Jim Keller has made Tenstorrent into a potential $8 billion to $10 billion takeover target for Qualcomm, according to a Tom's Hardware report citing The Information.
The reported talks are not a deal. Tom's Hardware said discussions are ongoing and that there is no guarantee Qualcomm and Tenstorrent will reach an agreement. Reuters, in a MarketScreener item, said it could not independently verify The Information's report. That distinction matters because the reported price is large enough to change how Tenstorrent is viewed: not as another AI accelerator hopeful raising around a technical roadmap, but as one of the more expensive strategic targets in Qualcomm's history.
Keller is the reason this story has weight beyond the number. Tenstorrent is now led by Keller, who became CEO, according to Data Center Dynamics. He previously held senior chip roles at Intel, AMD, Apple, and Tesla, per the same report. Tenstorrent's own site describes its work as AI graph processors, high-performance RISC-V CPUs, configurable chiplets, and a software stack.
That combination is what makes the reported Qualcomm interest plausible and hard to price. Tenstorrent is not just selling an AI accelerator card. Tenstorrent says it builds and sells AI computers using Tensix cores, licenses AI and RISC-V intellectual property, and uses open-source software stacks as part of the strategy. In a December 2024 funding announcement, Tenstorrent said it had closed more than $693 million in Series D funding at a $2 billion pre-money valuation, led by Samsung Securities and AFW Partners, with participation from XTX Markets, Export Development Canada, Healthcare of Ontario Pension Plan, LG Technology Ventures, Hyundai Motor Group, Fidelity Management and Research Company, Baillie Gifford, Bezos Expeditions, and others.
A reported $8 billion to $10 billion acquisition range would therefore represent a step-change from Tenstorrent's last publicly disclosed valuation. The structure of the reported Qualcomm price is not established, including whether a deal would be cash, stock, mixed consideration, or tied to performance milestones.
Qualcomm already has pieces of this roadmap
Qualcomm does not need Tenstorrent because Qualcomm lacks an AI story. Qualcomm unveiled AI200 and AI250, rack-scale data-center inference products, on October 28, 2025. Qualcomm said AI200 is expected to be commercially available in 2026, AI250 in 2027, and both are part of an annual data-center AI inference roadmap.
Qualcomm has also been buying around the data-center stack. Qualcomm completed its $1.4 billion acquisition of NUVIA in March 2021, using the deal to accelerate internally designed CPUs that later became central to Qualcomm's higher-performance compute roadmap. In December 2025, Qualcomm acquired Ventana Micro Systems, saying Ventana's RISC-V expertise would complement Qualcomm's Oryon CPU work and broaden Qualcomm's RISC-V capabilities.
That is the tension in the reported Tenstorrent approach. If Qualcomm is buying finished product overlap, the price looks demanding. If Qualcomm is buying Keller's team, RISC-V AI architecture, compiler work, and the option to compress several years of internal development, the comparison moves closer to NUVIA than to a simple product acquisition.
Tom's Hardware made that point directly: Qualcomm already has AI acceleration IP and CPU IP, so the cleaner explanation is talent and future architecture. Qualcomm has used acquisitions this way before. NUVIA gave Qualcomm not only CPU IP, but a team led by former Apple CPU architect Gerard Williams III. Tenstorrent would be a larger, riskier version of that play, spanning AI accelerators, CPUs, interconnect, compilers, and systems engineering.
The inference market is rewarding credible alternatives
The reported talks land in a market where inference has become the financing story for AI infrastructure. Groq said in September 2025 that it raised $750 million at a $6.9 billion post-money valuation. d-Matrix said in November 2025 that it raised $275 million at a $2 billion valuation. SambaNova announced more than $350 million in strategic Series E financing in February 2026 alongside its SN50 chip and an Intel collaboration.
Those rounds are not perfect comparables. Each vendor is selling a different mix of silicon, systems, cloud access, software, and customer economics. But they frame why Tenstorrent could command strategic interest even before Tenstorrent has the scale of established GPU suppliers. Production inference turns hardware into a unit-economics problem: latency, throughput, memory movement, power draw, software maturity, and cost per token matter as much as peak benchmark claims.
Tenstorrent has also shown the rough edges that come with building a new accelerator stack. In February 2026, Tom's Hardware reported that Tenstorrent was reducing Blackhole p150 cards from 140 to 120 Tensix cores through firmware and future shipments, with Tenstorrent saying existing users should expect a 1 percent to 2 percent performance drop. That is not fatal for an early hardware platform, but it is a reminder that a $10 billion price would be underwriting execution risk, not just a slide deck.
What Qualcomm would really be buying
Qualcomm's AI200 and AI250 roadmap says Qualcomm wants to compete in data-center inference. Ventana says Qualcomm wants RISC-V expertise inside its CPU engineering base. Tenstorrent sits at the intersection of those two moves.
The open question is whether Tenstorrent becomes a standalone architecture bet inside Qualcomm or another pool of engineering talent folded into Qualcomm's existing roadmap. The answer would determine whether Keller's open-source, RISC-V-centered strategy survives a Qualcomm acquisition or gets absorbed into a broader Qualcomm data-center platform built around its existing CPU and accelerator work.
For Tenstorrent, the strategic logic is clear even if the reported price is not. Tenstorrent has spent years arguing that AI infrastructure needs alternatives to GPU-centric systems and proprietary software stacks. Qualcomm is one of the few buyers with the balance sheet, chip design organization, and customer relationships to turn that argument into a full data-center product line. The cost of doing that, if The Information's reported range is right, may be as much as $10 billion before Qualcomm proves Tenstorrent's technology can scale inside Qualcomm's own roadmap.