Three Harvard Dropouts' AI Chip Startup Etched Raises $300M at $10.3B Valuation
Three young founders who dropped out of Harvard in 2022 just landed the largest Series C in Sequoia Capital's history — $300 million, valuing their AI inference chip startup Etched at $10.3 billion, doubling in seven months.
Three Key Takeaways
Sequoia leads, top-tier capital piles in. The round was led by Sequoia Capital, with Andreessen Horowitz, Jane Street, SK Hynix, and Diffusion participating. Existing investors Peter Thiel and Jump Trading doubled down. This is the largest Series C investment Sequoia has ever made. Etched's previous round closed in late 2025 at roughly a $5 billion valuation; seven months later, that number has doubled to $10.3 billion.
Not a PowerPoint chip — a product with real orders. Etched's core product, Sohu, is an inference chip purpose-built for the Transformer architecture. Rather than a general-purpose GPU, it burns the Transformer compute pipeline directly into silicon. The company disclosed an order book exceeding $1 billion and has clients actively running live tests. That puts Etched in a different category from most chip startups: it is a vendor with a shipping product, not just a roadmap.
Betting on a structural shift to inference. As AI moves from training to inference — the phase where deployed models process user requests — inference compute demand is exploding. General-purpose GPUs waste significant silicon on inference workloads. Etched's bet is that a purpose-built chip can crush general-purpose solutions on inference price-performance. SK Hynix's participation is also notable: the world's second-largest memory chipmaker joining signals supply-chain-level endorsement of Etched's chip-memory co-design approach.
WangDou's Take
A chip startup hitting a $10 billion valuation would normally scream bubble. But Etched's numbers look different: a $1 billion order book, clients running live tests, and Sequoia writing the biggest Series C check in its own history. The three Harvard dropouts founded the company in 2022, when Sequoia actually rejected them. Four years later, Sequoia is leading the round — that reversal alone tells you how fast the inference chip market heated up. The core logic is simple: Nvidia's GPU is a Swiss Army knife — does everything, optimal at nothing. Etched's Sohu is a scalpel — does one thing (Transformer inference) and does it to the extreme. The problem is the Swiss Army knife maker is not going to sit and watch you carve up the cake — Nvidia's Blackwell is closing the inference gap fast. Etched's window is probably two to three years. If they cannot reach production scale by then, they will get steamrolled.
Source: TechCrunch, GlobeNewsWire, TheNextWeb