AI hardware startup Etched announced on Tuesday that it raised an additional $700 million at a $21 billion valuation, with Jane Street leading the round after the quant fund tested and purchased the company’s AI systems. The valuation step-up is unusually fast even by AI standards: Etched was valued at $5 billion in December, then raised a $300 million Series C at a $10.3 billion valuation in July, and now investors have doubled its valuation to $21 billion in about a month.

Etched sells its technology as full systems it calls “frontier inference clusters,” a category where Nvidia markets its own integrated offerings as AI factories. Co-founder and COO Robert Wachen told TechCrunch that investor enthusiasm stems from two custom-designed components aimed at speeding up inference, the process that runs after a user submits a prompt. He explained that inference has two stages: prefill and decode. The prefill phase is mathematically and compute-intensive, requiring the system to understand the prompt and its context. The decode phase is memory-intensive, generating the output tokens the user sees.

To address the prefill stage, Etched built a chip that operates at low voltage, letting it pack in more transistors without the heat problems typical of high-end AI chips, which allows it to process more tokens faster. For the decode phase, Etched created a new type of memory and interconnect it calls cluster-scale memory. Wachen said this lets many chips connect together and use a shared memory pool with very low latency. The company promises higher speeds and lower costs as a result.

Etched still faces the perception from its early days that it etches a specific model into its chips, meaning each chip is custom-designed to run one frontier model. That was the original intention, but Wachen says it is no longer the case, and Etched‘s systems can run any frontier model.

In the blog post announcing the round, Jane Street said: “We tested the chip and are pleased with the early results. Etched‘s unique approach to inference delivers the precision we will need to support our most demanding workloads. We’re excited to now have our own rack running in our datacenter.” Other investors in Etched include Kleiner Perkins, Sequoia Capital, Andreessen Horowitz, Peter Thiel, Tiger Global, Bain Capital Ventures, Neo, Stripes, Primary, Positive Sum, Diffusion, Argo, and Blackstone.

The round underscores how much capital is flowing into specialized inference infrastructure, particularly from firms that run compute-heavy workloads and are willing to test hardware directly. The speed of the valuation increase reflects investor belief that Etched‘s architectural approach, which separates prefill and decode processing, can deliver meaningful performance gains in AI serving. At the same time, the company must convince the broader market that its custom hardware is broadly compatible with frontier models and not locked to a single architecture.