TD Cowen: Microsoft walked away from new data center projects in the US and Europe that had been set to consume 2 gigawatts, a sign of AI computing oversupply
The analysts, who rattled investors with a February note highlighting leases Microsoft had abandoned in the US … Bluesky: @mattmday Bluesky: Matt Day / @mattmday : Hmm: “We continue to believe the lease cancellations and deferrals of capacity points to data center oversupply relative to its current demand forecast.” www.bloomberg.com/news/article...
Context & Ripple Effects
This extends TD Cowen's earlier report that Microsoft had cancelled a substantial amount of US data-center leases. The new claim broadens the retrenchment to projects in both the US and Europe and frames it as a demand-versus-capacity issue rather than an isolated lease decision.
The later coverage arc shows infrastructure commitments continuing to be revised: Microsoft was subsequently reported to have frozen additional planned capacity, while broader AI buildout coverage raised questions about capacity limits and returns. This makes project timing and contract flexibility central to interpreting headline AI-infrastructure spending.
First-order effects
- Microsoft's withdrawal removes up to 2GW of prospective demand from the affected data-center projects, leaving developers, landlords, and power-planning counterparts to reassess those specific sites and contracts.
- The move strengthens TD Cowen's near-term case that planned AI compute capacity was ahead of Microsoft's current demand forecast, despite the company's broader AI investment program.
Second-order effects
- Data-center operators and infrastructure suppliers exposed to prospective Microsoft capacity may face slower conversions of planned projects into committed revenue, increasing the value of customers and contracts with firmer demand signals.
- Other hyperscalers and AI infrastructure buyers gain a clearer incentive to phase commitments, favor flexible leases, and distinguish immediate compute requirements from longer-dated capacity options.
Third-order effects
- If similar cancellations persist across major buyers, AI infrastructure may shift from a build-first race toward a more selective, demand-led deployment cycle, with speculative projects bearing more execution risk.
- The episode underscores that a power-constrained long-term buildout can still produce local or temporary compute oversupply: capacity duration and customer commitment quality become as important as aggregate gigawatt targets.
The trend: AI data-center investment is moving toward stricter matching of long-duration power and construction commitments with demonstrable near-term compute demand.