Hyperscalers project $660–690B spend

Microsoft, Amazon, Alphabet, Meta and Oracle are on track to spend between $660 billion and $690 billion on capital expenditure in 2026, according to a February 12 analysis from Futurum. The figure has circulated widely among investors and operators because it puts a single number on the AI buildout now underway across cloud, networking and data-center campuses. The same discussion has centered on a physical limit rather than a software one: power. Microsoft Chief Executive Satya Nadella said in a late-2025 interview that the industry’s “biggest issue” was power and the ability to finish builds fast enough near available electricity supply. ### Which companies are included in the $660 billion to $690 billion figure? Futurum said the estimate covers Microsoft, Alphabet, Amazon, Meta and Oracle. The firm said Amazon accounts for about $200 billion of projected 2026 capex, Alphabet for $175 billion to $185 billion, Meta for $115 billion to $135 billion, Microsoft for $120 billion or more, and Oracle for $50 billion. Those figures add up to the $660 billion to $690 billion range cited in the analysis. The February 12 note said the “vast majority” of that spending is directed at AI compute, data centers and networking. Futurum also said the combined annual commitment had risen from about $380 billion in 2025 to the projected 2026 range in roughly 18 months. ### Why has power become the central constraint? (futurumgroup.com) Satya Nadella said on the Bg2 Pod that the industry’s “biggest issue” was not a compute glut but power and the ability to get data-center builds done close to available electricity supply. Data Center Dynamics, which reported the remarks, said Nadella described AI GPUs sitting in inventory because Microsoft did not yet have enough power to install them. (futurumgroup.com) Goldman Sachs said on May 1 that the scale of the AI buildout is highly sensitive to “power, labor, and equipment bottlenecks.” The bank said those constraints can elongate deployment timelines and, in stress scenarios, feed back into demand-side concerns. Goldman described its work as a scenario framework rather than a forecast, but said market commentary has already featured estimates of $4 trillion to $8 trillion of total capital investment over the next five years. (datacenterdynamics.com) ### Where does the 5 gigawatts versus 12 to 16 gigawatts claim come from? Archdesk, citing Sightline Climate and BloombergNEF, said roughly 16 gigawatts slated to complete in 2026 were being tracked across about 140 projects, but only about 5 gigawatts were actively under construction at the time of that analysis. The report said 30% to 50% of that pipeline could slip into 2027 or later. (goldmansachs.com) Tech-Insider, citing Bloomberg reporting and industry trackers, similarly said only about 5 gigawatts of roughly 12 gigawatts of U.S. AI data-center capacity announced for 2026 was under active construction. It attributed the gap to grid bottlenecks, electrical equipment shortages and project delays. (archdesk.com) ### What is behind the Microsoft backlog number? Microsoft said on May 1 that commercial remaining performance obligation rose 99% to $627 billion in its fiscal third quarter. That measure is broader than Azure alone and includes contracted revenue to be recognized later across Microsoft’s commercial business. (tech-insider.org) CNBC reported after Microsoft’s January 28 earnings that year-end commercial remaining performance obligation stood at $625 billion, helped by OpenAI’s $250 billion cloud commitment during the quarter. Separate industry commentary has described about $80 billion of Azure demand as effectively unfilled because capacity has not been built fast enough, but Microsoft has not framed that figure in its earnings release as a standalone Azure backlog metric. (microsoft.com) ### How do other forecasts compare? Goldman Sachs said on May 1 that its baseline model implies $765 billion in annual AI capex in 2026, rising to $1.6 trillion in annual capex by 2031. The bank said those figures include chips, data centers and power infrastructure, and depend heavily on assumptions about chip replacement cycles, data-center complexity and bottlenecks. (cnbc.com) Microsoft is expected to report fiscal fourth-quarter results in late July, while Amazon, Alphabet and Meta are due to report second-quarter results on their usual summer earnings cycle. Those updates will give investors the next company-by-company read on whether the projected 2026 spending totals and power-constrained build schedules are holding. (goldmansachs.com)