The Trial and the Bill
Qualcomm acquired Modular for $3.9 billion to build GPU alternatives to Nvidia's CUDA. Meanwhile, cloud infrastructure capex is surging as AI labs bet heavily on compute layers. The market structure for AI is being built in silicon and software stacks, not just model weights.
The AI market structure is being built in silicon and software stacks — not just model weights.
Qualcomm's $3.9B Modular Bet
Qualcomm's acquisition of Modular (announced at $3.9 billion) is a direct move to build competitive GPU programming alternatives to Nvidia's CUDA ecosystem. Modular's Mojo language enables GPU programming with performance competitive to CUDA, reducing vendor lock-in and lowering the cost of deployment for organizations building on non-Nvidia silicon.
This signals that the AI infrastructure wars aren't won by models anymore — they're won by who owns the stack beneath the model. Qualcomm is betting they can own the inference layer and edge deployment. It's a structural play, not a model play.
Capex is the New Bottleneck
According to Bridgewater Associates analysis, Microsoft, Amazon, Alphabet, and Meta collectively are expected to invest approximately $650 billion in AI infrastructure during 2026 — up sharply from 2025. Memory and storage component prices are surging due to demand, creating supply-side constraints that are pushing capex higher across the board.
For builders: capex budgets are being exhausted faster than expected. Data center queues are extending. GPU availability is becoming the scarce resource, not model capability. Teams that locked in compute capacity early in 2026 have optionality. Teams that didn't are queuing.
What This Means
The frontier model releases (Fable 5, Sonnet 5, GPT-5.6) are happening inside a constrained infrastructure environment. The labs can ship new models faster than teams can deploy them. The bottleneck isn't model capability anymore — it's execution infrastructure.
The Founder's Take
Qualcomm is buying their way into the AI wars by acquiring the layer below the models. The capex bill is accelerating because the demand curve is steeper than supply. The labs are shipping faster, but the infrastructure can't keep up. The team that wins won't be the one with the best model — it'll be the one with the infrastructure to deploy it at scale.
— Megan
Get next week's note in your inbox.
🧠 Learn about AI from the frontlines. Follow us for AI updates in 10 minutes or less. Subscribe to Field Notes from the AI Frontier to stay ahead of AI news and updates. A founder's take as we consolidate the latest turning points and pivotal news from the top voices in AI. Learn how to build with AI and how to adapt.
🚀 Follow the future of AI with us.