JPMorgan just picked SambaNova to run AI inside its own walls. Banks are done trusting the cloud with sensitive models.

SambaNova just closed $1B at an $11B valuation.

Five months ago, Intel tried to buy the company for $1.6B.

The real story isn't the money. It's the JPMorgan Chase deal.

JPMorgan selected SambaNova as its "inference-infrastructure partner." SN40L and SN50 systems will power secure, on-premises AI inference at the bank.

CEO Rodrigo Liang: "Banks of the caliber of JP Morgan are now building their own private, secure infrastructure to run inference on their most sensitive models."

This is the signal. Regulated industries are done renting GPU time from hyperscalers for workloads that touch customer data, trading models, and risk engines.

The cloud-first AI strategy just hit a wall called compliance.

Liang confirmed SambaNova targets three customer types: sovereign clouds, neoclouds, and enterprises building for their own use. Saudi Aramco and Japanese firms are already customers.

Intel is a backer and co-development partner. SoftBank is the first SN50 deployment partner. The chip ships second half 2026.

$1B goes to supply chain. Liang called it the key constraint on fulfilling orders over the next 12 months.

Audit your inference architecture now. If your most sensitive AI workloads run on shared cloud infrastructure, you're behind every bank that's already building private. The gap between cloud-dependent enterprises and those running their own inference is widening fast.