AI chip startup Etched more than doubles valuation to $10.3B in new $300M round

Etched Inc., an innovative startup focused on artificial intelligence chip technology, has successfully raised $300 million in its latest funding round, bringing its valuation to an impressive $10.3 billion. This marks a significant increase, more than doubling its worth since December of the previous year. The Series C funding round was spearheaded by the prominent venture capital firm Sequoia, with participation from notable investors including SK Hynix Inc., Andreessen Horowitz, Jane Street, and Diffusion.

The core of Etched's offering is a chip specifically optimized for AI inference, a process that enables AI models to generate responses based on prompts. Unlike traditional graphics processing units (GPUs) that are designed to handle both AI training and inference tasks, Etched's chip focuses solely on inference. This specialization allows for a more efficient design, which the company believes will be crucial as demand for AI capabilities continues to grow.

Nvidia Corp.'s graphics cards are widely recognized for their ability to manage both AI training and inference workloads. However, Etched argues that by concentrating on inference, it can deliver superior performance. The company's chip will be integrated into a complete appliance that includes custom cooling components and interconnects, ensuring optimal operation.

The inference process begins with what is known as prefill processing, where the AI model interprets the user's prompt. This step relies heavily on matrix multiplications, which are complex mathematical operations that differ from standard multiplication. The efficiency of these calculations is directly linked to the chip's clock frequency; higher frequencies allow for more calculations per second. However, traditional GPUs face limitations in clock frequency due to heat generation, which can hinder performance.

Etched's chip addresses this thermal bottleneck through a technology referred to as LVI, which stands for Low Voltage Inference. By minimizing the processor's voltage, LVI effectively reduces the chip's temperature, allowing it to operate at higher clock frequencies compared to conventional graphics cards. This innovation is expected to enhance the chip's overall performance during the inference process.

Following the prefill phase, the inference workflow progresses to the decode stage, where the AI model generates a response token by token. To accelerate this process, Etched employs a mechanism called Cluster Scale Memory. This technology enables all accelerators within a rack to share a common memory pool, eliminating the need to duplicate data for each accelerator. As a result, the workflow for processing prompts becomes significantly more efficient.

Gavin Uberti, co-founder and CEO of Etched, emphasized the necessity for a fresh approach to AI infrastructure, 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."

Looking ahead, Etched plans to begin shipping its first racks this summer. To facilitate this, the company is establishing a surface mount technology (SMT) production line within an 80,000 square-foot facility located near its headquarters in San Jose. This facility will also serve as a prototyping lab, further supporting the development of its innovative chip technology.

In summary, Etched's recent funding success and subsequent valuation increase underscore the growing interest and investment in specialized AI hardware. As the demand for efficient AI inference solutions rises, the company's unique approach may position it as a key player in the evolving landscape of AI technology.

Why it matters

The substantial funding and valuation increase for Etched Inc. highlight a significant trend in the tech industry: the growing demand for specialized hardware solutions tailored to artificial intelligence tasks. As AI applications become more prevalent across various sectors, the need for efficient processing capabilities is paramount. Etched's focus on optimizing chips for inference rather than training reflects a strategic shift that could influence how future AI technologies are developed and implemented.

For creators and technologists, the implications of this development are profound. The ability to process AI tasks more efficiently can lead to faster and more responsive applications, enhancing user experiences and expanding the possibilities for innovation. Furthermore, as companies like Etched push the boundaries of chip design, it may inspire a new wave of creativity in AI application development, encouraging technologists to explore novel use cases and solutions that were previously considered impractical due to hardware limitations.

Overall, the advancements made by Etched and similar startups could pave the way for a new era of AI capabilities, where efficiency and performance are no longer constrained by traditional hardware limitations.