Amazon is spending $220 billion this year on AI infrastructure and still can't keep up—which means the cloud war just became a capacity war.
The Summary
- AWS hit $42.2 billion in Q2 revenue, its fastest growth in 18 quarters, with AI spending last quarter exceeding all prior quarters combined
- Amazon will spend $220 billion in 2026 on infrastructure and still won't have enough capacity to meet demand
- The PGA Tour is ditching physical server trucks for AWS-powered AI broadcasts, a signal that industries beyond tech are now cloud-native by default
- Jassy sees a path to $1 trillion in annual cloud revenue, but only if Amazon can build data centers faster than customers want to use them
The Signal
AWS revenue of $42.2 billion in Q2 represents the fastest quarterly growth Amazon's cloud division has seen since late 2022. More telling: AI platform spending in the last quarter alone exceeded the cumulative total of every prior quarter since AWS launched its AI services. That's not incremental adoption. That's a step function.
The constraint isn't demand. It's supply. Andy Jassy says Amazon will deploy $220 billion in capital expenditures this year, the vast majority aimed at data centers, GPUs, and networking infrastructure. Even at that scale, Amazon can't build fast enough. Customers are pre-committing to capacity that doesn't exist yet.
"Amazon will spend $220 billion this year and still won't have enough capacity to meet demand."
The PGA Tour example cuts through the abstraction. For years, the Tour traveled with semi-trucks full of servers to power live broadcasts at each event. Now they're running AI-powered broadcasts entirely on AWS, eliminating the physical footprint and gaining capabilities that weren't possible with on-premise hardware. This isn't a tech company optimizing its stack. It's a sports league treating cloud AI as baseline infrastructure.
That shift—industries outside tech assuming cloud and AI as the default, not the upgrade—is what Jassy is betting on for the $1 trillion cloud market. The math works if you believe two things:
- Every industry will run intelligence workloads, not just SaaS companies
- The cost of running those workloads stays high enough that cloud revenue grows faster than efficiency improvements shrink it
The implication here:
- For enterprises: If AWS is capacity-constrained despite $220 billion in capex, you're negotiating from a position of weakness. Lock in commitments now or get rationed later.
- For competitors: Google and Microsoft face the same build-versus-demand gap. Whoever solves it first—whether through better chip deals, faster permitting, or nuclear power agreements—wins the next five years.
- For the rest of us: When live sports broadcasts run on AI inference and a golf tournament doesn't need a truck full of servers, we've crossed into Web4. The infrastructure layer is solved. Now we build on top of it.
The Implication
The cloud wars are over. The capacity wars have begun. AWS, Google Cloud, and Azure aren't competing on features anymore. They're competing on who can turn dirt into data centers fastest. If you're building anything that needs serious compute, your strategy for the next 24 months is securing capacity commitments, not optimizing code. And if you're watching from the sidelines, understand this: the $1 trillion cloud Jassy is talking about isn't speculative. It's constrained only by how fast concrete can dry.