Qualcomm is in early talks to acquire Tenstorrent, the RISC-V AI chip company led by legendary chip designer Jim Keller, in a deal valued between $8 billion and $10 billion. The Information first reported the discussions in June, and the talks have resurfaced repeatedly since, with no agreement confirmed and both companies staying quiet.
The price would be a steep jump from Tenstorrent’s last publicly speculated valuation of roughly $3.2 billion at the end of last year, when the company sought $800 million in fresh funding. Qualcomm declined to comment on rumors. Tenstorrent did not respond to requests for comment.
Why Qualcomm wants out of its own shadow
Qualcomm built its business on smartphone chips, and that business is cyclical. The company has spent years pushing into data center processors and automotive silicon to reduce its dependence on handsets, with mixed results. An AI accelerator startup with a working product line would give it something it currently lacks: credibility in the training and inference market Nvidia dominates.
Qualcomm already sells AI200 and AI250 data center accelerators built on its Hexagon neural processing units, due to ship in 2026. It is also developing its own server CPUs, reportedly on the Arm instruction set. But the company’s data center story remains thin compared with Nvidia, AMD or even Broadcom, and its recent acquisition of RISC-V CPU designer Ventana Micro shows where it thinks the next opening is.
Tenstorrent would push that bet much further. The startup builds AI accelerators, developer workstations and enterprise clusters on the open-source RISC-V architecture, and licenses its chip designs and IP to others. Bernstein analyst Stacy Rasgon saw the strategic logic but flagged the risks: the valuation, and whether Keller would stay.
“While obtaining Jim Keller on the payroll would be a coup for any company, we would not plan on him staying for long as his typical behavior is to leave public companies behind in fairly short order once arriving,” Rasgon wrote.
The Keller factor
Keller is the reason this deal gets attention beyond its price tag. He led the teams behind AMD’s Zen architecture and Apple’s A-series iPhone chips, worked on Tesla’s autonomous driving silicon, and joined Tenstorrent in late 2020 after leaving Intel. He is one of the few chip architects whose name alone moves acquisition rumors.
He is also a repeat leaver. Keller’s stints at Apple, Tesla, AMD and Intel each lasted a few years before he moved on. Any Qualcomm deal would be buying his current company, not a long-term commitment from the man himself, and integration would fall to Qualcomm’s existing engineering leadership.
| Company | Keller’s role | Notable work |
|---|---|---|
| AMD | Lead microarchitect | Zen CPU architecture |
| Apple | Senior director | A4 through A7 iPhone chips |
| Tesla | VP of engineering | Autopilot hardware |
| Intel | Senior VP | x86 and AI programs |
| Tenstorrent | CEO since 2021 | RISC-V AI accelerators |
The Arm hedge
There is a second motive buried in the architecture. Qualcomm’s relationship with Arm has been strained for years, culminating in a lawsuit over Qualcomm’s Nuvia acquisition that Qualcomm won decisively in late 2024. Buying Tenstorrent would deepen Qualcomm’s RISC-V exposure and give it a hedge if its dependence on Arm licensing ever becomes a problem again.
Qualcomm has been building RISC-V capability for some time. The Ventana acquisition in late 2025 brought data center-grade RISC-V CPU design in house. A Tenstorrent deal would add AI accelerators on the same architecture, plus chiplet and die-to-die interconnect IP picked up through Tenstorrent’s 2025 purchase of Blue Cheetah Analog Design.
The fit is not perfect. Qualcomm would end up with two AI accelerator lines and three data center CPU approaches, one Arm-based and two RISC-V-based. That is a lot of parallel bets for a company that has struggled to turn its data center ambitions into meaningful revenue. Bernstein made no changes to its Qualcomm model on the news, holding a Market-Perform rating while waiting for more clarity.
What Tenstorrent brings on its own
Tenstorrent has survived on a different model from most AI chip startups. Instead of burning capital on a single giant training cluster play, it sells PCIe cards, developer workstations and enterprise clusters, and licenses its RISC-V processor designs and AI hardware SDKs to companies that want to build their own silicon. Licensing revenue is steadier than hardware sales, and the IP business gives an acquirer something to integrate beyond a product line.
The company raised $693 million in a December 2024 Series D round backed by Bezos Expeditions, Samsung Securities and AFW Partners, valuing it above $2.6 billion at the time. In July 2025 it acquired Blue Cheetah Analog Design, whose die-to-die interconnect IP it had already been licensing, and it has worked with AI infrastructure software provider Moreh on a data center offering. Press reports in May 2026 had surfaced Tenstorrent as a takeover target for both Intel and Qualcomm before The Information’s June report.
For Qualcomm, the acquisition would rank among the most expensive in its history. The company’s largest deals, Atheros in 2011 and Nuvia in 2021, each transformed parts of its business: Atheros turned a modem supplier into a connectivity company, and Nuvia’s team became the core of Qualcomm’s Snapdragon PC chips. Tenstorrent would be a bet on the same scale.
What happens next
The talks remain early and the price could change or the discussions could fall apart entirely. It is also unclear whether the valuation would include performance-based milestone payments, a structure Qualcomm has used in past chip startup acquisitions to bridge the gap between a startup’s asking price and its proven revenue.
What is clear is the direction of travel. Qualcomm’s Snapdragon Summit opens September 22 in Maui, where the company will unveil its next flagship mobile chips amid double-digit price hikes driven by rising costs. A $10 billion acquisition would signal that the company sees its future less in phones and more in the infrastructure layer underneath the AI buildout, where the margins and the growth actually are.
Shares of Qualcomm fell about 1 percent in extended trading when the talks were first reported, a muted reaction that reflects both the uncertainty and the market’s skepticism about whether Qualcomm can execute in data center silicon. The deal, if it comes, would need to change that perception as much as it changes the product lineup.