I Squared Capital Bets $1 Billion on AI Inference Data Centers

On May 27, I Squared Capital signed an agreement to acquire ten data centers from Cogen Fiber, a Cogent Communications Holdings subsidiary, for $225 million in cash. The facilities span 259,000 square feet across nine U.S. markets: Chicago, Atlanta, Phoenix, Los Angeles, Kansas City, Baltimore, Houston, Nashville, and Stockton, California.

I Squared plans to invest up to $1 billion in building and expanding U.S. data centers designed for artificial intelligence computing. The capital is earmarked for targeted improvements and additional acquisitions.

Each facility sits near internet exchanges and offers multi-carrier connectivity. Co-founder and Managing Partner Gautam Bhandari identified location, power, and connectivity as the three variables determining long-term data center value.

All acquired assets include expansion capacity and support liquid-cooling-enabled configurations, a system that uses liquid to cool computer chips. Bhandari said demand for high-density, low-latency facilities will grow as AI shifts from model training to inference, when systems are used by people and businesses daily.

I Squared oversees $60 billion in assets across global infrastructure businesses, with a portfolio of at least 90 companies in utilities, energy, environmental infrastructure, transportation, and social infrastructure. The firm employs 300 people across offices in Miami’s Brickell Financial District, Abu Dhabi, London, Munich, New Delhi, São Paulo, Singapore, Sydney, and Taipei.

AI data centers require enormous amounts of electricity to run and cool the powerful computers that train and operate artificial intelligence systems, making power access and cooling capacity central to their viability.

The $225 million acquisition price implies roughly $870 per square foot for assets with expansion potential and liquid cooling readiness. That is below replacement cost for new build-to-suit AI data centers, which can exceed $1,500 per square foot in primary markets, per CBRE data.

I Squared is betting that inference-phase computing will generate sustained demand for distributed, low-latency facilities. Training workloads concentrate in hyperscale campuses; inference workloads must sit closer to end users to minimize latency.

The nine markets selected are secondary or tertiary data center hubs, not Northern Virginia or Silicon Valley. That suggests I Squared is targeting cost-advantaged locations with available power and fiber connectivity, avoiding the capacity constraints and premium pricing of Tier 1 markets.

Power availability is the binding constraint for new data center development. Utilities in Northern Virginia, for example, have imposed moratoriums on new connections due to grid capacity limits. I Squared’s strategy of acquiring existing facilities with expansion capacity sidesteps that bottleneck.

Liquid cooling is another differentiator. Traditional air-cooled data centers cannot efficiently dissipate heat from the latest AI chips, which draw 700 watts or more per processor. Liquid cooling reduces energy consumption for cooling by up to 40%, per Uptime Institute data.

The deal signals that institutional capital sees AI infrastructure as a long-duration, inflation-hedged asset class. I Squared’s $60 billion infrastructure platform provides the patient capital needed to execute a $1 billion buildout without the quarterly earnings pressure that public REITs face.

Private equity’s appetite for data centers has surged. Global data center investment reached $48 billion in 2025, up from $32 billion in 2023, per JLL. I Squared’s commitment is one of the largest single-platform bets on inference-stage infrastructure.

The question for lenders and co-investors is whether inference demand materializes at the scale and pace that underwrite these returns. Hyperscalers like Microsoft, Amazon, and Google are spending heavily on training infrastructure. Inference is still nascent, but the unit economics improve as AI applications proliferate.

I Squared’s $225 million cash acquisition of ten facilities is a measured entry point. The $1 billion total commitment, if executed, will test whether secondary-market, liquid-cooled, inference-ready data centers can deliver infrastructure-grade returns. The answer will shape capital flows into the next phase of AI buildout.