Etched revealed on Tuesday that it secured an additional $700 million, boosting its valuation to $21 billion. This funding round was spearheaded by Jane Street, a renowned quantitative fund that invested after evaluating and acquiring the startup’s AI hardware.
The speed of this valuation increase is remarkable, even within the rapidly growing AI sector. Etched’s valuation has surged from $5 billion in December, to a $10.3 billion Series C in July, to now an astounding $21 billion – an increase of almost $11 billion in just one month.
Etched provides its AI technology as complete systems, which it terms “frontier inference clusters.” (In contrast, Nvidia, a competitor to Etched, refers to its comprehensive systems as AI factories.)
Robert Wachen, Etched’s co-founder and COO, explained to TechCrunch that the strong investor interest stems from Etched’s development of two innovative components, built from the ground up, to accelerate inference — the computational phase following a user’s prompt submission.
“Inference involves two distinct stages,” Wachen elaborated, “namely prefill and decode.” The “prefill phase” is mathematically and computationally demanding, requiring the system to interpret the user’s prompt and its context. Following this, the memory-intensive “decode” phase is where the system produces output tokens, which constitute the final answer presented to the user.
Etched has innovated a prefill chip designed for low-voltage operation, enabling a higher density of transistors without the overheating issues common in other advanced AI chips. This design facilitates quicker processing of more tokens. Additionally, for the decode process, Etched developed a novel memory type and interconnect, which it terms cluster-scale memory.
“This technology enables multiple chips to link up and access a common memory pool with exceptionally high speed and minimal latency,” Wachen stated. Etched asserts that this innovation will lead to enhanced performance and reduced expenditures.
Etched continues to address the initial misconception that its chips are designed to specifically embed and run only a single frontier model. While this was an early concept, it is no longer accurate; Etched’s current systems are capable of executing any frontier model.
In a blog post detailing the latest funding round, investment firm Jane Street commented, “Having rigorously tested the chip, we are very satisfied with its initial performance. Etched’s distinctive inference methodology provides the exact precision necessary for our most intensive workloads. We’re excited to now operate our own rack within our datacenter.”
Further investors contributing to Etched’s funding round comprise Kleiner Perkins, Sequoia Capital, Andreessen Horowitz, Peter Thiel, Tiger Global, Bain Capital Ventures, Neo, Stripes, Primary, Positive Sum, Diffusion, Argo, and Blackstone.