The growth of AI is unfolding through a highly connected ecosystem, but what physically powers the system is the constraint.
AI demand begins with model platforms (e.g., OpenAI, Anthropic, Google) and flows downstream through the compute and systems layer, where NVIDIA, AMD, and Broadcom supply chips integrated into AI servers by the major cloud providers. These fleets are deployed across hyperscaler and specialist campuses (e.g., Equinix, Digital Realty, CoreWeave), supported by real estate developers, EPC partners, and capital providers (e.g., CBRE, JLL, KKR, BlackRock). Reliable operation depends on utilities delivering firm, redundant power and on policymakers (e.g., DOE, EPA) accelerating what can be permitted and financed.
In 2025, it became apparent that capital, data center site selection, and multi-supplier models and compute are rapidly coming together to amount to actual data center buildouts as demonstrated by notable initiatives such as:
- Microsoft is on pace for roughly $80B in FY25 AI-data-center spend
- Apple committed $500B in U.S. investment over four years
- OpenAI + Oracle jointly announced ~4.5 GW of “Stargate” capacity
- An OpenAI + Oracle + SoftBank consortium outlined five U.S. sites targeting ~7 GW
- BlackRock’s acquisition of Aligned Data Centers positioned the platform as a core AI-capacity vehicle
However, power has become the limiting factor in turning AI data center plans into real projects. Data center electricity demand could rise to 7–12% of total U.S. consumption by 2030, yet connecting new capacity to the grid often takes 3–5 years, and even longer for 200-megawatt-plus campuses. Additional barriers to providing power include shortages of power transformers and switchgear, rising construction costs of $7–12 million per megawatt, lengthy permitting processes, skilled labor shortages, and site-level limitations such as water access, fiber connectivity, and available Tier-1 land.
Across the industry, consensus is clear that energy is the constraint. Campuses will only be built where utilities can deliver large, reliable capacity on a predictable timeline. In the future, the advantage will go to these developers and platforms that secure power first, standardize designs to improve buildout timelines, and diversify their providers that supply compute.
