Sequoia just bet $300M that your Nvidia inference contract is overpriced.

Etched closed a $300M Series C at $10.3B valuation.

Sequoia's largest-ever Series C.

$1B in pre-orders from companies deploying inference at scale — not vaporware.

This isn't a chip story. It's a procurement event.

Etched built a transformer-specific ASIC that splits inference into two phases — prefill and decode — and optimizes each with custom silicon.

Low-voltage inference for the compute-heavy prefill stage.

Cluster-scale memory for the token-heavy decode stage.

The result: more tokens per dollar, more tokens per watt, lower latency.

Google is reportedly building the same thing internally with Frozen v2 for Gemini.

When the hyperscalers start designing their own inference silicon, the writing is on the wall for通用 GPU pricing.

Your cloud provider's inference bill is about to face competition it hasn't seen in three years.

$1B in pre-orders means real enterprise customers are already locking in alternative supply.

Audit your inference spend now.

If your AI projects run on reserved GPU clusters or spot instances, get quotes from alternative silicon before your next budget cycle.

The companies that locked in long-term Nvidia contracts at 2025 prices are about to feel the squeeze.

The ones that didn't are about to get leverage.