The AI boom is driving unprecedented physical capital formation, with global data center investment projected to hit at least $3 trillion through 2030, peaking near $750 billion annually. A single modern 200 MW AI training data center costs approximately $8.2 billion, with facility and power infrastructure making up one-third of this cost, and IT equipment accounting for the rest. Hyperscaler capital expenditures are also surging, with Microsoft, Amazon, Meta, Google, and Oracle projected to spend $655 billion in 2026, a 58% year-over-year increase, absorbing nearly all of their operating cash flow.
Financing for this massive build-out presents a challenge, as only 48% of the estimated $2.9 trillion needed between 2025 and 2028 will come from internal cash flows. The remaining 52% will rely on external financing, including $800 billion in private credit, $200 billion in corporate debt, and $150 billion in securitized instruments. Private credit providers like Blue Owl, Blackstone, Apollo, Brookfield, and KKR are already dominant, holding 60-70% of the $200-$250 billion deployed in the data center space.
Power is emerging as the most significant bottleneck. Bloomberg New Energy Finance estimates a 19-gigawatt power shortfall for AI data centers by 2035 if current growth rates continue. For example, Australia's data center electricity use is expected to rise sevenfold by 2036, reaching 34 terawatt-hours and 13% of the National Electricity Market's power. In the US, there's been a nearly doubling of gas-fired power development for data centers. The situation is so critical that Texas Governor Greg Abbott ordered an audit of all new data center connections to the ERCOT grid, potentially delaying 49.8 GW of new demand, almost 20% of the US development pipeline, and costing projects up to $15 billion.
Industry experts and analysts echo these concerns. Canaccord Genuity analyst George Gianarikas notes that the ambitions of data center companies are unlikely to be met at their expected pace due to equipment, permits, people, and growing public resistance leading to protests and moratoriums. Despite these challenges, demand remains exceptionally high, with hyperscalers reporting that demand far exceeds supply. Amazon CEO Andy Jassy forecasts AWS to become a $1 trillion revenue business, citing striking demand for 2028, and CoreWeave CEO Michael Intrator stated that near-term capacity is effectively sold out, pointing to a persistent systemic disequilibrium.
The Texas audit highlights the growing tensions between data center expansion and grid reliability, as the 474 GW queue of large-load requests (90% data centers) is more than five times the state's record peak electricity demand. QTS Data Centers, a Blackstone portfolio company, has endorsed the audit, emphasizing the need for clear guardrails, transparency, accountability, and responsible growth. The potential delays could result in significant revenue losses, with BNEF estimating up to $15 billion in cumulative losses by Q1 2027 for a full AI-compute mix scenario, based on GPU rental costs of approximately $1.76 billion per GW per month.
