Big Tech capEx hits $715B

Datacenter spending is becoming the main event in Big Tech earnings. Not ad growth. Not devices. Not even model releases. In the last few days, Amazon, Alphabet, Meta, and Microsoft all made the same point in slightly different ways: they are going to spend staggering amounts of money in 2026 to build AI capacity, and they do not look ready to slow down. (microsoft.com) ### What actually changed? The big shift is that this stopped being a vague “AI investment” story and became a hard capex story with company-level numbers. Amazon is holding to about $200 billion in 2026 capital spending. Alphabet updated its range to $180 billion to $190 billion. Microsoft is pointing to about $190 billion. Meta raised its own range to $125 billion to (microsoft.com) figure circulating in some charts. (cnbc.com) ### Why are the numbers so huge? Because AI infrastructure is not just GPUs anymore. These companies are buying chips, networking gear, servers, land, power equipment, and entire datacenter campuses. Alphabet said Q1 capex was $35.7 billion, with the overwhelming majority going into technical infrastructure for AI opportunities across the company. That is the tell — this is core plumbing now, not a side project. (abc.xyz) ### Why now? Demand is outrunning supply. Sundar Pichai said Google is compute-constrained in the near term and that cloud revenue would have been higher if Google could meet all the demand. Microsoft said its AI business has already passed a $37 billion annual revenue run rate, up 123% year over year. Once customers are lining up faster than you can provision capacity, capex stops looking optional. (cnbc.com) ### Is this mostly training, or something else? More and more, it is about inference and production use — serving AI to real customers at scale. Google Cloud revenue jumped 63% past $20 billion in Q1 2026, and Microsoft Cloud revenue reached $54.5 billion, up 29%. Those are not science-project numbers. They suggest the spending is being pulled by live workloads in cloud and enterprise software, not just by internal model races. (cnbc.com) ### Why is Meta in this group? Meta is different because it is not primarily selling cloud capacity the way Amazon, Microsoft, and Google do. But it still needs enormous compute for ranking, recommendations, generative ads, Llama, and consumer AI products. It also keeps pushing deeper into custom hardware — including a March partnership with Arm to build CPUs for AI-optimized datacenters (cnbc.com) is architectural. (about.fb.com) ### What is the catch? Cash flow. CNBC noted that the four hyperscalers were already on track for close to $700 billion in 2026 spending back in February, and that the buildout would hit free cash flow hard. Amazon looked like the sharpest example, with analysts projecting negative free cash flow this year as spending ramps. Investors can tolerate giant bills when growth is obvious — but they still punish any sign that returns may take longer than promised. (cnbc.com) ### So is the $715 billion number wrong? Basically, it looks stale. It may have been a reasonable snapshot before the latest earnings updates. But after Meta’s raise and Alphabet’s updated range — and with Amazon and Microsoft both near $200 billion — the current combined figure is closer to $725 billion than $715 billion. That is the cleaner takeaway from the latest disclosures. (c([cnbc.com)at does this mean underneath the headline? It means AI’s bottleneck has moved from models to infrastructure. The winners are not just model labs. They are also chip vendors, networking suppliers, datacenter builders, utilities, and the cloud teams that can keep all this hardware running. When four companies are willing to spend roughly three-quarters of a trillion dollars in one yea(cnbc.com)come from. (cnbc.com) The bottom line is simple: this is no longer an AI hype cycle measured in demos. It is an industrial buildout measured in concrete, copper, power, and racks. And after last week’s earnings, the spending bar just moved higher. (microsoft.com)