Etched secured $700 million in August 2026 funding at a $21 billion valuation, led by Jane Street, which also became its first customer after testing and deploying the startup’s inference hardware racks.
Etched raised $700 million at a $21 billion valuation in a round led by Jane Street. This more than doubled the company’s valuation from the $10.3 billion set in its $300 million Series C just weeks earlier (July 23, 2026, led by Sequoia). That Series C itself had doubled the prior $5 billion valuation from a $500 million round closed in December 2025. Total capital raised now stands at approximately $1.9 billion.
Jane Street led the round after testing Etched’s hardware and becoming its first customer. The quantitative trading firm received its first rack last month and is deploying the technology into production workloads. Jane Street stated: “We tested the chip and are pleased with the early results. Etched’s unique approach to inference delivers the precision we will need to support our most demanding workloads. We’re excited to now have our own rack running in our data center.”
Participants include Kleiner Perkins, Sequoia Capital, Andreessen Horowitz, Peter Thiel, Tiger Global, Bain Capital Ventures, Neo, Stripes, Primary, Positive Sum, Diffusion, Argo, and Blackstone. Earlier backers and angels include SK Hynix, VentureTech Alliance (a TSMC-linked fund), Hudson River Trading, Jump Trading, Two Sigma, Ribbit Capital, Radical Ventures, Stanley Druckenmiller, and AI figures such as Geoffrey Hinton, Andrej Karpathy, Fei-Fei Li, Noam Brown, Arthur Mensch, and others.
The capital supports ramping production toward gigawatt scale, including factories, global supply chains, fleet software, and related infrastructure. Etched has opened a Taiwan factory and built a data center, test house, and NPI prototyping lab in its San Jose office. It is also constructing three generations of hardware in parallel.
What is Etched?
Etched, founded in 2022 by Harvard dropouts and Thiel Fellows Gavin Uberti (CEO), Robert Wachen (President/COO), and Chris Zhu, builds full rack scale “frontier inference clusters.” These co-designed systems integrate chips, packages, PCBs, cold plates, interconnects, software, and manufacturing methods optimized for high throughput, low latency, cost, and power efficiency on both prefill and decode phases of inference. The systems target frontier models, including many trillion parameter mixture of experts (MoEs), long context, and agentic workloads, and can run a range of architectures (transformers, Mamba, and others).
Key technical innovations include:
- Low Voltage Inference (LVI): Math blocks run at under half the voltage of most AI chips. This reduces power and heat (power scales with the square of voltage), enabling higher FLOPs density and sustained utilization. The company claims trillion parameter sparse MoEs can run at 80%+ of peak FLOPs without thermal throttling. Achieving this required co-design across transistors, splittable math arrays, circuit techniques, tiling/scheduling algorithms, power delivery, VRMs, packaging, and cooling.
- Cluster Scale Memory (CSM): A proprietary ultra low latency, high bandwidth interconnect creates a shared low latency memory pool across the scale-up domain. An HBM/SRAM hybrid design aims to deliver high capacity and low mem2mem latency simultaneously, addressing bottlenecks in pure HBM systems (for decode speeds) and pure SRAM systems (for density and capacity). This supports high throughput and interactivity without some of the cost, yield, reliability, or thermal tradeoffs of alternatives.
The A0 silicon returned successfully from TSMC’s N4P process earlier in 2026 (first pass silicon success in under three years from seed). Early customer tests and internal validation claim state of the art throughput, latency, and power efficiency. Etched reports more than $1 billion in customer contracts from frontier AI companies, clouds, and others. First racks began shipping in summer 2026; a 2-megawatt cluster operates at the San Jose headquarters for customer testing.
Recommended: CAIS Raises $170 Million In Series D Funding Round
The team exceeds 400 engineers, drawn from NVIDIA (including platform and HGX/DGX experience), Google TPU, Broadcom, SK Hynix, TSMC, Apple, and others. Leadership includes Mark Ross (CTO, ex-Cypress), Brian Loiler (VP Platform, longtime NVIDIA), David Munday (VP Software, TPU background), and specialists in production, ASIC architecture, and finance. The company emphasizes vertical integration and the principle that “production is the product.”
Inference (running trained models) is viewed as a massive and growing infrastructure market, with competition centered on tokens per dollar and tokens per watt. Etched positions its full system approach against general purpose GPU systems (such as those from NVIDIA) by specializing for inference workloads while maintaining flexibility across model types. The rapid valuation trajectory, from $5 billion (December 2025) to $10.3 billion (July 2026) to $21 billion (August 2026), reflects investor conviction in execution speed, first pass silicon success, early customer validation (especially a sophisticated quant firm putting hardware into production), and the scale of booked demand.
The funding and first delivery mark a transition from development and validation into volume production and broader customer rollout. Etched is targeting gigawatt scale deployment to expand access to frontier models, currently limited to a small fraction of the global population. Challenges ahead include scaling manufacturing yields, supply chain reliability, software/fleet management, and competing in a capital intensive market where historical semiconductor successes and failures both exist. The combination of technical claims, customer traction, and investor quality positions the company as one of the more advanced specialized inference hardware efforts emerging after the generative AI wave.
Please email us your feedback and news tips at hello(at)superbcrew.com
