Etched Raises $700M at $21B as Inference Race Accelerates

Etched has raised $700M in new funding at a $21B valuation, with Jane Street leading the round. Kleiner Perkins, Sequoia, Andreessen Horowitz, Tiger Global, Bain Capital Ventures, Neo, Primary, Stripes, Positive Sum, and Blackstone also participated, while the financing stage was not disclosed.

The headline is enormous, but the customer relationship is more revealing. Etched shipped its first rack to Jane Street in July 2026, and the trading firm says it is deploying that system in its own workloads after testing the chip.

That pairing turns the announcement into more than another expensive vote of confidence in AI infrastructure. The investor writing the largest check is also the first disclosed customer testing whether Etched can convert custom silicon into production systems that deliver useful inference economics.

What Happened

The August 18 financing values Etched at $21B, more than double the $10.3B valuation attached to its $300M Series C less than one month earlier. Etched says it has now raised $1.9B in total, although multiple earlier financings were unannounced and the company did not give the latest round a Series label.

Jane Street led the round after becoming Etched's first disclosed customer. The quantitative trading firm said it was pleased with early chip results and is running an Etched rack in its data center, giving the financing a direct link to a real deployment rather than a purely financial thesis.

The syndicate reaches across venture capital, growth capital, and technically sophisticated market participants. Kleiner Perkins, Sequoia, Andreessen Horowitz, Tiger Global, Bain Capital Ventures, Neo, Primary, Stripes, Positive Sum, and Blackstone are all named in the announcement, with Etched's existing backers and new participants converging around the same bet on specialized inference infrastructure.

From First Silicon to First Customer

Etched was founded in 2022 by Gavin Uberti, Robert Wachen, and Chris Zhu after they left Harvard. Gavin Uberti is the company's CEO, Robert Wachen is President, Chris Zhu remains a co-founder, and Etched's current leadership page identifies Mark Ross as CTO.

The company reported first-pass A0 silicon on TSMC's N4P process in under three years from seed funding. In June 2026, Etched emerged from stealth with a working chip, more than 400 employees, and over $1B in signed customer contracts, then followed with a $300M Series C led by Sequoia on July 23.

The first rack delivery is the bridge between those milestones and the current valuation. A successful tapeout proves a design can become silicon, while a customer deployment begins the longer test of reliability, software maturity, system integration, production yield, and economics under real workloads.

Why Etched Is Building the Whole System

Etched describes its product as a frontier inference cluster, not a standalone accelerator. The company co-designs chips, memory, packages, circuit boards, interconnects, cooling, racks, software, and manufacturing methods because inference performance is constrained by the behavior of the full system.

Its Low Voltage Inference technology targets higher compute density within a fixed power envelope. Cluster Scale Memory uses a hybrid SRAM and HBM subsystem to create a shared, low-latency memory pool across the cluster, aiming to serve both compute-heavy prefill and memory-heavy decode workloads.

Etched says its systems can run large mixture-of-experts models and non-transformer architectures such as Mamba. Those capabilities and performance claims still need broader independent validation, but they matter strategically because inference workloads are changing faster than a hardware company can redesign around one narrow model architecture.

Why This Funding Matters

AI training created the first major wave of accelerated-computing demand, but inference is where trained models meet users, agents, applications, and recurring operating costs. As usage expands, buyers care about throughput, latency, power consumption, memory access, deployment reliability, and tokens per dollar, not just theoretical peak performance.

That makes inference a credible opening for new systems companies, even in a market dominated by Nvidia's hardware and software ecosystem. The opening is not a guarantee, because semiconductor history is full of technically impressive products that struggled with manufacturing, software support, customer adoption, or the capital required to scale.

Etched now has the money, facilities, and disclosed customer demand to face that test. The company operates engineering and prototyping infrastructure in San Jose, opened an 80,000-square-foot, 10 MW facility in Milpitas, established a factory presence in Taiwan, and says it is developing three hardware generations in parallel.

What This Signals for AI Infrastructure

The $21B valuation reflects investor conviction that inference will support a major new infrastructure layer. It also sets a demanding standard: Etched must turn early racks and contracted demand into repeatable production, durable customer workloads, and economics strong enough to justify switching from established platforms.

Jane Street's dual role as lead investor and first disclosed customer makes this round unusually concrete. It does not settle the competitive question, but it shows that a technically demanding buyer tested the product, accepted a deployment, and increased its financial exposure to the company.

The next phase will be measured less by financing velocity than by manufacturing execution. Etched has already proven it can attract capital at remarkable speed; the harder proof is shipping systems at scale while the models, memory demands, power constraints, and competitive landscape keep moving.