Cerebras offers a fundamentally different approach better suited for the largest AI workloads. The company claims its systems can train models 10 to 100 times faster than GPU-based alternatives, though at a premium price point.
Revenue figures in the filing show explosive growth: $310 million in 2025, up from $78 million the prior year. But the company remains unprofitable, burning through cash as it scales manufacturing and competes for talent in the overheated AI labor market.
"The valuation is aggressive but not insane given the AI infrastructure gold rush," said Michael Rodriguez, managing partner at Sequoia Capital, an early Cerebras investor. "The question is whether they can maintain differentiation as Nvidia iterates and hyperscalers build custom silicon."
That competitive threat is real. Google, Amazon, and Microsoft are all developing proprietary AI chips, potentially limiting Cerebras' addressable market. Meanwhile, Nvidia's upcoming Blackwell architecture promises to narrow the performance gap.
The IPO timing is strategic. AI infrastructure spending shows no signs of slowing, with enterprises racing to build out GPU clusters and specialized AI hardware. Morgan Stanley estimates the AI chip market will reach $400 billion annually by 2027, up from roughly $50 billion today.
Cerebras counts Citigroup, GlaxoSmithKline, and several national laboratories among its customers—a diverse base that reduces dependence on any single sector. The company has also secured partnerships with major cloud providers, though details on revenue contribution remain limited.
Investors will be scrutinizing gross margins closely. Wafer-scale manufacturing is complex and expensive, with lower yields than traditional chipmaking. If Cerebras can't demonstrate a path to profitability at scale, the valuation multiple will face pressure.
The offering comes as tech IPOs have roared back after a brutal 2023-2024 drought. Reddit and Instacart both saw strong aftermarket performance, emboldening other venture-backed companies to test public markets.
For Cerebras, the IPO represents validation of an audacious bet: that the AI boom requires not just more GPUs, but fundamentally new chip architectures. Whether Wall Street agrees with that $26.6 billion price tag will become clear in the coming weeks.
