Jim Keller wanted to be clear: Tenstorrent is not for sale.
Two weeks after The Information reported that Qualcomm was in advanced talks to acquire his RISC-V AI chip startup for as much as $10 billion, the veteran chip architect used a Tokyo media event on June 30 to shut the rumor down. Tenstorrent, he said, is not in acquisition discussions and intends to keep building on its own.
"We are focused on building our own business," Keller told reporters, adding that the company routinely talks with many partners across strategic areas but has no deal to sell itself. The remarks, delivered while Keller was in Japan discussing the company's IP-licensing and high-end datacenter roadmap, snuffed out one of the summer's louder chip-M&A stories almost as quickly as it had ignited.
What Sparked the Rumor
The Information kicked things off on June 15, reporting that Qualcomm had opened talks to acquire Tenstorrent in a deal valued between $8 billion and $10 billion. Neither company confirmed the report, and the discussions were described as early and uncertain. Still, the logic was easy to follow, and outlets from The Register to Tom's Hardware quickly amplified it.
Qualcomm has spent the past year assembling the pieces of a datacenter AI strategy built on open, royalty-free silicon. In December 2025 it acquired Ventana Micro Systems, a RISC-V CPU designer, and it separately spent $3.92 billion on Modular, an AI deployment and compiler platform meant to smooth the software path onto non-Nvidia hardware. Buying Tenstorrent would have slotted a proven RISC-V AI accelerator line on top of that stack, handing Qualcomm a credible data-center product and one of the industry's most recognizable chip designers in a single stroke.
Investors treated the report as plausible but not thesis-changing. Qualcomm shares slipped roughly 1% after the story broke, then normalized within days as the lack of confirmation sank in.
The Chip at the Center
Tenstorrent's appeal starts with its hardware. The company's Galaxy Blackhole platform packs 32 accelerators, each carrying 768 RISC-V cores, into a single 6U data-center enclosure. That is a dense, standards-based building block aimed squarely at the training and inference workloads that Nvidia currently dominates, and it is built entirely on an architecture that carries no licensing tax.
That last point is the strategic heart of the story. RISC-V is an open instruction-set architecture that anyone can implement without paying royalties, in contrast to Arm's licensed cores or Intel and AMD's proprietary x86. For hyperscalers spending tens of billions a year on AI silicon, an open ISA offers something they have wanted for a decade: the ability to design custom chips without being locked to a single vendor's roadmap or price list.
Keller, whose résumé includes foundational work at AMD, Apple, and Tesla, has bet Tenstorrent on exactly that opening. Selling now, before the RISC-V AI market matures, would arguably mean cashing out early on the thesis he has spent years building.
Why It Matters
The denial is a small event that sits on top of a very large shift. The AI compute market has been effectively a one-vendor economy, with Nvidia's CUDA software moat reinforcing its hardware lead. RISC-V is emerging as the most serious structural challenge to that arrangement, not because any single chip out-benchmarks an Nvidia GPU today, but because an open architecture lets buyers escape the lock-in altogether.
"The interest in RISC-V is really about control," one semiconductor analyst noted of the datacenter push. "Hyperscalers don't just want a cheaper chip. They want an architecture nobody else owns." That instinct explains why Qualcomm was willing to be linked to a $10 billion price tag, and why Keller can plausibly argue Tenstorrent is worth more independent than absorbed.
For Qualcomm, the episode underscores how far the company is reaching beyond its smartphone-modem core. Ventana, Modular, and the reported Tenstorrent overture all point the same direction: toward a full-stack RISC-V AI business that could challenge both Nvidia's hardware and the Arm-centric server ecosystem. Whether it buys that stack or builds it, the ambition is now unmistakable.
For the broader industry, Keller's refusal keeps one of the few independent, well-funded RISC-V accelerator makers out of a hyperscaler-adjacent giant's hands, at least for now. That preserves optionality for cloud buyers who would rather license Tenstorrent's IP than watch it disappear inside a competitor.
What to Watch
Three things will tell whether this stays a denied rumor or resurfaces. First, funding: Tenstorrent has leaned on strategic investors, and a fresh raise or valuation mark would signal how much runway Keller has to stay independent. Second, Qualcomm's next move; having bought Ventana and Modular, its next RISC-V acquisition or product announcement will reveal whether Tenstorrent was a one-off target or part of a pattern. Third, design wins: if a major hyperscaler publicly commits to Galaxy Blackhole or Tenstorrent IP, the "build our own business" strategy gets a lot more convincing, and a lot harder to buy out.
For now, the answer from Tokyo is no. In a chip market where everything is suddenly for sale at nine or ten figures, that alone is worth noting.
“We are focused on building our own business.”— Jim Keller, CEO, Tenstorrent