Etched's AI hardware hits $21B valuation in a month

Etched, an AI hardware startup, has seen its valuation skyrocket from $5B in December to $21B after raising $700M in its latest funding round, led by Jane Street. This rapid valuation increase is driven by Etched's innovative approach to AI inference, a critical process in AI model deployment. The company designs custom chips that optimize the two stages of inference: the compute-intensive prefill phase and the memory-intensive decode phase.

Etched's breakthrough lies in its custom-designed chips for each phase. The prefill chip operates at low voltage, enabling higher transistor density without overheating, while the decode phase benefits from a new type of memory and interconnect called cluster-scale memory. This allows multiple chips to share a memory pool at ultra-low latency, promising faster speeds and lower costs. Etched has also dispelled early perceptions that its chips were model-specific; its systems can now run any frontier model, broadening its market appeal.