Etched Raises $300M at a $10.3B Valuation to Build Frontier AI Inference Clusters

AI chip company Etched has raised $300 million at a $10.3 billion valuation, led by Sequoia Capital, to build inference clusters designed specifically for the world’s most demanding AI models. Andreessen Horowitz, Jane Street, Diffusion, Argo, and memory maker SK Hynix also joined the round, announced on 23 July 2026.

The San Jose company frames its mission simply: to run the world’s inference, the work of actually serving AI models to users, at gigawatt scale.

The access problem behind the raise

Etched anchors its case in a single statistic. Fewer than 1% of the global population currently has access to the most advanced AI models.

Closing that gap is not primarily a distribution challenge. It is a physics and manufacturing one. Serving frontier models to billions of people requires far more tokens per watt and many gigawatts of additional compute, neither of which existing hardware delivers cheaply enough at scale.

What “frontier inference clusters” means

Rather than shipping a standalone chip, Etched co-designs the entire system: chips, packages, PCBs, cold plates, interconnects, racks, software, and manufacturing methods.

The target workloads are the hardest ones in production today, including many-trillion-parameter mixture-of-experts models, long-context tasks, and agentic systems, across both prefill and decode.

The company has described two core architectural approaches:

• Low voltage inference (LVI): running the chip’s math blocks at under half the voltage of typical AI chips, which the company says allows trillion-parameter sparse MoEs to run at over 80% of peak FLOPs without thermal throttling.
• Cluster Scale Memory (CSM): a shared, low-latency memory pool across the scale-up domain using a proprietary high-bandwidth interconnect, combining HBM capacity with SRAM-like decode speed.

These are company-stated performance claims, not independently benchmarked results.

Production is the product

Etched’s operating philosophy is that manufacturing capability is the differentiator, not the design alone. The company says its A0 silicon returned from TSMC’s N4P process earlier this year, and it has begun fabrication of hundreds of millions of dollars worth of inference clusters.

To shorten iteration cycles, it has built a 10-megawatt lab fifteen minutes from its office, opened a factory in Taiwan, and set up a data center, test house, and prototyping lab at its San Jose headquarters. Engineers have lived overseas for months to co-design with supply-chain partners.

Traction and backing

Etched reports over $1 billion in customer contracts, with its first racks shipping this summer. Beyond the newly announced round, the company has previously raised $800 million across four unannounced financings, including a strategic investment from VentureTech Alliance.

Its team now numbers more than 400 engineers drawn from NVIDIA, Google’s TPU program, Broadcom, SK Hynix, and TSMC.

The founders and leadership

Etched was co-founded by Gavin Uberti (CEO), Robert Wachen (President), and Chris Zhu. Uberti and Wachen are both Harvard Thiel Fellows.

The senior team carries deep silicon experience. CTO Mark Ross was previously CTO of Cypress Semiconductor, acquired for $9.4 billion. VP of ASIC and Architecture Saptadeep Pal co-founded Auradine and worked on NVIDIA’s H100, A100, and V100 architecture teams. VP of Software David Munday built the TPU software team across TPU v1 to v5 at Google.

Why inference is the new battleground

The AI infrastructure conversation has long centered on training, but the economics are shifting. Inference is where compute demand compounds as models reach real users, and where cost per token determines whether AI products are viable at scale.

Etched is one of several companies betting that this layer needs purpose-built hardware rather than general-purpose GPUs. With $300 million more in the bank, a $10.3 billion valuation, and racks shipping this summer, its thesis is about to meet the market.

Source: Etched announcements