SambaNova Targets $10B Valuation With Intel’s CEO in the Chair

SambaNova Systems is in talks to raise as much as $1 billion at a $10 billion valuation, roughly five times what the San Jose, California-based AI chip maker commanded in its February Series E round. The round, first reported by The Information and confirmed publicly by Executive Chairman Lip-Bu Tan, would be the company’s largest to date. It also ties the startup’s near-term trajectory to a tightly bound set of stakeholders: the chief executive of Intel in the chairman’s seat, Vista Equity Partners carrying a $3.5 billion compute commitment, and SoftBank anchoring the first deployment of the startup’s new SN50 chip.

General Atlantic is in talks to lead the round, per The Information, and the target price would push the company past the $5.1 billion valuation it carried after its April 2021 Series D, the previous high water mark.

What changed since February

The Series E closed on February 24, raising more than $350 million. Vista Equity Partners and Cambium Capital led the round, with Intel Capital joining as a strategic participant. The implied valuation ranged from $2.2 billion to $4.8 billion depending on the source.

Behind the valuation jump sits a single line item that did not exist in February: a $3.5 billion compute commitment from Vista Equity Partners. The commitment backs Vector Core Compute (VC2), a new enterprise inference cloud Vista and Cambium launched in May. The cloud stitches together Intel Xeon CPUs for orchestration, SambaNova SN40 RDUs for the “decode” phase of inference, and NVIDIA Blackwell GPUs for the compute-heavy “prefill” phase. New investors in the Series E also included Battery Ventures, accounts advised by T. Rowe Price Associates, and Saudi First Data.

Together AI, which serves 400 trillion tokens per month, is VC2’s first commercial customer. Vista’s 90-plus portfolio companies, which collectively serve more than 2.5 million enterprise customers, have secured early access, giving SambaNova a demand floor to price its next round against.

The Intel CEO’s second chair

Tan is the thread that ties the three companies together. He was an early SambaNova investor and now serves as the chip maker’s executive chairman, a role he has held alongside his position running Intel since March 2025. The arrangement has put Intel in the unusual position of being both a customer and an equity holder in a company its CEO chairs.

Intel said Tan recused himself from negotiations on the SambaNova partnership, with Kevork Kechichian, the chipmaker’s EVP and general manager of the data center group, running the talks in his place. The arrangement, a company spokesman told The New York Times, “speak[s] to well-established governance procedures to ensure decisions are made in Intel’s best interests.” For SambaNova, the dual role is more than optics. Intel increased its stake during the Series E round and paired the investment with the multi-year Xeon-based collaboration to deliver cloud-scale AI inference built around Intel Xeon processors.

The SN50 chip and its HBM trade-off

SambaNova’s pitch to that market is built around the energy efficiency of its Reconfigurable Dataflow Unit, or RDU. The fifth-generation SN50 chip, announced in February and shipping later this year, runs on the same dataflow architecture as its predecessor but uses HBM2E memory, skipping the HBM3E that supplies NVIDIA’s top accelerators.

“The key innovation in our dataflow architecture is that we can overlap computation and communication, meaning that the increase in compute and network bandwidth means that we don’t need to have the latest and greatest specs in other areas,” SambaNova co-founder and CEO Rodrigo Liang told The Next Platform. “Especially HBM, where we don’t suffer from the congested supply chains on HBM3E and instead use HBM2E, while still outperforming all other chips when it comes to the speed/throughput frontier,” he added.

The numbers SambaNova published in February put the SN50 at 5X faster than competitive chips and a 3X lower total cost of ownership than GPU clusters for agentic AI workloads. A full rack of SN50s draws 30 kilowatts at peak, a footprint that drops to 15 to 20 kilowatts under power capping. Those figures put the SN50 inside the air-cooled envelope of standard enterprise data centers, where NVIDIA’s denser racks often require liquid cooling. The 15 to 20 kilowatt footprint under power capping sits inside that envelope. The 30 kilowatt peak figure sits above it, and SambaNova relies on power capping to keep the rack inside the air-cooled limits under load.

SoftBank and Aramco lead the customer list

The customer list answers the obvious question of who buys chips from a private company with a $10 billion paper valuation. The deals are not yet at the scale of NVIDIA’s hyperscaler contracts, but they are concentrated enough to give a private investor a credible path to revenue.

• First named customer. SoftBank Corp. will deploy the SN50 inside its next-generation AI data centers in Japan, with the chip serving sovereign and enterprise inference across Asia-Pacific.
• Saudi partnership. Aramco is working with SambaNova on hardware and software for large language model development, with the deal disclosed in February.
• First VC2 customer. Together AI, serving 400 trillion tokens per month, signed on as the first commercial customer of the Vista-backed VC2 inference cloud.
• Vista portfolio access. More than 90 Vista portfolio companies have early access to the VC2 cloud, serving more than 2.5 million enterprise customers and 750 million users in aggregate.

SambaNova has not disclosed the dollar value of any of the contracts.

SambaNova’s place in a $106 billion market

The size of the prize underwrites the $10 billion ask. The global AI inference market was valued at $106 billion in 2025 and is projected by multiple research houses to grow at a compound annual rate above 19 percent through the end of the decade, with NVIDIA extending its lead in tandem.

• NVIDIA’s share of the inference chip market has climbed to 74% from 66% over the past year, per The Information’s reporting.
• Groq, the inference-focused startup that emerged as SambaNova’s closest architectural rival, was acquired by NVIDIA for $20 billion in December 2025.
• Cerebras, the wafer-scale chip maker, priced its IPO on May 13, 2026, in what Forbes called the biggest tech IPO of the year.
• OpenAI announced a joint effort with Broadcom this month to co-design AI inference chips. Meta Platforms is also working on its own inference silicon.

That is the field SambaNova is bidding to lead, with one notable difference from the rivals named above. Cerebras is public, Groq is now part of NVIDIA, and the hyperscalers are building for themselves. SambaNova is the rare non-acquired, non-public alternative at scale, with a public anchor in SoftBank and a board chaired by the CEO of a Fortune 100 chipmaker.

Where the new money goes

The new round, if it closes at the reported $10 billion target, would roughly double SambaNova’s previous all-time high valuation of $5.1 billion, set after a $676 million Series D in April 2021 led by SoftBank’s Vision Fund 2. It would also bring the company’s cumulative capital raised since its 2017 founding above the $1.48 billion it had banked by February.

That capital funds the SN50 chip’s production ramp, scheduled for later this year, plus the scaling of SambaNova’s own inference cloud and the deepening of enterprise software integrations for agentic AI workloads. SambaNova has identified that last segment, agentic AI, as the wedge against NVIDIA. The deployment plan mirrors the product strategy. SambaNova is selling the SN50 as a complement to NVIDIA’s GPUs in enterprise inference stacks, with the SN50 handling workloads where NVIDIA’s racks are not the right fit.

Customers are asking for more choice and more efficient ways to scale AI. By combining Intel’s leadership in compute, networking, and memory with SambaNova’s full-stack AI systems and inference cloud platform, we are delivering a compelling option for organizations looking for GPU alternatives to deploy advanced AI at scale.

That framing, from Intel’s Kevork Kechichian, is the same one Liang has been making to investors for months. Both are counting on the inference market growing fast enough to fit two dominant accelerator architectures at once. NVIDIA’s 74 percent share, and its December 2025 purchase of Groq, suggest the incumbent is planning for that possibility too.

Frequently Asked Questions

How much is SambaNova trying to raise and at what valuation?

SambaNova is in talks to raise between $800 million and $1 billion at a post-money valuation of approximately $10 billion, per The Information. General Atlantic is in talks to lead the round, which would be the largest in SambaNova’s history at the top of that range.

When did SambaNova’s previous funding round close?

The Series E closed on February 24, 2026, raising more than $350 million with Vista Equity Partners and Cambium Capital leading and Intel Capital participating. The new $10 billion target represents roughly a fivefold jump from the implied valuation at the time of that round. New investors in that round also included Battery Ventures, accounts advised by T. Rowe Price Associates, and Saudi First Data.

What is the SN50 chip and when does it ship?

The SN50 is SambaNova’s fifth-generation AI chip, built on its Reconfigurable Dataflow Unit (RDU) architecture. SambaNova claims 5X the maximum speed of competitive chips and a 3X lower total cost of ownership than GPU clusters for agentic AI workloads, with the SN50 set to begin shipping to customers later this year and SoftBank Corp. as the first named customer.

What is the $3.5 billion Vista compute commitment?

Vista Equity Partners and Cambium Capital launched Vector Core Compute (VC2) in May 2026, an enterprise inference cloud that combines Intel Xeon CPUs, SambaNova RDUs, and NVIDIA Blackwell GPUs. The launch is backed by a $3.5 billion compute commitment to SambaNova. Vista’s 90-plus portfolio companies, which together serve more than 2.5 million enterprise customers, have secured early access to the platform.