All four AI hyperscalers have more than tripled their capital expenditures (capex) over the last five years. Microsoft, an early investor in OpenAI (the organization behind ChatGPT), has more than tripled its capex spending, fueled by growing demand for data centers.
Combined capex spending at these companies rose by 84% from 2024 to 2025 alone, going from $224 billion to $413 billion. Results through the second quarter of 2026 show that pace is accelerating further: the four companies together spent $171.2 billion in the quarter, up 78.6% from $95.8 billion in the second quarter of 2025.
Each company also raised its full-year 2026 capex outlook during earnings season. Microsoft expects roughly $190 billion for calendar 2026.
- Alphabet raised its guidance to $195 billion to $205 billion, up from $180 billion to $190 billion.
- Meta narrowed its range to $130 billion to $145 billion.
- Amazon raised its guidance twice this year, most recently to $220 billion on its July 30, 2026 earnings call, up from the $200 billion figure it gave in February.
- Combined, the four companies' capex guidance sums to roughly $745 billion for 2026, up from the $600 billion to $700 billion range implied by earlier guidance.
The figures represent total capex for each company. A significant majority reflects AI infrastructure and data centers; each company described its 2026 spending as primarily driven by AI and cloud build-out, but the figures also include other non-AI capital assets.
AI spending is split between chips and infrastructure
The breakdown of AI data center investment reveals that hardware, primarily GPUs and other AI chips, accounts for the majority of total spend, with physical infrastructure representing the remainder.
- Chips account for roughly 60% of total AI data center investment, according to McKinsey's 2025 analysis of the projected $5.2 trillion in AI infrastructure spending through 2030. The remaining 40% covers power systems, cooling, construction, networking, and land.
- AI spending is split nearly evenly between chips and other infrastructure at the company level, according to New Street Research. Microsoft and Alphabet spend more on chips as a share of total AI capex than Meta and Amazon, based on their analysis, a pattern likely driven by how deeply each company has integrated AI into its core products. Both Microsoft and Alphabet have confirmed roughly two-thirds and 60%, respectively, of their most recent quarterly capex went to chips over physical infrastructure.
- Goldman Sachs estimates roughly $180 billion of 2026 hyperscaler spending will go toward GPUs and AI accelerators. Nvidia (NVDA +2.93%), the leading GPU designer, has benefited from strong demand from hyperscalers in recent years.
Data center construction spending has grown 129% in two years
Private data center construction spending in the United States reached a seasonally adjusted annual rate of $59.3 billion per month in May 2026 (preliminary), up 129% from $24.2 billion in December 2023, according to U.S. Census Bureau construction spending data.
- The pace of growth has been consistent and steep. In January 2024, the monthly rate was $26.4 billion. By January 2025, it had climbed to $45.3 billion. By May 2026, the rate reached $59.3 billion, the highest on record and still climbing.
- The trend is relatively new. Monthly data center construction spending averaged roughly $9–10 billion from 2020 through early 2022. Spending quickly ramped up in early 2023 and has not plateaued.
- The data captures physical construction only, not IT equipment purchases. It is reported as a seasonally adjusted annual rate — meaning the monthly figure is scaled to reflect what spending would be over a full year at that pace. These are Census Bureau estimates and are subject to revision.
