Five months ago, the rumor mill had Intel circling SambaNova with a check for $1.6 billion. A nice exit, sure, but founder Rodrigo Liang apparently had other ideas. Like raising $1 billion at an $11 billion valuation instead. That's not a pivot — that's a power move.
The Palo Alto-based AI chip company announced Wednesday it closed a Series E round that pushes its total funding past the $2.5 billion mark. The round was led by SoftBank Vision Fund 2, with participation from new and existing backers including BlackRock, Intel Capital, and Temasek. Yes, Intel Capital — the venture arm of the very company that was supposedly trying to buy the startup for a fraction of what those shares are now worth.
The timing raises eyebrows. AI chip funding has been frothy for two years running, but $1 billion rounds are still rare birds. Even rarer? Closing one just months after acquisition talks collapse. SambaNova isn't just surviving the AI chip wars — it's betting it can win them.
The chip that thinks different
SambaNova doesn't make the kind of GPUs that Nvidia has turned into the oil of the AI age. Its secret sauce is a reconfigurable dataflow architecture — a fancy way of saying the chip rewires itself on the fly to match the model it's running. That's unlike fixed-architecture GPUs that force software to bend to hardware; SambaNova's hardware bends to the software.
The company claims its SN40L chip delivers up to 20x performance gains on certain AI workloads compared to Nvidia's A100. Skeptics point out that benchmarks are cherry-picked and that Nvidia's software ecosystem is a moat that's hard to cross. But Liang told investors the new cash will go toward building a full-stack platform, not just silicon. Translation: they're coming for the whole pie, not just a slice.
“We're not a chip company. We're an AI systems company. The chip is just the engine — the full stack is the car.” — Rodrigo Liang, SambaNova CEO
That full-stack ambition includes a cloud service where customers can rent SambaNova's hardware and software by the hour. Think of it as AWS for non-Nvidia AI compute. Early adopters include the U.S. Department of Energy, Lawrence Livermore National Laboratory, and a handful of Fortune 500s. But enterprise traction remains thin outside of early-access hype.
Intel's what-could-have-been
Intel's rumored $1.6 billion offer would have been a steal — in hindsight, anyway. The $11 billion valuation is nearly 7x that number, and that's before SambaNova has even turned a profit. The startup's revenue is estimated at around $200 million annually, according to a person familiar with the matter. That's a 55x price-to-sales multiple. For context, Nvidia trades at about 25x sales. You do the math.
Intel Capital's participation in the round suggests Pat Gelsinger's crew isn't bitter about being spurned. Maybe they see SambaNova as a hedge against Nvidia's dominance. Or maybe they're just hoping to buy in cheaper now than they would have to acquire the company later. Either way, the message is clear: Intel wants a seat at the AI table, even if it's not the head of it.
The deal also underscores a shift in AI chip investing. SoftBank's Vision Fund 2 has been on a tear, pouring billions into AI infrastructure. Earlier this year, it led a $1.5 billion round for Graphcore, another Nvidia challenger. The Japanese conglomerate is betting that the AI boom will require more than one chip architect — and that the winner won't be decided by raw performance alone, but by who controls the stack.
The $11 billion question
Can SambaNova justify the valuation? The company's growth is real but not stratospheric. Revenues have doubled year over year, but from a low base. Its customer concentration is high — the DOE contract accounts for a significant chunk of revenue. And Nvidia isn't sitting still. The H100 GPU, successor to the A100, is already in production, and its software library, CUDA, remains the default for AI developers.
Still, the narrative is shifting. Big tech cloud providers — Amazon, Google, Microsoft — are all developing their own AI chips. They want alternatives to Nvidia, not because Nvidia's chips are bad, but because dependency on a single supplier is a risk no corporation wants to take. SambaNova, with its programmable architecture, could position itself as the flexible alternative that hyperscalers license for internal use. That's the bull case.
The bear case? SambaNova is a hardware company trying to act like a software platform, and that's a notoriously expensive pivot. Its cloud service is still nascent, and competing with AWS and Azure on compute rental is a price war few startups survive. Meanwhile, open-source alternatives like RISC-V-based chips are gaining momentum, threatening to commoditize the hardware layer entirely.
Liang, though, isn't losing sleep. “We're building the operating system for AI,” he told TechCrunch in an interview. “That's not a chip problem. That's a strategy problem. And we have the strategy.”
Closing thoughts
SambaNova's $1 billion raise is a bet that AI hardware will fragment — that the era of one chip to rule them all is ending. If Liang is right, his company will be one of a handful of players that own a slice of the AI compute pie. If he's wrong, the $11 billion valuation will look like the peak of a bubble.
But here's what sticks with me: Intel was ready to buy this company for $1.6 billion. Now its venture arm is investing at a valuation nearly 7x higher. Whether that's smart capital allocation or FOMO dressed up as strategy, we'll find out soon enough. The AI chip race is just getting started, and SambaNova just bought itself a bigger engine.
