Etched Unleashes New Era for AI Inference with $800M Funding, $1B Contracts
July 8, 2026, 3:03 pm
Google
Location: United States, New York
AI hardware innovator Etched emerged from stealth. It announced $800 million in funding. The company secured over $1 billion in customer contracts. Its valuation stands at $5 billion. Etched specializes in high-performance AI inference chips. These chips power rack-scale computing systems. The technology features Low-Voltage Inference and Cluster-Scale Memory. These innovations prevent thermal throttling. They ensure ultra-low-latency memory access. Etched's systems run large AI models efficiently. The firm leverages TSMC's N4P process for chip manufacturing. Production is ramping up. First rack shipments are expected this summer. Etched aims to redefine AI datacenter capabilities. It offers a powerful alternative for demanding AI workloads. This launch marks a significant moment in the AI semiconductor race.
Etched, an AI inference chip startup, made its public debut. The company launched with substantial financial backing. It raised $800 million in venture capital. This funding came across multiple undisclosed rounds. The latest investment valued Etched at $5 billion. This round closed in December 2025. It secured backing from prominent investors. VentureTech Alliance participated. This fund has ties to Taiwan Semiconductor Manufacturing Co. (TSMC). More than a dozen other backers joined. These include leading AI figures. Geoffrey Hinton, Fei-Fei Li, and Andrej Karpathy lent their support. Peter Thiel and Stanley Druckenmiller also invested. Other key firms included Jane Street, Hudson River Trading, Jump Trading, Two Sigma, Stripes, Ribbit Capital, Radical Ventures, Primary VC, and Positive Sum. This robust financial foundation fuels Etched's ambitious plans.
The company has already secured significant customer demand. Etched announced over $1 billion in signed customer contracts. This reflects strong market confidence. Enterprises seek powerful AI solutions. Etched positions itself to meet this demand. Its rack-scale systems are now undergoing validation with customers. First product shipments are slated for this summer. This rapid progression highlights Etched’s operational efficiency.
Etched focuses exclusively on AI inference. Inference is the stage where AI models process user requests. It constitutes a major operational cost for AI companies. Traditional GPUs optimize for both training and inference. Etched’s singular focus brings significant advantages. It allows for reduced power consumption. Unnecessary training-optimized circuits are removed. Engineers can instead dedicate more circuitry to inference. This boosts processing speeds for real-time AI applications.
The company’s technology incorporates several performance enhancements. A key innovation is Low-Voltage Inference (LVI). AI chips often face thermal throttling. High computational loads generate excessive heat. Chips must then lower their clock rates. This slows down inference significantly. Etched’s LVI technology mitigates this issue. It reduces the need for thermal throttling. The company claims its chip can run massive AI models. A trillion-parameter model can operate at "80%+ peak FLOPs." This occurs without reducing clock rates. This translates to a considerable inference speedup. Etched reports its FLOP density as several times higher. It outperforms existing AI processors.
Another breakthrough is Cluster-Scale Memory. AI workloads demand immense memory. Fast access to this data is critical. Etched’s appliance uses a hybrid memory design. It combines SRAM and HBM memory. SRAM offers the fastest RAM available. It stores the most crucial data. HBM memory provides greater capacity. It handles other information. This dual-memory approach optimizes both speed and scale.
AI chips within a rack often need shared memory access. Etched developed a custom interconnect. This facilitates seamless data movement. The system features a shared memory pool. It processes requests with lower latency. This surpasses earlier technologies. Cluster-Scale Memory avoids trade-offs. It bypasses limitations of SRAM-only or 3D DRAM chips. This ensures high throughput and interactivity.
Etched’s product is a comprehensive system. It is a rack-scale inference appliance. This system integrates multiple chips. They are installed on custom circuit boards. Custom cold plates are also part of the design. These components are vital for liquid cooling. Cold plates channel heat from chips. They direct it into the rack’s coolant system. This holistic approach ensures optimal performance and thermal management.
Chip production relies on TSMC’s advanced N4P process. This is an enhanced five-nanometer node. It offers 11% better performance than the original. Etched’s first prototype chips rolled off TSMC’s N4P lines earlier this year. This signifies successful silicon development. The company is now actively ramping up production.
Etched was founded in 2022. Gavin Uberti serves as CEO. Rob Wachen is co-founder. Both are Harvard alumni. They received Thiel Foundation fellowships. Their vision was to build specialized AI chips. The initial fundraising was challenging. Major investors hesitated despite detailed proposals. Market conditions have since transformed dramatically.
The company boasts a robust team. Over 400 professionals comprise its workforce. They hail from top technology firms. These include NVIDIA, Broadcom, Google TPU, and SK Hynix. Experts from HFT quant firms also contribute. This talent pool brings deep expertise. It covers hardware design, software optimization, and high-performance computing.
Etched has established significant operational infrastructure. It built a Taiwan factory. This supports 24/7 engineering efforts. Its San Jose headquarters houses critical facilities. These include a data center, a test house, and an NPI prototyping lab. This integration puts design, validation, and production under one roof. The company has a clear path to gigawatt-scale operations by 2027.
Etched enters a competitive landscape. Nvidia dominates the AI chip market. Other players like Cerebras and Groq also innovate. Tech giants like Amazon, Google, and Microsoft develop their own AI chips. OpenAI has also announced custom silicon plans. Etched's focused approach and technical breakthroughs offer differentiation. Its emphasis on power efficiency and sustained performance for inference is critical. Datacenter operators increasingly prioritize these metrics.
The company’s systems are designed for modern AI models. They support DeepSeek, Qwen, Mamba, and Llama. They accommodate models of all shapes. Arbitrarily large numbers of parameters are handled. This versatility ensures future-proofing. Etched aims to power the next generation of AI services. Its technology reduces operational costs. It accelerates deployment of advanced AI applications. This makes Etched a key player in the evolving AI hardware ecosystem.
Etched, an AI inference chip startup, made its public debut. The company launched with substantial financial backing. It raised $800 million in venture capital. This funding came across multiple undisclosed rounds. The latest investment valued Etched at $5 billion. This round closed in December 2025. It secured backing from prominent investors. VentureTech Alliance participated. This fund has ties to Taiwan Semiconductor Manufacturing Co. (TSMC). More than a dozen other backers joined. These include leading AI figures. Geoffrey Hinton, Fei-Fei Li, and Andrej Karpathy lent their support. Peter Thiel and Stanley Druckenmiller also invested. Other key firms included Jane Street, Hudson River Trading, Jump Trading, Two Sigma, Stripes, Ribbit Capital, Radical Ventures, Primary VC, and Positive Sum. This robust financial foundation fuels Etched's ambitious plans.
The company has already secured significant customer demand. Etched announced over $1 billion in signed customer contracts. This reflects strong market confidence. Enterprises seek powerful AI solutions. Etched positions itself to meet this demand. Its rack-scale systems are now undergoing validation with customers. First product shipments are slated for this summer. This rapid progression highlights Etched’s operational efficiency.
Etched focuses exclusively on AI inference. Inference is the stage where AI models process user requests. It constitutes a major operational cost for AI companies. Traditional GPUs optimize for both training and inference. Etched’s singular focus brings significant advantages. It allows for reduced power consumption. Unnecessary training-optimized circuits are removed. Engineers can instead dedicate more circuitry to inference. This boosts processing speeds for real-time AI applications.
The company’s technology incorporates several performance enhancements. A key innovation is Low-Voltage Inference (LVI). AI chips often face thermal throttling. High computational loads generate excessive heat. Chips must then lower their clock rates. This slows down inference significantly. Etched’s LVI technology mitigates this issue. It reduces the need for thermal throttling. The company claims its chip can run massive AI models. A trillion-parameter model can operate at "80%+ peak FLOPs." This occurs without reducing clock rates. This translates to a considerable inference speedup. Etched reports its FLOP density as several times higher. It outperforms existing AI processors.
Another breakthrough is Cluster-Scale Memory. AI workloads demand immense memory. Fast access to this data is critical. Etched’s appliance uses a hybrid memory design. It combines SRAM and HBM memory. SRAM offers the fastest RAM available. It stores the most crucial data. HBM memory provides greater capacity. It handles other information. This dual-memory approach optimizes both speed and scale.
AI chips within a rack often need shared memory access. Etched developed a custom interconnect. This facilitates seamless data movement. The system features a shared memory pool. It processes requests with lower latency. This surpasses earlier technologies. Cluster-Scale Memory avoids trade-offs. It bypasses limitations of SRAM-only or 3D DRAM chips. This ensures high throughput and interactivity.
Etched’s product is a comprehensive system. It is a rack-scale inference appliance. This system integrates multiple chips. They are installed on custom circuit boards. Custom cold plates are also part of the design. These components are vital for liquid cooling. Cold plates channel heat from chips. They direct it into the rack’s coolant system. This holistic approach ensures optimal performance and thermal management.
Chip production relies on TSMC’s advanced N4P process. This is an enhanced five-nanometer node. It offers 11% better performance than the original. Etched’s first prototype chips rolled off TSMC’s N4P lines earlier this year. This signifies successful silicon development. The company is now actively ramping up production.
Etched was founded in 2022. Gavin Uberti serves as CEO. Rob Wachen is co-founder. Both are Harvard alumni. They received Thiel Foundation fellowships. Their vision was to build specialized AI chips. The initial fundraising was challenging. Major investors hesitated despite detailed proposals. Market conditions have since transformed dramatically.
The company boasts a robust team. Over 400 professionals comprise its workforce. They hail from top technology firms. These include NVIDIA, Broadcom, Google TPU, and SK Hynix. Experts from HFT quant firms also contribute. This talent pool brings deep expertise. It covers hardware design, software optimization, and high-performance computing.
Etched has established significant operational infrastructure. It built a Taiwan factory. This supports 24/7 engineering efforts. Its San Jose headquarters houses critical facilities. These include a data center, a test house, and an NPI prototyping lab. This integration puts design, validation, and production under one roof. The company has a clear path to gigawatt-scale operations by 2027.
Etched enters a competitive landscape. Nvidia dominates the AI chip market. Other players like Cerebras and Groq also innovate. Tech giants like Amazon, Google, and Microsoft develop their own AI chips. OpenAI has also announced custom silicon plans. Etched's focused approach and technical breakthroughs offer differentiation. Its emphasis on power efficiency and sustained performance for inference is critical. Datacenter operators increasingly prioritize these metrics.
The company’s systems are designed for modern AI models. They support DeepSeek, Qwen, Mamba, and Llama. They accommodate models of all shapes. Arbitrarily large numbers of parameters are handled. This versatility ensures future-proofing. Etched aims to power the next generation of AI services. Its technology reduces operational costs. It accelerates deployment of advanced AI applications. This makes Etched a key player in the evolving AI hardware ecosystem.