AI chip startup Etched has closed a new $700 million funding round at a valuation of $21 billion, with the investment led by quantitative trading firm Jane Street following the firm's hands-on testing and purchase of Etched's hardware. The announcement came on Tuesday and marks yet another dramatic leap in the company's financial standing.

The pace of Etched's valuation growth is striking even against the backdrop of an AI investment landscape already known for its excesses. The company was valued at $5 billion as recently as December. By July, it had raised a $300 million Series C at a valuation of $10.3 billion. Now, just one month later, investors have pushed that figure to $21 billion - a jump of nearly $11 billion in roughly thirty days.

What Etched Actually Sells

Rather than selling individual chips, Etched delivers its technology as complete, integrated systems it refers to as "frontier inference clusters." The terminology echoes similar language used by competitors in the space - Nvidia, for instance, markets its own full-system offerings under the term "AI factories." The framing reflects a broader industry shift toward selling end-to-end infrastructure rather than standalone components.

Co-founder and Chief Operating Officer Robert Wachen explained that investor enthusiasm stems from a core technical achievement: Etched has engineered two entirely new components, both designed from the ground up to accelerate inference - the computational work that occurs after a user submits a query or prompt to an AI system.

The Two-Stage Inference Problem

"Inference is built in two stages," Wachen said, describing the architecture as consisting of a prefill phase and a decode phase. Each stage has distinct computational demands, and Etched has built dedicated hardware to address both.

The prefill phase is mathematically intensive and requires significant compute resources. During this stage, the system processes and interprets the incoming prompt, including any surrounding context. The decode phase, by contrast, is memory-intensive and is where the system generates the actual output - the tokens that form the response a user ultimately reads.

To handle the prefill stage more efficiently, Etched developed a chip that runs at lower voltage than conventional high-end AI processors. This design choice allows the chip to incorporate more transistors without generating the heat that typically limits the density of competing hardware. The practical result is a chip capable of processing more tokens at greater speed.

For the decode stage, Etched engineered a new type of memory architecture paired with an interconnect system the company calls cluster-scale memory. According to Wachen, the system "allows many chips to connect together and use a shared memory pool at a very, very fast, low latency." The company says the combined effect of these two innovations translates into higher throughput and reduced operating costs for customers running large-scale AI workloads.

Clearing Up an Early Misconception

Etched continues to push back against a persistent misconception that dates to its earliest days as a company. When it first launched, Etched's original concept involved etching a specific model directly into its chips - essentially creating hardware that was purpose-built to run one particular frontier AI model. That approach generated both interest and skepticism, as it implied a lack of flexibility.

That original direction has since been abandoned. Etched's current systems are designed to run any frontier model, giving customers the flexibility to deploy whichever AI model best fits their needs without being locked into a single architecture.

Jane Street's Endorsement

Jane Street, the quantitative investment firm leading this round, offered a direct assessment of the hardware in a public statement accompanying the funding announcement. "We tested the chip and are pleased with the early results," the firm said. "Etched's unique approach to inference delivers the precision we will need to support our most demanding workloads. We're excited to now have our own rack running in our datacenter."

The endorsement carries weight given Jane Street's reputation as a technically sophisticated institution with demanding computational requirements. The firm's decision to not only invest but also deploy the hardware in its own data center signals a level of confidence that goes beyond a financial bet on the company's potential.

Broad Investor Support

Jane Street's participation adds to an already substantial list of backers. Etched's existing investors include some of the most recognized names in venture capital and finance. The company counts Kleiner Perkins, Sequoia Capital, Andreessen Horowitz, Peter Thiel, Tiger Global, Bain Capital Ventures, Neo, Stripes, Primary, Positive Sum, Diffusion, Argo, and Blackstone among its supporters. The breadth of that investor base - spanning traditional venture firms, hedge funds, and private equity - reflects the scale of interest in companies building specialized AI infrastructure.