The prevailing wisdom on Wall Street has been simple: AI workloads are moving to the cloud, and whoever owns the cloud wins. JPMorgan Chase just complicated that story in a meaningful way — and investors who understand what happened this week are better positioned than those who don’t.
SambaNova Systems, an AI chip startup that Intel reportedly tried to acquire for roughly $1.6 billion less than a year ago, raised $1 billion this week at an $11 billion valuation. The anchor announcement that drove that re-rating: JPMorgan Chase selected SambaNova as its inference-infrastructure partner, deploying the company’s systems to power secure, on-premises AI inference inside the bank’s own walls. That’s a deliberate choice not to route sensitive data through AWS, Azure, or Google Cloud — and it signals a real business need that the cloud-only narrative overlooks. JPMorgan holds client data, proprietary trading strategies, and financial records that it simply cannot expose to third-party servers. Banks, hospitals, government agencies, and defense contractors face the same constraint. For them, the cloud’s economics are attractive in theory but unacceptable in practice when data sovereignty is non-negotiable.
What this means for retail investors: the AI infrastructure trade is not a winner-take-all game for the hyperscalers. On-premises AI inference is quietly becoming a second front — and it favors companies that specialize in purpose-built, secure AI hardware. For public-market investors, this is worth tracking in a few ways. First, it reinforces the long thesis on semiconductor companies building inference-optimized chips — Nvidia (NVDA) remains the dominant name, but the competitive landscape is widening. Second, it adds a new angle to large-bank stocks like JPMorgan (JPM): institutions deploying proprietary AI infrastructure are building durable competitive moats in financial services. Third, agentic AI — the kind that runs continuously inside enterprise operations — consumes dramatically more compute than single-shot queries. Estimates from Gartner suggest agentic workflows use 5x to 30x more tokens per task. That’s a sustained demand driver that benefits both cloud and on-premises AI providers. The JPMorgan deal is a reminder that the AI boom has more than one road to monetization.
