Etched has raised another $700 million to reach a $21 billion valuation.
The funding round took place on Tuesday and was led by Jane Street following a test purchase of the startup’s hardware.
The speed of the deal
Even within the artificial intelligence sector, this valuation jump is unusually rapid. The company was worth $5 billion in December. It secured a Series C round for $300 million at a $10.3 billion valuation in July. Investors have now doubled that figure to $21 billion in just one month.
Etched sells its technology as complete systems known as frontier inference clusters. Its rival, Nvidia, refers to similar full systems as AI factories.
Hardware built from scratch
Co-founder and COO Robert Wachen explained to TechCrunch that investor enthusiasm stems from two new components designed specifically to accelerate inference. This is the computing process that occurs after a user submits a prompt.
“Inference is built in two stages,” Wachen said, “prefill and decode.” The mathematically and compute-intensive prefill phase requires the system to understand the prompt and its context. The memory-intensive decode phase generates output tokens, which form the actual answer the user sees.
Etched developed a prefill chip that runs at low voltage. This design allows for more transistors without the typical heat issues found in other high-end AI chips. Consequently, the chip processes more tokens at higher speeds. The company also created a new type of memory and interconnect for the decode process, which it calls cluster-scale memory.
“It allows many chips to connect together and use a shared memory pool at a very, very fast, low latency,” Wachen said. The outcome, Etched claims, is improved speeds and reduced costs.
Clearing up a misconception
Etched is still dealing with the perception from its early days that it etched a particular model into its chips. This implied each chip was custom-designed for a single frontier model. That was the original plan, but it is no longer accurate. Etched’s systems can now run any frontier model.
In the announcement blog post, investment firm Jane Street stated: “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.”
Other investors in Etched include Kleiner Perkins, Sequoia Capital, Andreessen Horowitz, Peter Thiel, Tiger Global, Bain Capital Ventures, Neo, Stripes, Primary, Positive Sum, Diffusion, Argo, and Blackstone.
What it means
For teams building and running large models, the shift moves the bottleneck away from custom silicon constraints. The new hardware allows clusters to share memory pools with minimal delay. This means developers can swap models without redesigning their infrastructure. The focus remains on raw throughput and cost efficiency rather than locking into a single algorithm.
