Etched Raises $300M Series C at $10.3B Valuation, Led by Sequoia Capital
The AMW Read
Updates the AI chip player map with a $300M round at $10.3B valuation, signaling capital acceleration in specialized inference hardware; cross.§D triggered by round size ($300M under $500M cap but $10.3B valuation is well above $5B threshold), cross.§H as chip design is primary subject
Etched Raises $300M Series C at $10.3B Valuation, Led by Sequoia Capital
Etched, an AI chip startup founded by Harvard dropouts Gavin Uberti, Chris Zhu, and Robert Wachen, announced a $300 million Series C round led by Sequoia Capital, with participation from a16z, SK Hynix, Jane Street, and Diffusion. The round values the company at $10.3 billion, more than doubling its $5 billion valuation from a $500 million Series B in December 2025. Etched develops specialized inference chips and systems optimized for large language model workloads, separating prefill and decode into distinct hardware-optimized pipelines: low-voltage inference for prefill and cluster-scale memory with shared SRAM pools for decode. The company has already fabricated first silicon at TSMC N4P process and announced over $1 billion in customer contracts for its rack-scale inference systems, supporting models including DeepSeek, Qwen, Mamba, and Llama.
This funding round is significant because it demonstrates escalating capital concentration in the AI silicon segment, where specialized inference hardware is becoming a high-stakes battleground. Etched’s approach directly targets the compute economics bottleneck of LLM inference, aiming to displace general-purpose accelerators (Nvidia GPUs) with application-specific chips that optimize for the two-phase workload of transformer models. The rapid valuation growth from $5B to $10.3B in seven months signals strong investor conviction that dedicated inference silicon can capture substantial enterprise inference spend, particularly as model usage scales and inference costs become the dominant total-cost-of-ownership factor. The participation of SK Hynix, a major memory manufacturer, also highlights the growing interdependence between chip design and memory-interface innovation for AI workloads.
From an AI market analysis perspective, Etched is now positioned as one of the most valuable independent AI chip companies, challenging the narrative that only hyperscaler-backed efforts (Google TPU, AWS Trainium, Microsoft Maia) can succeed in specialized silicon. The $1B+ in customer contracts provides rare pre-revenue validation, reducing the risk profile of the custom-chip bet. However, the company faces formidable execution challenges: TSMC’s N4P process delivers competitive performance but is not bleeding-edge (N2), and the inference chip market remains volatile as model architectures evolve — a risk that sunk earlier startups like Cerebras in the training segment. The round also reinforces the capital-compression arc in AI infrastructure, where only firms able to command $300M+ rounds at billion-dollar valuations survive the compute capex war.
