Sequoia leads $300M into Etched: the ASIC transformer is back in the spotlight

Context

Sequoia Capital leads a $300M funding round for Etched, a US startup specializing in AI chips (Tech in Asia, July 24, 2026). Since 2024, Etched has advocated a counterintuitive thesis: rather than a general-purpose GPU, an ASIC designed for the Transformer architecture, with an order of magnitude difference in performance per dollar and per watt when the load is purely LLM inference.

The Data

• Round amount: $300M (Tech in Asia, July 24, 2026).
• Lead investor: Sequoia Capital.
• Positioning: Dedicated Transformer ASIC (inference, prospectively training).

Analysis

Etched's bet is narrow and subject to an obvious risk - if the dominant architecture deviates from pure Transformer (Mamba, RWKV, or hybrid), the silicon becomes outdated. But the opposite is also true: as long as Transformer remains the reference, a dedicated ASIC will have a structural advantage of an order of magnitude over GPUs. Sequoia bets that the window remains open for 3-5 years, which is enough to write a success story at 10-30x.

Scenarios

• Base: Etched delivers its first silicon in 2027, with 2 to 3 hyperscalers in early access (Amazon Trainium/AWS Neuron style).
• Optimistic: Captures 3-5% of the LLM inference market by 2029, exit or IPO at $15-25B.
• Pessimistic: The Transformer architecture shifts to a hybrid (SSM+Attn), Etched pivots or is absorbed at a low premium.

Implications for the Professional

For a CTO building a large-scale inference platform, Etched enters the list of "credible alternatives to watch" alongside Cerebras, Groq, SambaNova, Tenstorrent. None of these players has yet demonstrated a massive replacement of Nvidia - but each is targeting a segment (batch offline, low-latency, edge).

Signals to Watch

• Tape-out date and availability of first silicons.
• Signing of a hyperscaler partner (AWS, Azure, GCP, Oracle).
• Nvidia's reaction to the pricing of the Blackwell Ultra inference generation.

Our Take

Etched's bet is the riskiest and most telling of the AI ASIC wave. Sequoia is putting capital behind a specific architectural choice at a time when the market is beginning to distinguish training and inference in terms of silicon. If Etched delivers, it's one of the few real events of disruption of the Nvidia/AMD duopoly in three years.