Etched raises $300M Series C at $10.3B valuation, values inference chips above general training

Etched, an AI chip startup founded by three Harvard dropouts in 2022, closed a $300 million Series C at a $10.3 billion valuation—the highest ever for a Sequoia-led Series C. Lead backer is Sequoia Capital, with participation from Andreessen Horowitz, Jane Street, Diffusion, and memory chip giant SK Hynix. The company nearly doubled its $5 billion valuation from December 2025 in just seven months.

Etched builds inference-only chip clusters optimized to run large language models, not train them. Unlike NVIDIA's general-purpose GPUs that handle both training and serving, Etched specializes in transformer-based inference at minimal power and latency. The startup emerged from stealth in June with a working chip already in production and $1 billion in signed customer orders. It now employs 400 engineers from NVIDIA, Broadcom, Google's TPU group, and SK Hynix itself.

The core innovation addresses two inference bottlenecks: power and memory. Low Voltage Inference runs computation blocks at half typical voltage, cutting heat and enabling higher transistor density. Cluster Scale Memory creates a shared SRAM pool across entire racks via proprietary interconnect. SK Hynix's participation validates Etched's memory architecture as inference becomes the dominant workload and centralized training economics diverge from distributed serving.

For infrastructure architects, Etched represents a direct alternative to NVIDIA's inference pricing and power envelope. The $1 billion order book and near-immediate production deployment matter more than valuation—customers are moving to production, not evaluation. This validates the thesis that inference at scale requires a rebuilt compute stack, not just incremental GPU improvements, making specialized inference hardware a distinct market category.