- Lead. Qualcomm confirmed the all-stock acquisition of AI software company Modular on 24 June, valued at roughly $3.92 billion, while simultaneously advancing separate talks to acquire chip designer Tenstorrent for $8–10 billion — a combined bet approaching $14 billion on breaking Nvidia’s dominance in AI computing.
- Fact. Tenstorrent, led by legendary chip architect Jim Keller, builds AI accelerators on RISC-V architecture paired with an open compiler, offering customers an alternative to Nvidia’s proprietary CUDA software stack.
- Stake. If the Tenstorrent deal closes, Qualcomm would own both the software layer (Modular) and an open-source hardware stack, positioning it as the most credible challenger to Nvidia in the AI data centre market.
The Modular acquisition
Qualcomm confirmed the Modular deal on 24 June in an all-stock transaction priced at approximately $3.92 billion based on its closing share price. Modular, which builds a modular AI compiler and inference platform, addresses one of the most persistent friction points in the industry: deploying AI models across heterogeneous hardware from different chip vendors. Its software allows developers to write once and run on multiple processor architectures without recompiling for each target — a direct challenge to Nvidia’s CUDA ecosystem, which locks workloads to Nvidia silicon.
The acquisition was announced separately from the ongoing Tenstorrent discussions, first reported by Reuters on 15 June. Qualcomm shares fell approximately 1% in extended trading following the Tenstorrent report, though analysts noted that the strategic rationale — owning both hardware and compiler stack — is coherent at the valuation levels in discussion.
The Tenstorrent talks
Tenstorrent, founded in 2016, develops AI accelerators built on RISC-V, the open-source instruction set that has attracted substantial investment as an alternative to Arm-licensed cores. Jim Keller, who previously designed chips at Apple and oversaw Tesla’s autonomous-driving silicon programme, leads the company. His involvement is one reason investors have assigned the startup a premium: Keller has an unusually consistent track record of building processors that outperform incumbents, and his focus at Tenstorrent has been on systems specifically optimised for training and running large AI models.
The talks value Tenstorrent at $8–10 billion, though the deal terms remain in flux and could include milestone-based payments. Neither Qualcomm nor Tenstorrent commented publicly. The FTC’s posture toward semiconductor M&A in 2026 is a variable; the regulator opened an antitrust probe into Arm earlier this year as consolidation in chip design accelerated.
Challenging Nvidia’s position
Nvidia currently commands the AI accelerator market through a combination of hardware performance, software ecosystem lock-in via CUDA, and supply chain advantages. Its H100 and H200 GPUs remain the reference architecture for most large-scale AI training runs, and its transition to Blackwell has not meaningfully disrupted demand. Qualcomm’s approach — combining Modular’s open compiler with Tenstorrent’s RISC-V accelerators — targets the software lock-in directly: if workloads can run efficiently on Qualcomm hardware without code changes, the switching cost disappears.
Whether the strategy succeeds depends heavily on performance benchmarks that have not yet been published at Tenstorrent’s current generation. But the scale of Qualcomm’s commitment signals that the company regards the AI data centre opportunity as large enough to justify acquisitions that together represent a substantial fraction of its annual revenue. The combination would make Qualcomm the first major challenger to Nvidia with a credible full-stack alternative — hardware, software, and developer tooling — in a single portfolio.