TD Cowen: Microsoft cancels leases for a substantial amount of US data centers, potentially due to over-investment fears; Microsoft plans to spend $80B in 2025
- The company has pledged $80 billion toward computing capacity — Wall Street is questioning AI demand over the longer-term
Context & Ripple Effects
Microsoft's reported lease cancellations complicate its earlier $80B FY2025 data-center buildout plan, which was framed around capacity for AI workloads and cloud applications.
The report became an early marker of a broader pullback: subsequent coverage said Microsoft had walked away from projects in the US and Europe expected to consume 2 gigawatts. Together, the coverage highlights a gap between announced infrastructure spending and the timing of capacity commitments.
First-order effects
- Microsoft can reduce near-term leased data-center commitments while retaining its stated plan to spend heavily on computing capacity; lessors and project developers lose anticipated demand where leases are canceled.
- The cancellations give investors a concrete reason to test whether AI-related capacity is being committed ahead of durable demand, rather than treating headline capex plans as a direct proxy for deployed infrastructure.
Second-order effects
- Data-center developers, utilities, and other capacity suppliers may need to reallocate or defer projects that had been oriented toward Microsoft demand, while competing cloud providers face greater scrutiny of their own build schedules.
- Microsoft's mix of owned construction and leased capacity becomes more consequential: lease flexibility can limit exposure to demand timing changes, but makes planned supply less certain for infrastructure counterparties.
Third-order effects
- If similar revisions persist, AI infrastructure could follow a more uneven capital cycle, with large spending announcements followed by selective cancellations or freezes as cloud providers match capacity to realized workloads.
- The pattern would shift attention from aggregate AI capex to execution quality: the location, contract structure, and utilization of capacity may matter more than announced totals.
The trend: AI infrastructure is entering a discipline phase in which hyperscalers continue investing but adjust individual capacity commitments as demand visibility evolves.