[URGENT] Why power, not AI, may decide the data center race
The global data center industry is entering a sustained expansion phase, with nearly 100 gigawatts of new capacity expected between 2026 and 2030, effectively doubling today’s footprint and pushing total investment toward $3 trillion.
Capacity is projected to grow at a 14 percent compound annual rate through 2030, driven primarily by hyperscale cloud providers and accelerating AI workloads, creating both structural opportunities and mounting pressure on power infrastructure, capital markets, and supply chains.
Hyperscalers will remain the dominant growth engine, pursuing a hybrid strategy of leasing capacity while continuing to self-build large-scale campuses to secure power access, control latency, and manage long-term costs.
By 2030, AI workloads could account for roughly half of all data center demand, up from about one-quarter in 2025, with a pivotal shift expected around 2027 as inference overtakes training as the primary driver of compute consumption.
This transition materially changes deployment economics, as inference workloads require geographically distributed infrastructure closer to end users, increasing demand for regional and edge facilities rather than centralized mega-clusters alone.
Regionally, the Americas will retain leadership, representing roughly half of global capacity and posting the fastest growth, supported by strong U.S. hyperscale investment, while APAC and EMEA expand at lower but still robust double-digit rates tied to cloud migration, sovereign AI requirements, and government-backed digital infrastructure.
Power availability has emerged as the sector’s binding constraint, with grid connection wait times exceeding four years in many core markets, forcing operators to adopt behind-the-meter generation, battery storage, and, in some jurisdictions, direct ownership of power assets.
In the U.S., natural gas is increasingly used as bridge or permanent on-site power, though sustainability concerns among large tenants may limit adoption, while EMEA and APAC lean more heavily on renewables and private-wire solutions to reduce cost and regulatory risk.
Capital intensity continues to rise, with average global construction costs climbing to more than $10 million per megawatt and forecast to increase further, excluding tenant-funded IT fit-outs that can exceed $25 million per megawatt for AI-optimized deployments.
Roughly $1.2 trillion in real estate asset value and nearly $870 billion in new debt financing will be required to deliver the next wave of capacity, alongside an additional $1 trillion to $2 trillion in tenant spending on GPUs, networking, and related infrastructure.
These economics are accelerating industry consolidation, raising barriers to entry, and favoring well-capitalized operators with proven execution, access to power, and balance sheet flexibility.
For investors and developers, the central risks now lie in power procurement, construction timelines, and the pace of AI application adoption, while strategic advantage will accrue to those who can align capital deployment with the shift from training to inference and the resulting geographic redistribution of demand.
Discover more from TBC News
Subscribe to get the latest posts sent to your email.
