Big Tech capex hits $725B

Data centers are the story here — not chatbots, not demos, not vibes. On April 29, Microsoft, Alphabet, Meta, and Amazon all gave investors fresh numbers that made one thing plain: the AI buildout is getting bigger, faster, and more expensive. Add their updated 2026 plans together and you land around $725 billion in capital spending. That is the real news. The AI race is no longer mostly about models. It is about who can afford to pour concrete, buy chips, wire power, and keep enough servers running to meet demand. (cnbc.com) ### Why did this jump now? Because the big cloud companies just reset expectations in the same 24-hour stretch. Microsoft said it expects roughly $190 billion of capex in calendar 2026, including about $25 billion from higher component pricing. Alphabet raised full-year guidance to $180 billion to $190 billion. Meta lifted its range to $125 billi(cnbc.com)ing plans still get to that roughly $725 billion combined figure. (microsoft.com) ### What are they actually buying? Mostly the boring, brutal stuff. GPUs and CPUs. Networking gear. Racks. Cooling. Land. Buildings. Power hookups. Microsoft’s finance chief said two-thirds of its recent capex was going into short-lived assets like GPUs and CPUs, which tells you this is not just a slow real-estate cycle. A lot of the money is going straight into compute that s(microsoft.com)road and more like stocking a warehouse with perishable inventory. (theglobeandmail.com) ### Why does component pricing matter so much? Because even the richest buyers are still supply constrained. Microsoft explicitly baked about $25 billion of higher component pricing into its 2026 capex outlook. Sundar Pichai also said Google is compute constrained in the near term. Basically, the companies are not only (theglobeandmail.com)t of it is inflation inside the AI supply chain. (microsoft.com) ### Why is Wall Street suddenly talking about $1 trillion? Because the 2026 numbers moved first. CNBC’s roundup of the earnings calls says both Evercore and Bank of America now see 2027 AI capex topping $1 trillion, with 2026 estimates rising into roughly the $800 billion to $900 billion range. That is the important shift. Last year, the debate was whether hyperscalers were ov(microsoft.com)nd for AI services and cloud capacity. (cnbc.com) ### Is this all about Nvidia? Not really — even if Nvidia still captures a huge slice. The bottleneck is moving downstream into the data center itself: power delivery, cooling, networking, and how fast operators can turn chips into usable compute. In other words, buying accelerators is only step one. Getting them installed and fed with electricit(cnbc.com)ction, and infrastructure vendors. (forbes.com) ### So what is the catch? Returns. Investors tolerated giant AI budgets when the numbers were smaller and the upside felt open-ended. At this scale, every quarter becomes a proof test. Cloud growth has to stay strong. AI products have to generate real revenue. Margins have to survive. If demand holds, these companies look early and aggressive. If demand cools, they will be sitting on some of the most expensive infrastructure bets in corporate history. (cnbc.com) ### What matters next? Watch two things — utilization and pricing. If the new capacity fills quickly, this spending cycle keeps going and may even accelerate into 2027. If customers start pushing back on AI costs, the story changes from expansion to efficiency. The bottom line is simple: Big Tech just told the market that AI is now an infrastructure arms race, and the bill is arriving faster than almost anyone expected. (cnbc.com)