AI's Trillion-Dollar Infrastructure Buildout is Fueling the Next Wave of Data Center Investment Opportunities

PR Newswire market commentary dated July 29, a 10-year colocation agreement between CleanCore Solutions and Cerebras Systems will dedicate a Tier 3 Minnesota campus entirely to AI compute, with 55 MW of utility power capacity and 40 MW of critical IT load at full buildout. The facility is 100% pre-leased under a contract valued at roughly $800 million over the initial term, with renewal options potentially pushing that figure above $3 billion. The deal extends CleanCore's strategy of building dedicated AI infrastructure and illustrates how accelerator vendors are locking in physical capacity years ahead of first workload deployment.

Compute commitments locked into the grid

The Minnesota campus moves Cerebras from its existing deployment model into a colocation arrangement sized for sustained high-density training and inference workloads. At 40 MW of critical IT load, the facility sits in a band that supports sustained wafer-scale operations within the interconnection limits of typical US utility feeds. Tier 3 design specifies concurrent maintainability — any single distribution path can be taken offline without halting compute operations. For a system architecture as thermally and electrically demanding as wafer-scale silicon, that redundancy envelope is the minimum fault tolerance required to keep clusters at target utilization rates without power capping.

The $800 million baseline contract, scaled across a decade, gives CleanCore a recurring colocation revenue profile that allows capex amortization over a multi-year operating horizon rather than spot-market tenancy. The structure — 100% pre-lease to a single anchor tenant — marks a departure from the diversified tenancy models that historically dominated wholesale data center deals. It indicates that accelerator-heavy architectures now carry enough demand visibility to justify single-tenant hyperscale builds, at least at the tens-of-megawatts tier.

Capacity arithmetic behind the buildout

JLL's projection, cited in the same PR Newswire release, places global data center capacity on a path from roughly 103 GW today to approximately 200 GW by 2030 — a near-doubling that would require on the order of $3 trillion in new infrastructure outlays. McKinsey's higher estimate pushes total worldwide spend toward $7 trillion by decade's end. The gap between the two figures is itself informative: it reflects differing assumptions about power provisioning timelines, land acquisition, and average rack density per facility. As accelerator power envelopes continue rising, the higher figure becomes more defensible.

At the market level, the hyperscale data center segment is projected to grow from approximately $31.4 billion in 2026 to over $52.5 billion by 2030, while the broader global AI infrastructure market is forecast to climb from roughly $75.9 billion to approximately $223.5 billion over the same window. The near-tripling of the AI infrastructure figure versus the roughly 1.7x expansion of the hyperscale segment implies that non-tenant spend — networking, switching fabric, power conversion, cooling, and accelerator silicon — captures an outsized share of incremental capex.

What to watch at the system level

Three variables will determine whether these projections hold. First, utility interconnection queues: a 40 MW IT load campus requires utility-grade substation work, and lead times for that work currently stretch into multiple years in most US grid regions. Second, the thermal envelope — as accelerator power density rises, air cooling gives way to direct-to-chip and immersion loops, each imposing facility-level design constraints that older colocation footprints cannot meet without retrofit. Third, silicon supply: any disruption at the wafer fabrication layer cascades directly into utilization at pre-leased sites.

A simplywall.st analysis dated July 30 notes that companies in cabling, power conversion, and storage — including Belden and Everpure — sit closer to the physical layer of this buildout than the headline chip vendors. Everpure, with approximately $3.9 billion in storage revenue, is positioned at the intersection of AI data gravity and on-premises persistence, though its elevated P/E and reliance on external funding remain open questions per the same analysis. The broader signal from the CleanCore–Cerebras deal is structural: in the current capex cycle, physical compute commitments are being priced and signed before the underlying silicon generations are fully productized.