Etched funding values AI chip startup at $10.3 billion
Etched raised a $300 million Series C led by Sequoia to scale inference hardware, more than doubling its valuation since December.
By Dominic Okoye · Staff Writer
· 3 min read
Etched funding reached $300 million in a new Series C round, valuing the AI chip startup at $10.3 billion as it prepares to ship its first inference hardware racks this summer. The round more than doubles the company’s valuation since December and gives Etched fresh capital for production at a time when AI infrastructure buyers are hunting for alternatives to general-purpose GPUs.
Sequoia led the financing, according to Etched. SK Hynix, described as the world’s largest supplier of memory for AI chips, also participated, along with Andreessen Horowitz, Jane Street and Diffusion. Etched did not disclose revenue, customer commitments or headcount in its announcement.
The company is pitching a narrower approach than Nvidia Corp.’s broadly used graphics processors. Nvidia GPUs are built to handle both AI training and inference. Etched says its chip is designed only for inference, the stage when a deployed model responds to prompts, and that this focus lets it improve efficiency versus hardware built for multiple workloads.
What is Etched building?
Etched is developing an AI inference processor that it plans to sell inside a rack-scale appliance with custom cooling and interconnects. Inference is the compute work that happens after a model has been trained, when the system takes a user prompt and generates an answer.
The company describes two main pieces of its technical approach. The first is aimed at the prefill step, where a model processes the prompt before generating output. That work relies heavily on matrix multiplications, and Etched says the number of those operations a chip can perform is tied to clock frequency.
Etched claims its LVI mechanism reduces processor voltage and heat, allowing higher clock frequencies than conventional GPUs that must hold down clock rates to manage thermals. The company has not provided independent benchmark data in the announcement to support the performance claims.
The second piece is aimed at the decode phase, where the model generates the answer token by token. Etched says its Cluster Scale Memory system lets accelerators in a rack use shared memory, reducing the need to send separate copies of the same data to each accelerator. The company says that makes prompt processing more efficient.
Why the round matters for AI infrastructure
The size of the round and the valuation increase show that investors are still willing to price specialized AI hardware aggressively, especially for inference. Training has dominated AI infrastructure spending, but inference capacity becomes the recurring cost center once models are deployed at scale.
Etched co-founder and Chief Executive Gavin Uberti said the industry needs new infrastructure rather than incremental changes to current hardware, and framed the round as evidence that investors agree. That is Etched’s case, not proof that buyers will shift meaningful workloads away from incumbent GPU platforms.
The production plan is more concrete. Etched said it will begin shipping its first racks this summer. To speed assembly, the company is building a surface mount technology production line in an 80,000-square-foot facility near its San Jose, California headquarters. The same site will also serve as a prototyping lab.
The next test is execution: turning a high valuation and a specialized architecture into shipped systems. Etched has disclosed the round size, valuation, investors and manufacturing plans, but not unit economics, order volume or production capacity.
This story draws on original reporting from SiliconANGLE.
