Seven months ago, SambaNova was in quiet discussions to be sold to Intel for about $1.6 billion, a price that represented a steep fall from its 2021 peak valuation. The company had been struggling to raise money and grow sales against Nvidia's dominance. Intel's CEO, who is also SambaNova's chairman, was reportedly close to a deal.
That deal fell apart. SambaNova instead raised $350 million in February and has now closed a further $1 billion at an $11 billion valuation, led by General Atlantic, with backing from BlackRock, the Qatar Investment Authority, T. Rowe Price, and others. Intel stayed involved, as a partner and investor rather than acquirer. The jump from a near-acquisition at $1.6 billion to a fundraise valuing the company at $11 billion, in roughly five months, reflects how fast enterprise priorities around AI have shifted.
What changed is the question of where AI runs, and who controls it.
For the past few years, most companies tried AI by sending data to cloud services run by Amazon, Microsoft, or Google. That is fast to get started and requires no hardware. The trade-off is that your data leaves your building. For a law firm, a hospital, an insurer, or a bank, that trade-off is increasingly uncomfortable. A 2026 Deloitte survey found that 55% of enterprises are avoiding at least some AI use cases entirely because of data security concerns with cloud providers.
SambaNova sells physical hardware, server racks that sit inside a company's own data center. The AI runs locally. The data never leaves. SambaNova says its systems can be installed and running in as few as 90 days, which is quick for on-site infrastructure.
The company's chip, the SN50, is built differently from the chips Nvidia sells. Nvidia's chips are excellent at training AI models, which is the initial, expensive process of building an AI from scratch. SambaNova's chips are designed specifically for inference, which is the ongoing process of using a trained model to answer questions or complete tasks in production. As companies move from experimenting with AI to running it continuously across real business processes, inference is where most of the cost and activity sits.
JPMorgan Chase has selected SambaNova as its AI infrastructure provider, deploying the company's hardware for secure, on-premises AI inside the bank. SambaNova's CEO described this as a signal to the entire banking sector that the time has come to stop depending completely on cloud services for sensitive AI work. That is a reasonable read: when JPMorgan moves, compliance-focused industries watch.
The broader picture is one of a market splitting in two. Cloud AI remains the right choice for many tasks: quick deployments, public-facing applications, and anything where data sensitivity is low. On-site hardware makes more sense when the data is regulated, the volume is high enough to make fixed costs worthwhile, or when regulators and auditors need proof that data never left the organization's control. Financial services, healthcare, insurance, and government are the obvious early adopters of the on-site model.
For businesses that process large amounts of sensitive client data, the JPMorgan decision is worth tracking. The question is not whether to adopt AI, but where it runs and who controls it. That decision, once made, is hard to reverse: changing from cloud to on-site or back requires redesigning internal systems, which is a major project.
SambaNova's CEO said the company is seriously considering a US IPO in 2027. Given the pace of funding in this space and the JPMorgan validation, that timeline looks credible. Whether the company's valuation holds through to a public listing depends on whether the enterprise on-site AI trend keeps accelerating, or whether cloud providers find ways to satisfy the data sovereignty concerns that are currently driving businesses toward hardware like SambaNova's.