Etched, a startup that makes chips specifically designed to run artificial-intelligence models, said Tuesday it has closed a $700 million funding round that values the company at $21 billion, double what investors paid a month ago. The company also said it has completed its first product deliveries to Jane Street, the trading firm that was among the first to bet on its technology.
The numbers are striking even by the standards of the AI chip boom. A valuation that doubles in a month suggests investors believe Etched has crossed a threshold: from a promising design to a product people are actually paying for. The Jane Street deliveries are the evidence. The trading firm, known for its willingness to test exotic technology, has been evaluating Etched’s chip for AI-driven trading workloads, and the delivery marks the first time the startup’s hardware has shipped to a customer at scale.
Etched’s bet is a specific one. The company is building chips hard-wired for transformer models, the architecture behind most of today’s large language models. By baking the model’s computations directly into the silicon, Etched says its chip can run inference far more efficiently than general-purpose graphics processing units from Nvidia. The trade-off is that the chip does one thing, and one thing only: if the industry’s model architectures shift, the chip’s advantage could evaporate.
That bet puts Etched in direct competition with Nvidia, the most valuable chip company in the world. Nvidia’s GPUs dominate AI computing because they can run any model, and the company’s software ecosystem makes switching costly. Etched’s argument is that specialization beats flexibility at the scale of modern AI, where the same transformer architecture powers thousands of applications and the cost of inference is measured in millions of dollars a day.
The funding round brings together the two things every chip startup needs: capital and customers. Jane Street is the customer, a name that carries weight in quantitative trading and a reputation for evaluating technology on merit. The investors are the capital, and their willingness to pay $21 billion for a company that has only begun shipping product suggests the market sees a clear path for specialized inference chips. The round also gives Etched the balance sheet to negotiate long-term supply agreements with manufacturers, the other scarce resource in chip making.
The valuation math is aggressive. At $21 billion, Etched is worth more than many established semiconductor companies, despite having no meaningful revenue history in public filings and a product that is only now reaching customers. The doubling in a month reflects a market that is pricing scarcity: there are only a handful of companies with working silicon designed specifically for AI inference, and investors have concluded that the winners in this category will be worth far more than today’s prices.
The comparison investors make is with the early days of Nvidia’s own rise, when the company’s GPUs were dismissed as gaming hardware before the AI boom made them indispensable. Etched’s founders have leaned into that analogy, arguing that the specialized chip they have built is where the industry is heading. Skeptics note that Nvidia is already adding inference-specific features to its own products, and that the incumbent has the resources to absorb the threat or acquire it.
The broader trend favors Etched’s thesis. The cost of running AI models has become one of the largest line items in the technology industry, and every large company using AI is looking for ways to cut it. Specialized chips promise the biggest cuts, and the demand for them has attracted a wave of startups, from companies building memory-adjacent hardware to teams designing chips for specific workloads. Etched is among the most visible, and now among the best funded.
The real test is manufacturing. A chip design is a promise; a chip that ships in volume is a business. Etched has contracted with manufacturing partners and says its production schedule is on track, but the history of chip startups is littered with designs that never made it to scale. The company’s next tests will be measured in deliveries, not dollars.
For now, the market has spoken. Jane Street has the product, investors have the valuation, and Etched has the attention of an industry that is watching to see whether specialization can beat the generalist. The chip has been designed, the money has been raised, and the question of whether Etched’s bet pays off will be answered in the data centers where its silicon actually runs. In a market that doubles valuations in a month, the next number that matters will be the one in the delivery logs, not the term sheet.
