Qualcomm is buying its way into the one part of the AI stack it has never owned: the software. On June 24, the San Diego chipmaker announced it would acquire Modular Inc., a four-year-old startup whose platform lets AI models run across competing silicon, in an all-stock deal valued at roughly $3.9 billion. The transaction will be paid in up to 19.2 million shares of Qualcomm common stock, priced against a closing share price of $204.13, and is expected to close in the second half of 2026.
For Qualcomm, the logic is less about Modular's revenue than about what its software unlocks. The company has spent four decades building chips — for smartphones, for the edge, and lately for the data center — but it has lacked the developer-facing layer that turns silicon into something engineers actually want to deploy. Modular supplies exactly that: an open, AI-native stack that runs models across CPUs, GPUs, NPUs, and custom ASICs "without re-writes for each accelerator," in the company's words.
A software answer to a hardware problem
The price tag represents a steep markup. Modular, founded in 2022 by Chris Lattner and Tim Davis — both veterans of Google's AI infrastructure teams — raised $250 million in September at a $1.6 billion valuation, bringing total funding to roughly $380 million. The Qualcomm deal values the company at more than twice that figure in under two years, a premium that reflects how badly large chipmakers want a credible alternative to Nvidia's CUDA, the proprietary software ecosystem that has helped lock developers into Nvidia hardware.
Lattner is among the most consequential systems engineers of his generation. He created the LLVM compiler infrastructure and the Clang and MLIR projects, designed Apple's Swift programming language, and built AI and core systems at Apple, Google, SiFive, and Tesla. At Modular he set out to build two things at once: Mojo, a programming language aimed at squeezing maximum performance from both CPUs and GPUs, and MAX, a framework for serving generative AI models efficiently across heterogeneous hardware. Davis, a repeat entrepreneur, helped build large parts of Google Brain's infrastructure, including TensorFlow, XLA, and TF Lite.
"Modular was founded on the belief that AI needs a more open and efficient software foundation that can span diverse hardware and deployment environments," Lattner said in a statement. "Joining Qualcomm gives us the scale and platform reach to accelerate that mission. Together, we can make AI development more accessible and performant for developers, strengthen portability across hardware, and help grow an open ecosystem that broadens participation and speeds innovation."
The data-center play behind the deal
The acquisition is the clearest signal yet of how Qualcomm intends to push into the data center, a market where Nvidia's dominance is anchored as much in software as in silicon. The pitch is portability: if developers can write once and deploy anywhere — across accelerators from any vendor — then the standard objection to non-Nvidia hardware ("we'd have to rewrite everything") loses much of its force.
"This acquisition marks a pivotal moment not just for Qualcomm, but for the AI industry," said Cristiano Amon, president and CEO of Qualcomm. "As agentic AI scales across data centers and edge environments, the industry is moving toward disaggregated, multi-vendor architectures that demand a more open and modern software foundation. We believe the future belongs to developer-friendly, horizontal platforms that can run across diverse compute environments and give customers real choice in how and where they deploy AI."
That framing matters because of the economics Qualcomm keeps returning to. As inference scales, the company argues, efficiency rather than raw capability becomes the binding constraint — performance-per-watt drives the cost of inference, and cost determines what scales. Better software that wrings more out of existing silicon helps both at the edge, where Qualcomm already has a strong franchise, and in the data center, where it is the newcomer.
Modular does not arrive in isolation. Qualcomm used its Investor Day to detail a new data-center processor roadmap and to raise its fiscal 2029 non-handset revenue target to $40 billion. And the company has reportedly been in talks to acquire Tenstorrent — the RISC-V AI chip startup led by Jim Keller — at a valuation in the $8 billion to $10 billion range. Pair Tenstorrent's silicon with Modular's hardware-agnostic inference layer, and Qualcomm would have both halves of a full-stack challenge to Nvidia. That deal remains unconfirmed.
What to watch
The strategic case is coherent; execution is the open question. Modular's value depends heavily on its open-source community and its credibility as a vendor-neutral layer — qualities that can be difficult to preserve inside a silicon company with obvious incentives to favor its own chips. Both Qualcomm and Modular have stressed that the platform will stay "industry-friendly" and "vendor-neutral," but developers will judge that by behavior, not press releases.
Two things will reveal whether the bet pays off. First, whether the Modular team, and Lattner in particular, stays and keeps shipping. Second, whether a Tenstorrent deal materializes to give Qualcomm the data-center silicon that Modular's software is meant to make portable. The transaction is subject to customary regulatory approvals; for now, Qualcomm has bought the software layer. The harder question is whether it can build the ecosystem around it.
"As agentic AI scales across data centers and edge environments, the industry is moving toward disaggregated, multi-vendor architectures that demand a more open and modern software foundation."- Cristiano Amon, President and CEO, Qualcomm