Etched AI Raises $300M Series C at $10.3 Billion Valuation to Build Inference-Optimized AI Chips

Etched AI, an inference chip startup founded by three Harvard dropouts in 2022, has closed a $300 million Series C at a $10.3 billion valuation. The round was led by Sequoia Capital with participation from Andreessen Horowitz, SK Hynix, Jane Street, and Diffusion Capital, according to TechCrunch.

The valuation represents a 2x increase in seven months. Etched was valued at $5 billion in December after a $500 million round. The company says this is the highest valuation ever achieved in a Sequoia-led Series C.

What Etched Builds

Etched designs and sells full inference systems: custom chips, racks, and software bundled together. The company claims its architecture runs AI models faster and more cheaply than Nvidia GPUs by tackling the two stages of inference separately.

The first stage, prefill (processing the user’s prompt), uses a low-voltage chip design. Lower voltage generates less heat, which allows more transistors per die. The second stage, decode (generating output tokens), relies on a shared, cluster-scale memory system that addresses the memory bottleneck in token generation.

Co-founder and COO Robert Wachen told TechCrunch the systems support transformers, mixture-of-experts models like DeepSeek and Qwen, and non-transformer architectures including Mamba.

Scale and Traction

The numbers behind the valuation jump: Etched reported $1 billion in booked orders in June, has approximately 400 employees, operates a 2-megawatt data center, and is building a new 10-megawatt facility in Milpitas. Total funding to date exceeds $1.1 billion.

The investor list extends beyond institutional VCs. Angel backers include Andrej Karpathy, Geoffrey Hinton, Fei-Fei Li, Peter Thiel, Dylan Field, and Amjad Masad, according to TechCrunch.

The Inference Bottleneck for Agent Deployment

Inference is the computational process that runs every time a user submits a prompt or an autonomous agent makes a decision. As agent deployments scale from testing to production, the compute requirements shift from training (batch processing, scheduled) to inference (real-time, continuous). Every agent decision cycle, every tool call, every multi-step reasoning chain generates inference load.

Etched competes in this space against Nvidia (GPU dominance), Groq (LPU architecture), Cerebras (wafer-scale engines), and Google’s TPU program. The company’s bet is that inference-specialized silicon, purpose-built for the decode stage, will outperform general-purpose GPUs on cost and latency as agent-driven inference demand scales.

The risk is execution. Designing chips is one challenge. Manufacturing and deploying complete systems at scale, competing against Nvidia’s entrenched ecosystem of software tools and developer adoption, is a larger one. TSMC successfully manufactured Etched’s chips earlier this year, and first systems are currently in customer testing.

Etched is part of a broader week in AI infrastructure funding: Edith Yeung’s Silicon Valley Funding Weekly tracked 17 startups raising $1.13 billion across the week ending July 27.