Etched Hits $10.3B Valuation as AI Inference Chips Draw Fresh Investor Demand

Lead

Etched is becoming one of the most closely watched challengers in AI infrastructure. According to TechCrunch, the chip startup has closed a $300 million Series C round at a $10.3 billion valuation, led by Sequoia and joined by Andreessen Horowitz, SK Hynix, Jane Street, Diffusion Capital and earlier backers. The company was valued at about $5 billion in December, meaning its valuation has roughly doubled in seven months.

Founded in 2022 by three Harvard dropouts, Etched has built its pitch around a direct claim: AI inference can be accelerated without relying on GPUs. The company sells full systems rather than standalone chips, and it says its first silicon has been manufactured, its initial systems are being tested by customers, and it has booked $1 billion in orders.

Key points

• A major funding signal: Etched says this is the highest valuation ever for a Sequoia-led Series C. The round also includes strategic and financial investors whose names suggest interest across chips, AI software and capital markets.
• Not just for one LLM: One of the persistent criticisms of Etched has been that a transformer-focused chip could become too narrow. COO Robert Wachen told TechCrunch that the systems can run many model types, including Mixture of Experts models such as DeepSeek and Qwen, as well as non-transformer approaches like Mamba.
• Inference is the target: Etched focuses on the two stages of inference. The prefill stage processes the prompt and context and is compute-intensive. The decode stage generates output tokens and is constrained by memory capacity, bandwidth and latency.
• Two custom components: For prefill, Etched has designed a low-voltage inference chip that it says reduces heat and allows denser transistor packing. For decode, it has built what it calls cluster-scale memory, combining memory and interconnect technology so multiple chips can access a shared memory pool quickly.
• Scaling remains the test: The company now has around 400 employees, operates a 2MW data center at its office, and has opened a new 80,000-square-foot, 10MW facility in Milpitas. But mass production and broad delivery of rack-scale systems are still ahead.

Why it matters

Etched’s rise reflects a broader shift in AI infrastructure economics. Training frontier models remains expensive, but inference is where usage costs accumulate at massive scale. Every user prompt creates demand for fast computation, low-latency memory access and efficient energy use. If a specialized system can lower those costs while supporting many architectures, it could become a meaningful alternative to GPU-centric deployments.

The company is not without risk. Hardware startups must clear a long chain of hurdles after a successful tape-out: manufacturing yield, system reliability, software tooling, deployment support and customer trust. Private demos and early orders can create momentum, but the real proof will come when Etched ships systems at scale and customers measure them against existing infrastructure. For now, its $10.3 billion valuation shows that investors believe the inference bottleneck is large enough to justify a bold hardware bet.

Source: TechCrunch AI