Wealthpulse projects $630B–$750B capex
- Alphabet, Amazon, Meta, Microsoft and Oracle are being modeled to spend roughly $630 billion to $750 billion on 2026 capital expenditure. - CreditSights put the top-five total near $750 billion, while Bloomberg and Bridgewater estimated about $650 billion for four companies alone. - Power and memory are now the binding constraints on AI buildouts, not cash alone. (cnbc.com)
Artificial intelligence spending is no longer a software story alone. On 2026 forecasts, the biggest cloud companies are turning it into a construction, chips and power story. (know.creditsights.com) (www.bloomberg.com) CreditSights said Alphabet, Amazon, Meta, Microsoft and Oracle could reach about $750 billion of capital expenditure in 2026, up 67% from 2025. Bloomberg separately reported Alphabet, Amazon, Meta and Microsoft alone are on track for about $650 billion this year. (know.creditsights.com) (www.bloomberg.com) Those totals come from company guidance and analyst models released after early-2026 earnings. CNBC reported the four megacaps were expecting combined spending close to $700 billion after a round of raised capex plans. (www.cnbc.com) The underlying concept is simple: training and running large AI models requires huge clusters of graphics processors, networking gear and specialized memory, all housed in new data centers. The bill is not just for chips; it includes land, buildings, substations and long-term electricity supply. (www.cnbc.com) (www.datacenterfrontier.com) Analysts increasingly say roughly three-quarters of hyperscaler capex is now tied directly to AI infrastructure rather than traditional cloud expansion. MUFG put that AI share at about $450 billion out of more than $600 billion for the “big five” in 2026. (www.mufgamericas.com) (know.creditsights.com) The bottlenecks are physical. Data Center Frontier reported power availability has become the defining limit on AI data center growth, with campuses moving toward gigawatt scale and developers adding on-site generation. (www.datacenterfrontier.com) (www.bloomenergy.com) Even a single project can now require utility-scale electricity. Industry coverage in 2026 describes 100-megawatt to 300-megawatt AI facilities as the new planning unit, which is why siting has shifted toward places that can actually deliver power on time. (www.datacenters.com) (enkiai.com) Memory is another choke point. CNBC reported Micron said demand for AI memory had “far outpaced” industry supply, and Micron’s high-bandwidth memory output for 2026 was already committed. (www.cnbc.com) (finance.yahoo.com) That helps explain why these forecasts keep moving higher. Goldman Sachs said analyst estimates for 2026 hyperscaler capex had already risen to $527 billion by December 2025, and said supply bottlenecks or investor appetite were more likely constraints than balance-sheet capacity. (www.goldmansachs.com) Bridgewater added another market frame in February, estimating Alphabet, Amazon, Meta and Microsoft would invest about $650 billion in 2026 to scale AI infrastructure. Reuters said that figure was up from about $410 billion in 2025. (www.globalbankingandfinance.com) The thread running through all of it is that hyperscalers are no longer competing only on models and products. They are competing on transformers, substations, chip allocations and how fast they can turn land and power into working compute. (www.bloomberg.com) (www.datacenterfrontier.com)
