Qualcomm considers acquiring AI-chip startup Tenstorrent for $8-10 billion

Qualcomm is considering acquiring AI chip startup Tenstorrent for $8-10 billion — according to Reuters citing sources from The Information publication.

Who is Tenstorrent

Tenstorrent is a Toronto-based startup developing processors for artificial intelligence tasks on the open RISC-V architecture. The company was founded by Jim Keller — one of the most renowned chip architects in the industry: he participated in developing processors for Apple, AMD, Tesla and served as chief architect of Intel's x86 division. Under his leadership, Tenstorrent is developing two product lines — CPU cores on RISC-V and specialized AI accelerators for neural network inference, competing with Nvidia, AMD, Groq and Cerebras.

• The estimated deal value is $8-10 billion dollars
• Source of information — Reuters citing The Information
• Founder of Tenstorrent — former architect at Apple, AMD, Tesla and Intel Jim Keller
• The company specializes in AI accelerators and CPUs on RISC-V architecture

Why Qualcomm is entering the AI accelerator segment

Qualcomm has for decades been associated almost exclusively with mobile Snapdragon processors, but in recent years the company has been actively pushing into the data center chips market for artificial intelligence. Previously, Qualcomm announced a line of cloud inference chips — AI200 and AI250 — directly aimed at competing with Nvidia and AMD in the market for large language model computing.

Acquiring Tenstorrent would give Qualcomm a ready team of engineers with unique expertise and its own AI accelerator architecture — instead of spending many years developing a solution from scratch. This accelerated path to a new market through acquisition of a specialized startup was previously used by Intel, which bought Israeli AI chip developer Habana Labs in 2019.

Consolidation of the AI chip market

Demand for computing power for neural network training and inference continues to grow faster than supply, and major technology companies are choosing different strategies to enter this market. Some hyperscalers — Google, Amazon, Microsoft — develop their own chips (TPU, Trainium, Maia) in-house. Others, including Qualcomm and previously Intel, prefer to acquire promising startups with ready-made teams and architecture.

For independent developers like Tenstorrent, such consolidation means growing interest from major players in their developments — especially when the startup is led by an architect with the reputation and track record of Jim Keller.

The AI accelerator segment itself is still largely controlled by Nvidia, whose chips for neural network training and inference dominate the data center market. AMD, Intel, Qualcomm and a whole range of startups — Groq, Cerebras, SambaNova — have for years been trying to take at least some of this market from Nvidia by offering cheaper or more energy-efficient alternatives for inference. Acquiring Tenstorrent would strengthen Qualcomm's position in this competitive battle, giving the company its own, rather than licensed, AI accelerator architecture.

What this means

If the Tenstorrent deal goes through, Qualcomm will gain one of the fastest paths to establish itself in the AI accelerator segment for data centers — a market that analysts call one of the main growth drivers for the semiconductor industry in the coming years. For the industry, it's another signal: giants increasingly prefer to buy ready expertise than to spend years independently developing a competitive architecture.