What happened
Qualcomm said on June 24 that it agreed to buy Modular in an all-stock deal worth about $3.9 billion, or roughly 19.2 million Qualcomm shares. The announcement came at Qualcomm Investor Day, alongside a new data-center processor.
Modular brings the MAX inference engine, the Mojo programming language, its co-founders Chris Lattner and Tim Davis, and about 150 employees. The aim is to let models run across NVIDIA, AMD, and Qualcomm hardware. The deal is expected to close in the second half of 2026.
Why it matters
NVIDIA's real advantage is not only its chips but the CUDA software layer that ties workloads to its hardware. A credible portability runtime lowers switching costs and opens the door to cheaper inference on rival silicon.
The price also sets a roughly $4 billion market comparison for inference software portability, signaling how much buyers will pay to loosen that lock-in.
MintedBrain take
Portability layers rarely match native performance on day one, so treat vendor benchmarks with care. Still, teams planning multi-chip inference should track MAX and Mojo now rather than after the deal closes.
Discussion
Sign in to comment. Your account must be at least 1 day old.