Sequoia just wrote a $300M check that says your Nvidia inference contract is overpriced

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

Doubled in seven months.

Led by Sequoia — their highest-valued Series C ever.

Here is what matters: Etched sells inference-only chips. Not training. Not general-purpose GPUs. Purpose-built silicon that claims 20x the throughput of an H100 for serving deployed models.

And they already have $1B in pre-orders.

SK Hynix invested alongside Sequoia, a16z, and Jane Street. That is not passive capital. That is a memory supplier betting its HBM roadmap on a post-GPU inference market.

Your Nvidia contract was priced assuming no credible alternative existed for inference workloads. That assumption just expired.

First rack shipments this summer. 400 employees. New 80,000 sq ft production facility in California and a Taiwan factory running.

The inference era is not coming. It is funded, staffed, and shipping.

Audit your inference spend today. If more than 40% of your GPU bill goes to serving models rather than training them, your architecture has a pricing problem that will not survive another procurement cycle.