Lambda expands owned AI data centers

This build out turns Lambda from a broker of rented GPU capacity into an owner operator with more control over cost, timing, and customer commitments. That matters because frontier AI customers do not just buy GPUs, they buy guaranteed power, cooling, and floor space for multi year clusters. Owning or locking up the site lets Lambda match long contracts with long lived infrastructure, the same playbook that helped CoreWeave and Crusoe move upmarket.1 2 3 4 5 6

• Lambda has already put concrete projects behind the shift. Kansas City is planned to launch in early 2026 with 24MW, more than 10,000 NVIDIA Blackwell Ultra GPUs, room to scale past 100MW, and the whole initial deployment committed to one customer on a multi year deal.4
• The EdgeConneX expansion shows why long term leased space matters operationally. Chicago alone is a single tenant 23MW site, ready for service in 2026, using direct to chip liquid cooling plus air cooling, which is the kind of custom setup needed for dense AI racks rather than generic colocation.5
• This also changes Lambda’s business mix. It shut down low margin hardware sales in August 2025 and narrowed product focus around large training deployments, while revenue scaled from $425M at the end of 2024 to an estimated $760M at the end of 2025, still below CoreWeave but increasingly on the same infrastructure path.1 2 3 6

The next step is a fuller shift from developer cloud to utility scale AI infrastructure. If Lambda keeps converting anchor contracts into dedicated campuses and long leaseholds, it can sell not just GPU hours but reserved superclusters, which should make capacity a bigger competitive weapon than software alone.1 2 4 5