Digitimes flags $700B capex

Cloud infrastructure is the story here — not chatbots, not flashy demos, but the physical buildout underneath them. The big change last week is that Alphabet, Amazon, Meta, and Microsoft all came out of earnings still spending like the AI boom is getting bigger, not smaller. Put together, their 2026 capital expenditure plans now point to roughly $725 billion. That is an enormous number, but the real point is simpler: the biggest companies in tech have decided the bottleneck is still compute, and they are buying their way through it. (bloomberg.com) ### Why does this number matter? Capex is the money these companies spend on long-lived stuff — data centers, servers, networking gear, power systems, land, and all the other hardware needed to run AI at scale. When that number jumps, it usually means management thinks demand is durable enough to justify years of heavy spending. In this case, the sign(bloomberg.com)o $145 billion, Microsoft pointed to roughly $190 billion for calendar 2026, and Amazon kept its figure around $200 billion. (abc.xyz) ### What forced their hand? Cloud demand did. Amazon said AWS revenue rose 28% year over year to $37.6 billion in the March quarter, its fastest growth rate in 15 quarters. Alphabet said Google Cloud topped $20 billion and grew 63%. Microsoft said its AI business alone passed a $37 billion annual revenue run rate, up 123% from a year earlier. Those are not “maybe someday” numbers — they say customers are already paying up for AI capacity right now. (ir.aboutamazon.com) ### Is this mostly about GPUs? GPUs are the headline item, but the spend is broader than that. A frontier AI data center is basically a power plant, a networking project, and a real-estate project glued to a chip order. The catch is that every layer is constrained at once — chips, high-bandwidth memory, racks, cooling, transformers, and e(ir.aboutamazon.com)at is why the spending race looks defensive as much as offensive. (bloomberg.com) ### Why does Google stand out? Because Google’s cloud growth was the biggest surprise. A 63% jump to more than $20 billion reset the conversation from “who has AI products?” to “who is actually converting AI demand into cloud revenue fastest?” That does not mean Google has won the cloud market overall. But it does mean investors now have proof that AI workloads can move the revenue needle hard enough to justify even bigger infrastructure budgets. (cnbc.com) ### Where does that leave startups? In a tougher spot at the infrastructure layer. If the biggest four companies can spend hundreds of billions, absorb lower margins, and cross-subsidize AI from ads, software, and e-commerce, then pure-play challengers have less room to compete on raw compute. Basically, the stack is splitting. Hyperscalers own the expensive base layer, while smaller companies have t(cnbc.com)ps. This is less glamorous than “build your own foundation model,” but it is probably the more realistic lane. (bloomberg.com) ### Is there any catch? Yes — returns still have to show up. Investors are tolerating this because revenue is accelerating alongside the spend. If that link weakens, the mood changes fast. But right now the opposite is happening: cloud growth, AI run rates, and backlog are giving management teams cover to keep building. The market is still arguing over whether this is visionary or excessive. The companies themselves are acting like delay is the bigger risk. (ir.aboutamazon.com) ### Bottom line? The news is not just that Big Tech is spending more. It is that all four major U.S. hyperscalers are spending as if AI infrastructure is becoming the new core utility of the industry. Once that happens, scale stops being an advantage and starts looking like the whole game. (bloomberg.com)