Intel backs SambaNova’s $350 million raise as SN50 inference chip lines up SoftBank

February 25, 2026
Intel backs SambaNova’s $350 million raise as SN50 inference chip lines up SoftBank

SAN JOSE, Calif., February 24, 2026, 23:52 (PST)

  • SambaNova pulled in $350 million in fresh funding, with Vista Equity Partners and Cambium Capital taking the lead. Intel Capital also joined the round.
  • The startup inked a multi-year AI inference deal with Intel, putting SoftBank in place as the inaugural SN50 customer.
  • SambaNova is touting its SN50 chip and rack systems, aiming to undercut GPU-focused inference setups on cost.

SambaNova Systems, the AI chip company, has pulled in $350 million from investors, with Vista Equity Partners and Cambium Capital leading the round. The company also announced a multi-year deal with Intel to push AI inference systems to customers. According to SambaNova, SoftBank Corp is set to be the first to roll out the new SN50 chip inside its Japanese AI data centres.

The deal arrives as firms move away from pouring resources into training massive AI models, focusing instead on deploying them in production—a phase called inference, where specialized chips spit out answers and handle instant decisions. Suddenly, power consumption, latency, and expense are boardroom concerns for cloud vendors and major corporates alike.

Intel is once again pushing to keep its footing in data center infrastructure, even as customers increasingly diversify beyond just one vendor’s GPU-centric offerings. SambaNova is eyeing the chance to cash in on the hype around “agents”—a term that’s everywhere right now—by turning that chatter into actual sales and hardware out the door.

Vista’s involvement is notable—traditionally, the private equity shop has focused on enterprise software over silicon. SambaNova, for its part, plans to put the funds toward ramping up SN50 output, building out its SambaCloud platform, and tightening integrations with enterprise software.

Intel outlined a partnership centered on Xeon infrastructure, targeting AI-native companies, model vendors, large enterprises, and government agencies. According to the company, this deal fits alongside Intel’s current data-center GPU initiatives and leaves its overarching AI strategy intact.

SambaNova’s new SN50 is pitched for “agentic” AI—think systems that deconstruct a request into several steps, firing off numerous model calls and tool queries. Here, latency matters; even a one-second delay can ruin the user experience. The company, in a blog post, claims its Reconfigurable Dataflow Unit (RDU) architecture tackles data movement and memory issues head-on. SambaNova lined up SN50’s agent-oriented inference performance against Nvidia’s Blackwell B200. Sambanova

“The real race,” said Rodrigo Liang, co-founder and CEO of SambaNova, is building out data centres that can run agents with instant answers, while keeping costs in check to make AI pay. Intel’s data-centre head, Kevork Kechichian, pointed out that customers want greater choice and a way to roll out AI at scale without relying solely on GPUs. SambaNova, in a statement, added that its SN50 will ship later this year. Sambanova

SoftBank’s Hironobu Tamba said the company is putting together an “AI inference fabric” in Japan, aiming for both speed and the kind of “sovereignty” that comes from running everything within its own data centers. According to SambaNova, the rollout targets enterprise and sovereign clients across Asia-Pacific, backing both open-source and proprietary models.

Peter Rutten, research vice-president at IDC, said the SN50 “changes the tokenomics”—that is, the economics behind producing AI outputs—by offering high throughput while sticking to air-cooled power limits. Vista Capital partner Monti Saroya sees the startup ready for agentic systems that juggle several models and handle requests almost instantly.

Still, the core of the pitch is all about performance and cost—claims customers are likely to want to test themselves, in their own setups. Software support remains a big question in a market accustomed to Nvidia’s tools and developer base. Delays in hardware schedules aren’t unusual. And inference buyers, once their production workloads are locked to a given stack, typically don’t rush big changes.

Intel’s funding and the new partnership follow stalled acquisition talks with SambaNova, according to Reuters. CEO Lip-Bu Tan—who holds the executive chairman role at SambaNova—had earlier floated a $1.6 billion buyout of the startup, debt included.