Morgan Stanley sees $805B AI capex
- Morgan Stanley lifted its forecast for Amazon, Alphabet, Meta, Microsoft, and Oracle AI capex to $805 billion in 2026, with $1.1 trillion in 2027. - The key detail is what breaks first: not chips, but power access — substations, transformers, grid hookups, and the heavy site work. - That shifts the AI trade toward utilities, power gear, and civil works as software money turns into physical infrastructure.
AI capex is starting to look less like a software story and more like a construction story. That’s the real point buried inside Morgan Stanley’s new numbers. The bank now thinks Amazon, Alphabet, Meta, Microsoft, and Oracle will pour $805 billion into capex in 2026 and $1.1 trillion in 2027 — a scale that pushes AI spending past old dot-com peaks and into something more physical, slower, and harder to fake. (msn.com) ### Why are these numbers such a big deal? Because they are not normal revisions. Morgan Stanley has been ratcheting estimates higher fast enough that the 2026-27 spending outlook is up by about $630 billion versus six months ago. The bank also thinks hyperscalers could account for roughly 40% of all Russell 1000 cash capex over 2026 through 2028 — more than $2 trillion by themselves. (finance.yahoo.com) ### Who is actually spending this money? The core group is the big cloud and platform companies — Amazon, Alphabet, Meta, Microsoft, and Oracle. These are the companies racing to build AI data centers, buy accelerators, lease capacity, and lock in enough infrastructure to train and serve models at industrial scale. The impo(finance.yahoo.com)t. (msn.com) ### So what’s the bottleneck now? Power. Not in the abstract — in the painfully literal sense. Data centers need grid access, substations, transformers, switchgear, transmission links, cooling systems, water handling, and prepared sites. Morgan Stanley’s own energy team has been warning that U.S. data-cente(msn.com)singly whether a site can actually get energized, not whether a company wants more GPUs. (morganstanley.com) ### Why does that change the AI story? Because physical infrastructure runs on long lead times. You can order chips and servers, but you cannot instantly create a transformer factory slot, a new substation, or a transmission upgrade. The catch is that AI demand is colliding with an electric grid that was already under(morganstanley.com)eneration, microgrids, batteries, gas, and even nuclear-adjacent solutions are suddenly part of the conversation. (morganstanley.com) ### Is this still bullish for tech? Yes — but with a different shape. Morgan Stanley still sees semiconductor suppliers as the clearest near-term winners, with 2026 sales revisions for that group up about 60%. But the spending wave is broadening outward. If the hard part of AI is now energizing and cooling the buildin(morganstanley.com)r a lot more. (finance.yahoo.com) ### What’s the risk? The risk is that spending is accelerating faster than revenue. Morgan Stanley has flagged that revenue revisions are lagging while free-cash-flow estimates are drifting lower. Basically, the hyperscalers are front-loading an enormous fixed-cost base before anyone can prove exactly how fast AI services w(finance.yahoo.com)ys, cost overruns, and any wobble in demand. (finance.yahoo.com) ### Why should a regular reader care? Because this is one of those moments when “digital” demand turns into very non-digital consequences. More electricity demand. More industrial financing. More pressure on local grids. More leverage for the companies that make the boring parts — cables, switchgear, cooling, backup power, and land-ready sites. That is the hidden translation mechanism from AI hype into the real economy. (morganstanley.com) The bottom line is simple. Morgan Stanley’s $805 billion call is not just a bigger number. It is a clue about where the next constraint sits. AI is moving from code into concrete. (msn.com)
