AI Hardware Maker Etched Secures $700M at $21B Valuation Led by Jane Street

Artificial intelligence hardware startup Etched announced Tuesday that it has raised $700 million in a new funding round that values the company at $21 billion. Quantitative trading firm Jane Street led the investment after evaluating and purchasing the startup's hardware for deployment within its own facilities, as first reported by TechCrunch AI.

The valuation step-up marks a rapid increase in capital value for the chip designer. Etched was valued at $5 billion in December before raising a $300 million Series C round at a $10.3 billion valuation in July. The newest funding round effectively doubles the company's valuation from $10.3 billion to $21 billion, representing an equity growth of nearly $11 billion in approximately one month.

Etched delivers its technology to enterprise clients as integrated hardware systems that it terms "frontier inference clusters." The system-level product line competes directly with high-density hardware deployments from chipmaker Nvidia, which refers to its complete infrastructure systems as "AI factories." Rather than focusing on model training, Etched constructs its architecture specifically to handle AI inference, the computational process that occurs after a user submits an input prompt to an AI model.

Etched co-founder and Chief Operating Officer Robert Wachen told TechCrunch AI that investor enthusiasm reflects the startup's ground-up engineering of two primary hardware components meant to speed up inference processing. Wachen explained that inference tasks split into two distinct operational phases: the prefill stage and the decode stage.

The prefill phase is mathematically and compute-intensive, requiring the system hardware to analyze and comprehend the input prompt along with any associated contextual data. To handle this stage, Etched built a custom prefill chip engineered to run at lower voltage levels. The low-voltage design allows the chip to pack in higher transistor density without producing the severe heat problems typical of standard high-end AI processors, enabling the system to process token volumes faster.

The decode phase is memory-intensive and handles the generation of output tokens, which constitute the actual visible response delivered back to the user. For this stage, Etched developed a custom memory arrangement and chip interconnect framework called "cluster-scale memory." Wachen stated that this interconnect allows multiple chips to link together and utilize a shared memory pool at high speeds with low latency, which the company claims results in faster processing speeds and reduced compute costs.

The funding round also addresses early market perceptions regarding Etched's architecture. During its initial stages, the startup intended to physically hardcode, or "etch," specific individual frontier models directly into dedicated custom silicon. While that was its original product concept, Etched confirmed that its current hardware systems are flexible and capable of running any frontier artificial intelligence model.

Jane Street confirmed its capital investment and hardware adoption in a blog post accompanying the fundraising announcement. The quantitative trading firm stated that it tested the startup's chip and found the early performance results positive, noting that Etched's inference precision meets the requirements of its most demanding computational workloads. Jane Street added that it now operates its own active Etched hardware rack inside its datacenter.

In addition to Jane Street, Etched has built an investor base that includes prominent venture capital firms and private equity institutions. Previous and returning backers in the company include Kleiner Perkins, Sequoia Capital, Andreessen Horowitz, Peter Thiel, Tiger Global, Bain Capital Ventures, Neo, Stripes, Primary, Positive Sum, Diffusion, Argo, and Blackstone.