Funding and Market Position
Published 7/23/2026, 6:00:26 PM
Etched's $300M Series C raise, announced on July 23, 2026, at a $10.3 billion valuation, signals a shift from general-purpose GPU dominance toward specialized "Transformer-only" hardware. By securing over $1.1 billion in total funding and $1 billion in pre-orders, Etched is positioning its "Sohu" chip as a direct challenger to NVIDIA’s inference market share, with significant downstream effects on decentralized compute protocols.
Funding and Market Position
The Series C was led by Sequoia Capital, with participation from a16z, SK Hynix, and Jane Street [Source: https://www.eetimes.com/etched-raises-300m-with-1b-in-pre-orders/]. This round more than doubled Etched's valuation from $5B in late 2025 to $10.3B.
| Metric | Value | Source |
|---|---|---|
| Series C Raise | $300 Million | EE Times |
| Post-Money Valuation | $10.3 Billion | EE Times |
| Total Funding | ~$1.1 Billion | EE Times |
| Pre-order Backlog | $1 Billion+ | EE Times |
Technical Specialization: The Sohu Chip
Etched builds Application-Specific Integrated Circuits (ASICs) optimized specifically for AI inference workloads. Their flagship chip, Sohu, "etches" the Transformer architecture (used by Llama, GPT, and Claude) directly into the silicon [Source: https://thelogic.co/briefing/radical-ventures-gomez-hinton-back-etched-to-build-hardware-to-run-ai/].
- Performance: Etched claims Sohu is >10x faster than NVIDIA’s H100 for inference.
- Throughput: Capable of processing 500,000 tokens per second on Llama 70B, compared to approximately 25,000 on standard GPU setups.
- Efficiency: One Sohu server is marketed as providing the equivalent throughput of 160 NVIDIA H100 GPUs.
Impact on Crypto-AI Competition
The emergence of high-performance, low-cost ASICs like Sohu creates a new competitive dynamic for blockchain-based AI protocols:
- Commoditization of Compute: Decentralized Physical Infrastructure Networks (DePIN) like Akash (AKT) and Render (RNDR) currently rely on the scarcity and high cost of NVIDIA GPUs. If Etched successfully reduces inference costs by an order of magnitude, the "cheap compute" value proposition of these networks may shift from general GPU rental to the necessity of supporting specialized ASICs.
- Inference-Heavy Applications: Low-latency hardware enables more complex crypto-native AI use cases, such as Zero-Knowledge Machine Learning (ZK-ML) and real-time on-chain AI agents, which were previously cost-prohibitive on standard hardware.
- Protocol Pivot: Projects like Bittensor (TAO) or Fetch.ai (ASI) may see a shift where the competitive advantage moves from "who has the most GPUs" to "who has the most efficient specialized hardware."
Competitive Risks
While Etched has strong backing from figures like Geoffrey Hinton [Source: https://thelogic.co/briefing/radical-ventures-gomez-hinton-back-etched-to-build-hardware-to-run-ai/], it faces a "Transformer-only" risk. If the AI industry moves away from the Transformer architecture toward newer models (e.g., State-Space Models), Etched's ASICs could become obsolete, whereas NVIDIA's general-purpose GPUs would remain flexible.
Note: A separate project using the "Etched" branding (0xetch) launched a token in early 2026; this is a distinct entity from the hardware company and its security has not been independently verified.