Etched Reaches $10.3B Valuation, Redefining AI Inference Hardware
July 26, 2026, 9:34 am
Etched Inc. recently raised $300M in Series C funding, boosting its valuation to $10.3B. This AI chip startup specializes in purpose-built inference hardware, directly challenging general-purpose GPUs. Its core innovations, Low Voltage Inference and Cluster Scale Memory, optimize for power efficiency and memory performance. Etched addresses the growing demands of frontier AI models, aiming to revolutionize global AI infrastructure with its specialized systems and rapid production expansion. This substantial capital propels its mission to deliver gigawatt-scale AI compute.
AI demands relentless compute. General-purpose hardware, like graphics processing units (GPUs), once dominated this space. The landscape shifts. Training models and running them, known as inference, are distinct tasks. Inference rapidly becomes the larger, recurring workload. Billions of AI requests occur daily. Current hardware faces limitations for this massive scale. Power consumption, thermal management, and memory access pose significant hurdles. A new era demands purpose-built solutions. Specialized AI chips offer greater efficiency.
Etched, a leader in AI inference hardware, tackles these challenges head-on. Its Low Voltage Inference (LVI) technology represents a critical innovation. Traditional processors limit clock frequencies. They must avoid excessive heat generation. LVI minimizes processor voltage. This directly lowers chip temperature. Cooler chips operate at higher clock frequencies. More calculations per second become possible. LVI significantly boosts performance within existing power envelopes. It increases tokens generated per watt. This translates to more AI requests served with less electricity. Greater efficiency lowers operational costs for AI data centers.
Memory performance is another bottleneck for AI inference. Etched introduces Cluster Scale Memory (CSM). This is a hybrid memory architecture. CSM treats an entire computing cluster as a unified memory domain. It creates a shared pool of fast static random-access memory (SRAM). Proprietary interconnects ensure low-latency, high-bandwidth access. Data no longer requires separate copies for each accelerator. This streamlines prompt processing. Processors gain faster access to model parameters and intermediate data. CSM enhances latency. It avoids limitations of SRAM-only or three-dimensional DRAM designs. This innovation is crucial for complex AI models. Mixture-of-Experts (MoE) and Mamba models thrive on efficient memory access.
Etched's strategy stands apart. It focuses exclusively on inference. General-purpose chips attempt both training and inference. Etched optimizes for one. This specialization delivers lower latency, superior power, and cost efficiency. The hardware is architecture-agnostic. Customers run diverse AI models without commitment to a single design. The company supports leading models like DeepSeek and Qwen (MoE). It also supports non-transformer systems such as Mamba. This foresight prepares customers for evolving frontier model architectures. Demand for these specialized systems currently outstrips supply. Customers transition from technical evaluations into production deployments.
Etched recently secured $300 million in Series C funding. This round, led by Sequoia Capital, boosted its valuation to $10.3 billion. Other major investors include Andreessen Horowitz, Jane Street, Diffusion, and SK Hynix. Total funding now exceeds $1 billion. This marks a significant milestone for the company. The substantial capital fuels aggressive expansion. Manufacturing capacity will increase. Product development accelerates. Etched plans extensive deployment of frontier-scale inference clusters for customers. This investment validates the company's vision and its groundbreaking technology.
Physical infrastructure is expanding rapidly. Etched established a manufacturing facility in Taiwan earlier in 2026. A new 80,000 square-foot facility opened near its San Jose headquarters. Located in Milpitas, California, this 10-megawatt site houses a New Product Introduction (NPI) laboratory. An internal Surface Mount Technology (SMT) line facilitates component assembly and testing. These facilities shorten development cycles. They provide greater control over production and validation. The San Jose headquarters also includes a prototyping facility and a 2-megawatt data center. This internal data center operates the company’s hardware continuously. Etched aims for gigawatt-scale infrastructure deployment. It seeks to control how quickly it can produce, validate, and improve its systems.
Etched has rapidly grown to approximately 400 employees. Its team comprises engineers from industry giants. NVIDIA, Broadcom, Google's Tensor Processing Unit organization, and SK Hynix are represented. The company operates as an in-person organization. This fosters collaboration and rapid iteration. Recruitment continues, targeting top engineering talent. This experienced workforce drives the company's ambitious goals. Etched is building a defining company in inference hardware. It is leveraging its deep expertise to innovate at a foundational level.
Inference is becoming a monumental market. AI integrates into every sector. Etched envisions purpose-built compute powering the majority of this global inference. Its foundational approach is disrupting existing paradigms. The company offers a sustainable, economic solution for frontier AI. This represents a significant leap forward for artificial intelligence deployment. Etched shapes the future of AI infrastructure. It aims to deliver frontier inference without the typical trade-offs between speed, cost, and scale. The company believes incremental improvements to existing hardware cannot serve frontier AI sustainably. Etched builds its stack from first principles. It delivers unprecedented efficiency, one specialized chip at a time.
AI demands relentless compute. General-purpose hardware, like graphics processing units (GPUs), once dominated this space. The landscape shifts. Training models and running them, known as inference, are distinct tasks. Inference rapidly becomes the larger, recurring workload. Billions of AI requests occur daily. Current hardware faces limitations for this massive scale. Power consumption, thermal management, and memory access pose significant hurdles. A new era demands purpose-built solutions. Specialized AI chips offer greater efficiency.
Etched, a leader in AI inference hardware, tackles these challenges head-on. Its Low Voltage Inference (LVI) technology represents a critical innovation. Traditional processors limit clock frequencies. They must avoid excessive heat generation. LVI minimizes processor voltage. This directly lowers chip temperature. Cooler chips operate at higher clock frequencies. More calculations per second become possible. LVI significantly boosts performance within existing power envelopes. It increases tokens generated per watt. This translates to more AI requests served with less electricity. Greater efficiency lowers operational costs for AI data centers.
Memory performance is another bottleneck for AI inference. Etched introduces Cluster Scale Memory (CSM). This is a hybrid memory architecture. CSM treats an entire computing cluster as a unified memory domain. It creates a shared pool of fast static random-access memory (SRAM). Proprietary interconnects ensure low-latency, high-bandwidth access. Data no longer requires separate copies for each accelerator. This streamlines prompt processing. Processors gain faster access to model parameters and intermediate data. CSM enhances latency. It avoids limitations of SRAM-only or three-dimensional DRAM designs. This innovation is crucial for complex AI models. Mixture-of-Experts (MoE) and Mamba models thrive on efficient memory access.
Etched's strategy stands apart. It focuses exclusively on inference. General-purpose chips attempt both training and inference. Etched optimizes for one. This specialization delivers lower latency, superior power, and cost efficiency. The hardware is architecture-agnostic. Customers run diverse AI models without commitment to a single design. The company supports leading models like DeepSeek and Qwen (MoE). It also supports non-transformer systems such as Mamba. This foresight prepares customers for evolving frontier model architectures. Demand for these specialized systems currently outstrips supply. Customers transition from technical evaluations into production deployments.
Etched recently secured $300 million in Series C funding. This round, led by Sequoia Capital, boosted its valuation to $10.3 billion. Other major investors include Andreessen Horowitz, Jane Street, Diffusion, and SK Hynix. Total funding now exceeds $1 billion. This marks a significant milestone for the company. The substantial capital fuels aggressive expansion. Manufacturing capacity will increase. Product development accelerates. Etched plans extensive deployment of frontier-scale inference clusters for customers. This investment validates the company's vision and its groundbreaking technology.
Physical infrastructure is expanding rapidly. Etched established a manufacturing facility in Taiwan earlier in 2026. A new 80,000 square-foot facility opened near its San Jose headquarters. Located in Milpitas, California, this 10-megawatt site houses a New Product Introduction (NPI) laboratory. An internal Surface Mount Technology (SMT) line facilitates component assembly and testing. These facilities shorten development cycles. They provide greater control over production and validation. The San Jose headquarters also includes a prototyping facility and a 2-megawatt data center. This internal data center operates the company’s hardware continuously. Etched aims for gigawatt-scale infrastructure deployment. It seeks to control how quickly it can produce, validate, and improve its systems.
Etched has rapidly grown to approximately 400 employees. Its team comprises engineers from industry giants. NVIDIA, Broadcom, Google's Tensor Processing Unit organization, and SK Hynix are represented. The company operates as an in-person organization. This fosters collaboration and rapid iteration. Recruitment continues, targeting top engineering talent. This experienced workforce drives the company's ambitious goals. Etched is building a defining company in inference hardware. It is leveraging its deep expertise to innovate at a foundational level.
Inference is becoming a monumental market. AI integrates into every sector. Etched envisions purpose-built compute powering the majority of this global inference. Its foundational approach is disrupting existing paradigms. The company offers a sustainable, economic solution for frontier AI. This represents a significant leap forward for artificial intelligence deployment. Etched shapes the future of AI infrastructure. It aims to deliver frontier inference without the typical trade-offs between speed, cost, and scale. The company believes incremental improvements to existing hardware cannot serve frontier AI sustainably. Etched builds its stack from first principles. It delivers unprecedented efficiency, one specialized chip at a time.