AI chip startup Etched announced a $700 million funding round on Tuesday that values the company at $21 billion, more than doubling its valuation in less than a month as investors race to back specialized hardware for running AI models.
The round was led by Jane Street, the quantitative trading firm that became Etched’s first customer after testing the startup’s silicon in its own datacenter. Kleiner Perkins, Sequoia Capital, Andreessen Horowitz, Tiger Global, and Peter Thiel also participated, according to the company.
A Staggering Valuation Climb
The speed of Etched’s rise is remarkable even by the standards of the current AI boom. The San Jose-based company was valued at $5 billion in December, raised a $300 million Series C at $10.3 billion in July, and has now nearly doubled again to $21 billion, an increase of almost $11 billion in roughly four weeks.
In its announcement, Jane Street said it had installed Etched’s first shipped rack of hardware and was deploying the technology in its own workloads. The firm called Etched’s approach to inference one that delivers the precision needed to support its most demanding workloads.
Specialized Chips for Inference
Etched builds what it calls frontier inference clusters, complete systems designed specifically for the process of running trained AI models to generate responses. The startup has designed two custom components: a prefill chip that operates at low voltage to pack in more transistors without overheating, and a novel memory and interconnect system called cluster-scale memory that allows many chips to share a single memory pool at very low latency.
Co-founder and COO Robert Wachen told TechCrunch that inference happens in two stages: a compute-intensive prefill phase where the system processes the prompt, and a memory-intensive decode phase where it generates output tokens. Etched has built hardware optimized for both.
The company has more than 400 employees and a working chip, and has moved past its original plan to etch specific models into its silicon. Etched’s systems can now run any frontier model, addressing early skepticism that its hardware was too narrow to gain traction.
Backing the AI Infrastructure Stack
The funding underscores growing investor appetite for the infrastructure layer of the AI ecosystem. While much attention has focused on Nvidia’s dominance in AI training chips, Etched is betting that the inference market, where models are actually deployed for end users, will become equally large as AI applications scale.
Etched competes with Nvidia’s full-system offerings, which the GPU giant calls AI factories, as well as with other inference-focused startups. The company’s backers include Bain Capital Ventures, Stripes, Primary, Positive Sum, Diffusion, Argo, and Blackstone.
Sources: Reuters; TechCrunch; Wall Street Journal
discussion