Etched AI chip exits stealth with $800M raised
Etched emerged from stealth with $800M raised, a $5B valuation, and $1B in contracts for an inference-only AI chip built at TSMC, shipping summer 2026.
Etched, a four-year-old AI chip startup, came out of stealth on June 30 with $800 million raised across multiple undisclosed rounds. The most recent was $500 million at a $5 billion post-money valuation, and the company has $1 billion in signed customer contracts. TSMC manufactured the chip in early 2026; customer shipments are planned for summer 2026.
Etched is building inference chips, not training chips. Its pitch is that specializing silicon for transformer inference rather than general GPU compute yields substantially better performance per watt for running AI models after training. The company packages its chips into “frontier inference clusters” that bundle custom racks and proprietary software.
Investors include Jane Street, Hudson River Trading, Jump Trading, Two Sigma, Ribbit Capital, and Radical Ventures. Angel backers include Andrej Karpathy, Geoffrey Hinton, Fei-Fei Li, Stanley Druckenmiller, Peter Thiel, Arthur Mensch, and Scott Wu.
The founders are Gavin Uberti (CEO) and Robert Wachen, Harvard dropouts and Thiel Fellows who founded the company in 2022.
Why it matters
Inference (running models after training, not training them) is where most AI spending lands in production. Etched is betting that the inference workload is regular enough to warrant specialized silicon, similar to how TPUs serve training at Google.
The $1 billion in signed contracts before shipping is the most credible signal here; it suggests enterprise customers are willing to commit now rather than wait to see whether Nvidia maintains dominance. The chip story connects to Qualcomm’s Dragonfly data-center roadmap and the broader AI chip race analysis. For operators, the near-term practical read is: inference costs are likely to fall further and faster than GPU-only pricing would suggest.