SambaNova hits $11bn valuation as Nvidia-alternative demand surges

SambaNova has secured new funding that values the company at $11 billion, according to Bloomberg, capping a remarkable stretch where the AI chip startup has roughly quintupled its worth in just months. That figure edges up from the approximately $10 billion valuation reported when the round first took shape in late June.

The speed of this rerating tells the story as much as the final number. A company worth around $2 billion earlier this year now carries a paper valuation more than five times higher, a surge fueled by investors scrambling to back any credible Nvidia alternative. That same appetite has lifted rival bets, from Qualcomm’s move on Modular to a cluster of inference-focused startups.

SambaNova builds chips and systems engineered to run large AI models efficiently, using an architecture it calls a Reconfigurable Dataflow Unit rather than the graphics-derived design Nvidia popularized. Its pitch to customers centers on lower cost and lower power for the inference workloads that now consume the bulk of AI spending.

Money has followed that pitch. The company has raised close to $1.5 billion over its lifetime from a roster that includes SoftBank’s Vision Fund, Vista Equity Partners, Intel, GV, BlackRock, and Temasek.

The broader AI chip market is where the enthusiasm originates. As spending shifts from training models to running them, inference cost has become the metric that determines whether an AI product turns a profit, and buyers want more than one company setting the price.

SambaNova’s most recent prior round, a $350 million raise, closed only in February, making the valuation leap since then all the sharper. Rounds that once came a year or more apart are now landing within months as compute demand outstrips supply.

An unusual figure sits atop the company. SambaNova’s chairman is Lip-Bu Tan, who also serves as chief executive of Intel, a tie that links the startup to one of the incumbents it is partly trying to bypass.

That relationship has already produced hardware. Intel and Foxconn have joined SambaNova to build rackscale AI infrastructure, pairing its dataflow chips with the manufacturing muscle needed to ship systems at volume.

The backdrop to all of this is a market desperate for options. Nvidia still dominates AI training and much of inference, and its customers, cloud providers and model labs alike, have every incentive to fund a second source.

That explains why capital keeps flowing to challengers even before they prove they can take meaningful share. European entrants such as Fractile have drawn similar interest on the same thesis: inference at scale is too large and too expensive to leave to one supplier.

Whether SambaNova can convert a soaring valuation into durable revenue is the open question. Founded roughly nine years ago, it has cycled through pitches from training to enterprise inference as the market shifted beneath it.

It is not short of company at the front of that queue. A cluster of specialist firms, including Cerebras and Groq, is chasing the same opening, each arguing that its particular design suits inference economics better than a general-purpose GPU.

A valuation like this also raises the question of an exit. Numbers in the double-digit billions tend to point toward a public listing, and an $11 billion private tag is the sort of milestone that usually sharpens talk of an IPO.

Neither SambaNova nor its investors have detailed the exact size of the new round or how the proceeds will be spent. What the $11 billion tag makes clear is that, for now, the money is betting on the alternatives well ahead of the results.