Etched announced on Tuesday a $700 million funding round led by Jane Street, pushing its post‑money valuation to $21 billion. The startup was valued at $5 billion in December and raised a $300 million Series C at a $10.3 billion valuation in July, meaning its valuation has roughly doubled in just over a month.
Etched markets its offering as “frontier inference clusters,” comprising two custom‑designed blocks. A prefill ASIC operates at low voltage, allowing a higher transistor density without the thermal penalties typical of high‑power AI chips, which speeds the prompt‑understanding stage. For the token‑generation stage, the company introduced a new memory subsystem and interconnect dubbed cluster‑scale memory, enabling many chips to share a low‑latency memory pool. The architecture separates the mathematically intensive prefill phase from the memory‑intensive decode phase, and Etched says the system can run any frontier model rather than being locked to a single network.
Why this matters
The rapid valuation increase signals that investors view Etched’s split‑architecture approach as a viable path to lower inference cost and higher throughput, especially as large language models continue to grow in size and usage. While the company has not published benchmark numbers or context‑window specifications, the emphasis on low‑voltage prefill logic and a shared‑memory decode fabric suggests a focus on reducing energy per token and improving scalability across multi‑chip systems. If these claims hold in independent tests, Etched could shift the economics of serving frontier models, putting pressure on incumbent providers that rely on monolithic GPUs.