Etched has successfully secured an additional $700 million in funding, bringing its valuation to $21 billion. This fundraising round was led by Jane Street, following the investment firm’s testing and procurement of the startup’s AI hardware.
The rapid increase in valuation is striking, even in the AI sector. Just last December, Etched was valued at $5 billion and had its valuation rise to $10.3 billion in July after securing $300 million in Series C funding. Now, within a single month, that figure has surged to $21 billion, marking an increase of nearly $11 billion.
Etched provides its AI solutions through integrated systems referred to as “frontier inference clusters.” This approach is similar to what competitor Nvidia denotes as AI factories.
Robert Wachen, co-founder and COO of Etched, stated that the excitement among investors stems from the company’s innovative development of two new components specifically designed to enhance the efficiency of inference—the phase that occurs after a user inputs a prompt.
Wachen explained, “Inference consists of two phases: prefill and decode.” In the initial “prefill phase,” which is computationally intensive, the system grasps the prompt and its context. Following this, in the “decode” phase, the system generates output tokens that reflect the final answer presented to the user.
To enhance performance, Etched has designed a prefill chip that runs at a lower voltage, enabling it to integrate more transistors without the overheating issues commonly associated with high-performance AI chips. This advancement allows for quicker processing of tokens. Additionally, the company has developed what it calls cluster-scale memory—an innovative memory type and interconnect for the decoding process.
Wachen noted, “This enables multiple chips to link together and utilize a shared memory pool with exceptional speed and minimal latency.” As a result, Etched asserts that its technology offers increased speed and reduced costs.
Despite past misconceptions that its chips were tailored to specific models—suggesting a customization for singular frontier models—Etched now clarifies that its systems can accommodate any frontier model.
In a statement regarding the funding, Jane Street expressed satisfaction with the chip’s performance: “After our testing, we are happy with the initial outcomes. Etched’s distinctive inference method provides the accuracy necessary for our most intricate workloads. We are thrilled to have our own setup operational in our data center.”
A variety of investors are supporting Etched, including Kleiner Perkins, Sequoia Capital, Andreessen Horowitz, Peter Thiel, Tiger Global, Bain Capital Ventures, Neo, Stripes, Primary, Positive Sum, Diffusion, Argo, and Blackstone.
