Etched Raises $300M, Valued at $10.3B for AI Inference Chips

Etched Inc. has secured a substantial $300 million funding infusion, propelling its valuation to an impressive $10.3 billion. This latest Series C round, which more than doubles the company's previous valuation from December, signals robust investor confidence in its specialized approach to artificial intelligence hardware. The funding was spearheaded by Sequoia, with significant participation from industry heavyweights including SK Hynix Inc., a critical supplier in the AI memory market, alongside Andreessen Horowitz, Jane Street, and Diffusion.

The company is carving out a distinct niche by focusing its chip architecture exclusively on AI inference, the process of deploying trained AI models to generate outputs. This contrasts with general-purpose GPUs, like those from Nvidia Corp., which are engineered for both training and inference. By dedicating its silicon solely to inference, Etched aims to deliver superior efficiency and performance for this specific, high-demand workload. The firm plans to deliver its processors within integrated appliances featuring custom cooling and high-speed interconnects.

At the heart of Etched's innovation is its proprietary LVI technology, designed to circumvent the thermal limitations that often constrain traditional processors. By minimizing voltage, the LVI mechanism effectively reduces heat generation, allowing the chip to operate at higher clock frequencies. This is particularly advantageous during the initial "prefill" phase of AI inference, where models process user prompts through computationally intensive matrix multiplications. The ability to sustain higher frequencies directly translates to faster understanding of complex inputs.

Further enhancing its inference capabilities, Etched employs a mechanism called Cluster Scale Memory. This technology facilitates shared memory access across multiple accelerators within a rack. By eliminating the need to duplicate data for each processing unit, it streamlines the "decode" stage of inference, where AI models generate responses token by token. This shared memory architecture significantly boosts the overall efficiency of the prompt processing pipeline.

Etched co-founder and CEO Gavin Uberti emphasized the necessity of novel hardware solutions, stating, \"The infrastructure required to serve frontier AI sustainably and economically was never going to come from incremental improvements to existing hardware. This round reflects a growing industry conviction that the challenge demands a new entrant willing to rebuild the stack from first principles.\" The company is preparing to commence shipments of its first server racks this summer, supported by an in-house surface mount technology production line designed for rapid assembly.

This significant funding round underscores the escalating demand for specialized AI hardware. As AI models become more sophisticated and widely deployed, the need for efficient, cost-effective inference solutions intensifies. Etched's success highlights a broader market trend where dedicated hardware architectures are gaining traction to address the unique computational demands of AI, potentially reshaping the competitive dynamics within the semiconductor industry.