Etched raises 700M USD at 21B USD valuation for AI inference chips

Etched says it has raised 700M USD at a 21B USD valuation, giving the AI chip startup another major step-up as investors look for alternatives to general-purpose accelerators for inference workloads. Jane Street led the round after testing Etched hardware and installing the company’s first rack in its own datacenter.

The company says its systems are built for frontier model inference, the stage where models process a prompt and generate an answer. Etched says it designed separate hardware for the compute-heavy prefill phase and the memory-heavy decode phase, including a shared memory approach meant to connect many chips with low latency.

TechCrunch reported that Etched was valued at 5B USD in a previous round and then 10.3B USD before the latest financing. The speed of the jump matters because inference is becoming one of the central cost bottlenecks in AI: every ChatGPT-style query, coding-agent run, or enterprise assistant session depends on running models cheaply and quickly after training is done.

Etched is still early compared with Nvidia’s ecosystem, but the Jane Street deployment gives the round more weight than a purely speculative financing. If specialized inference clusters work at scale, AI infrastructure may become less about one universal chip and more about matching hardware to the economics of serving models.