Etched’s valuation doubles to $21B in a month
In a move that underscores the blistering pace of the artificial intelligence hardware sector, Etched has secured $700 million in fresh funding, catapulting its valuation to $21 billion. The round was spearheaded by the renowned quantitative trading firm Jane Street, which opted to back the startup after successfully integrating and testing Etched’s proprietary AI cluster hardware within its own infrastructure.
A Meteoric Rise in Valuation
The speed of Etched’s financial ascent is nothing short of extraordinary. Valued at $5 billion as recently as December, the company saw its valuation climb to $10.3 billion during a $300 million Series C round in July. Now, just one month later, that figure has effectively doubled, adding roughly $11 billion to its market capitalization in a matter of weeks.
Redefining Inference Architecture
Etched distinguishes itself by delivering complete "frontier inference clusters"—systems designed specifically to optimize the computing processes that occur after a user submits a prompt. According to Robert Wachen, co-founder and COO, the company’s competitive edge lies in two custom-engineered components designed to tackle the distinct stages of inference:
- Prefill Phase: Etched developed a specialized chip that operates at a lower voltage, enabling a higher density of transistors without the thermal management issues typically associated with high-end AI processors.
- Decode Phase: The company introduced a novel interconnect and memory architecture dubbed "cluster-scale memory," which facilitates a high-speed, low-latency shared memory pool across multiple chips.
"Inference is built in two stages: prefill and decode," Wachen explained. "The prefill phase requires the system to understand the prompt and context, while the decode phase is memory-intensive, generating the actual output tokens the user sees."
Dispelling Myths and Scaling Forward
While early industry skepticism suggested that Etched’s hardware was hard-coded to support only a single model, the company has clarified that its current systems are fully capable of running any frontier-grade model.
Jane Street expressed strong confidence in this pivot, noting in a blog post:
"We tested the chip and are pleased with the early results. 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 startup’s impressive cap table includes heavyweights such as Sequoia Capital, Andreessen Horowitz, Kleiner Perkins, Peter Thiel, Tiger Global, Bain Capital Ventures, Neo, Stripes, Primary, Positive Sum, Diffusion, Argo, and Blackstone. With this latest infusion of capital, Etched is positioned to aggressively scale its vision of faster, more cost-effective AI inference.