{"slug": "etched-raises-300m-series-c-at-10-3b-valuation-to-scale-gpu-free-inference-chips", "title": "Etched Raises $300M Series C at $10.3B Valuation to Scale GPU-Free Inference Chips", "summary": "Etched, a San Jose-based AI chip startup, raised $300 million in a Series C round at a $10.3 billion valuation, led by Sequoia Capital, to scale production of its Sohu transformer-specific ASIC that claims 20x inference speedup over Nvidia's H100. The company has now raised over $1 billion in total and is in discussions for a subsequent round at a $20 billion valuation.", "body_md": "# Etched Raises $300M Series C at $10.3B Valuation to Scale GPU-Free Inference Chips\n\n- Etched closed a $300 million Series C at a $10.3 billion valuation, led by Sequoia Capital — the firm's highest-valued Series C investment\n[[1]](https://www.manilatimes.net/2026/07/23/tmt-newswire/globenewswire/etched-raises-300m-at-a-103b-valuation-to-scale-production-of-frontier-scale-inference-hardware/2390496) - The round included a16z, Jane Street, SK Hynix, and Diffusion, bringing total funding past $1 billion\n[[1]](https://www.manilatimes.net/2026/07/23/tmt-newswire/globenewswire/etched-raises-300m-at-a-103b-valuation-to-scale-production-of-frontier-scale-inference-hardware/2390496) - The company's Sohu chip is a transformer-specific ASIC built on TSMC 4nm that claims 20x inference speedup over Nvidia's H100\n[[3]](https://www.spheron.network/blog/etched-ai-sohu-vs-nvidia-transformer-asic-inference/) - Etched has 400 employees and is scaling production at a new 80,000 sq ft California facility and a Taiwan factory\n[[1]](https://www.manilatimes.net/2026/07/23/tmt-newswire/globenewswire/etched-raises-300m-at-a-103b-valuation-to-scale-production-of-frontier-scale-inference-hardware/2390496) - Prospective customers have expressed roughly $1 billion in demand for Etched's inference hardware\n[[2]](https://www.investing.com/news/company-news/etched-seeks-20-billion-valuation-in-new-ai-chip-funding-round--wsj-reports-4799566)\n\nEtched, a San Jose-based AI chip startup building specialized inference processors that eliminate the need for GPUs, has raised $300 million in a Series C round at a $10.3 billion valuation. Sequoia Capital led the round — its highest-valued Series C investment — with participation from Andreessen Horowitz, Jane Street, SK Hynix, and Diffusion [1].\n\nThe deal doubles Etched's valuation from the roughly $5 billion mark it reached in a $500 million round led by Stripes in late 2025 [2]. The company has now raised more than $1 billion in total from investors including Peter Thiel, Jump Trading, Two Sigma, Ribbit Capital, and Radical Ventures\n\n.\n\n[[1]](https://www.manilatimes.net/2026/07/23/tmt-newswire/globenewswire/etched-raises-300m-at-a-103b-valuation-to-scale-production-of-frontier-scale-inference-hardware/2390496)Founded in 2022 by Harvard dropouts Gavin Uberti, Chris Zhu, and Robert Wachen, Etched emerged from stealth in late June with working silicon and over $1 billion in signed customer contracts. The company is betting that purpose-built inference ASICs will displace Nvidia's general-purpose GPUs for running deployed AI models — a market that is growing rapidly as enterprises shift spending from training to inference [3].\n\n## The Round\n\nThe $300 million Series C values Etched at $10.3 billion post-money, making it one of the most highly valued private semiconductor startups in the world [1]. Sequoia Capital anchored the round, with a16z, Jane Street, SK Hynix, and Diffusion also participating. SK Hynix's involvement is notable given the memory maker supplies the HBM3E chips used in Etched's hardware\n\n.\n\n[[1]](https://www.manilatimes.net/2026/07/23/tmt-newswire/globenewswire/etched-raises-300m-at-a-103b-valuation-to-scale-production-of-frontier-scale-inference-hardware/2390496)Etched is already in discussions for a subsequent round at a $20 billion valuation, led by existing investor Jane Street, according to a Wall Street Journal report. Neither that deal nor its terms have been finalized [2].\n\nThe company plans to use the capital to aggressively scale production and customer deployments, including expanding its new 80,000-square-foot facility near its headquarters and a 10-megawatt Milpitas site that houses an NPI lab and in-house SMT line [1].\n\n## The Technology\n\nEtched's core product is Sohu, a transformer-specific application-specific integrated circuit (ASIC) fabricated on TSMC's 4nm process node. Unlike Nvidia's GPUs, which are designed for general-purpose parallel computing, Sohu hard-wires the transformer computation graph — attention mechanisms, linear projections, softmax, and layer normalization — directly into silicon [3].\n\nThe company claims two key innovations: Low Voltage Inference (LVI), which increases FLOP density within existing power envelopes, and Cluster Scale Memory (CSM), a hybrid SRAM/HBM subsystem that creates shared memory pools across chip clusters via a proprietary ultra-low-latency interconnect [1].\n\nEach Sohu chip carries 144GB of HBM3E memory. Early benchmarks shared by the company show 500,000 tokens per second on Meta's Llama 70B model, compared with roughly 25,000 tokens per second from an equivalent Nvidia H100 configuration — a claimed 20x speedup [3]. Etched says the architecture is model-agnostic, supporting large mixture-of-experts models like DeepSeek and Qwen, as well as non-transformer architectures like Mamba\n\n.\n\n[[1]](https://www.manilatimes.net/2026/07/23/tmt-newswire/globenewswire/etched-raises-300m-at-a-103b-valuation-to-scale-production-of-frontier-scale-inference-hardware/2390496)## The Business\n\nEtched emerged from stealth on June 30 with what it described as working A0 silicon, a rack-scale 8-chip inference system, and over $1 billion in signed customer contracts [3]. The company now employs 400 people, drawing engineers from Nvidia, Broadcom, Google's TPU team, and SK Hynix\n\n.\n\n[[1]](https://www.manilatimes.net/2026/07/23/tmt-newswire/globenewswire/etched-raises-300m-at-a-103b-valuation-to-scale-production-of-frontier-scale-inference-hardware/2390496)The startup operates a 2-megawatt datacenter running Etched hardware around the clock, a factory in Taiwan, and the new California production facility. First rack shipments are scheduled for summer 2026 [3].\n\nEtched is entering a competitive field of Nvidia challengers in the inference chip market. Cerebras Systems, Groq, SambaNova Systems, and UK-based Fractile are all pursuing variations on specialized inference silicon [2]. But Etched's rapid fundraising trajectory and the caliber of its investor syndicate suggest significant market confidence in its approach.\n\n## Why It Matters\n\nThe AI hardware market is undergoing a structural shift as inference — running trained models to serve predictions and generate responses — overtakes training as the dominant workload. Nvidia's GPUs remain the default choice for both, but their general-purpose architecture means significant silicon is wasted on capabilities inference workloads don't require. That inefficiency creates an opening for ASICs like Sohu.\n\nEtched's $10.3 billion valuation, reached just four years after founding and before large-scale commercial deliveries, reflects investor conviction that the inference hardware market is large enough to support specialized entrants alongside Nvidia. The company's claim of $1 billion in customer demand, if converted to actual revenue, would represent a meaningful share of the nascent inference chip market [2].\n\nThe involvement of SK Hynix, a leading memory manufacturer, signals potential supply chain advantages. Memory bandwidth is a critical bottleneck for inference performance, and a strategic investor relationship with a top HBM supplier strengthens Etched's position as it scales production [1].\n\n## What's Next\n\nEtched faces the standard challenge of any pre-revenue hardware startup at a premium valuation: converting customer interest into shipped products and recurring revenue. No independent third-party benchmarks of the Sohu chip have been published, and pricing remains undisclosed [3].\n\nThe reported discussions for a follow-on round at a $20 billion valuation, led by Jane Street, suggest the company is moving quickly to secure additional capital before competitive dynamics shift [2]. With first racks scheduled to ship this summer, the next six months will test whether Etched's performance claims hold up under real-world production workloads.\n\n## Companies mentioned\n\n## Further sources\n\nThe stories that matter, in one email. Free — unsubscribe anytime.", "url": "https://wpnews.pro/news/etched-raises-300m-series-c-at-10-3b-valuation-to-scale-gpu-free-inference-chips", "canonical_source": "https://mlq.ai/news/etched-raises-300m-series-c-at-103b-valuation-to-scale-gpu-free-inference-chips/", "published_at": "2026-07-23 15:54:05.415919+00:00", "updated_at": "2026-07-23 15:54:07.502732+00:00", "lang": "en", "topics": ["ai-chips", "ai-infrastructure", "ai-startups"], "entities": ["Etched", "Sequoia Capital", "Andreessen Horowitz", "Jane Street", "SK Hynix", "Nvidia", "TSMC", "Sohu"], "alternates": {"html": "https://wpnews.pro/news/etched-raises-300m-series-c-at-10-3b-valuation-to-scale-gpu-free-inference-chips", "markdown": "https://wpnews.pro/news/etched-raises-300m-series-c-at-10-3b-valuation-to-scale-gpu-free-inference-chips.md", "text": "https://wpnews.pro/news/etched-raises-300m-series-c-at-10-3b-valuation-to-scale-gpu-free-inference-chips.txt", "jsonld": "https://wpnews.pro/news/etched-raises-300m-series-c-at-10-3b-valuation-to-scale-gpu-free-inference-chips.jsonld"}}