Big Tech locks up nearly $1.09 trillion in AI data centers

Microsoft, Amazon, Alphabet, Meta and Oracle have committed to nearly $1.09 trillion in long-term leases for data centers that are still being built or are not yet operational, according to Reuters reporting cited by ITzine.

The figure is almost four times the $285 billion in related lease obligations currently recognized on the companies' balance sheets. The difference comes from accounting rules: leases are generally recorded as ordinary lease liabilities once a facility is available for use, while commitments for future sites appear separately in financial-statement notes.

That makes the headline number a measure of contracted future capacity rather than data centers already running. It also shows why the companies' reported infrastructure commitments can rise sharply before the associated facilities begin generating revenue.

Oracle carries the largest disclosed commitment

Oracle accounts for the biggest disclosed share. The company has reported $260 billion in future data-center lease commitments, compared with $37.89 billion in lease obligations already recognized.

Most of Oracle’s new facilities are expected to become operational during fiscal 2027 through 2029. The leases run for 15 to 19 years, although Oracle warned that the duration and terms of its leases may not match the contracts it signs with customers. Its filing identifies that mismatch as a financial risk if demand falls below expectations.

Recommended reading

SpaceX targets carriers as Musk forecasts $1 trillion revenue

The source does not provide a company-by-company breakdown of the remaining commitments, nor does it identify how much of the aggregate total is tied to US versus European sites. Oracle also has not disclosed per-kilowatt or per-rack rates for the future facilities, the exact start date and location of each lease, or whether contracts include power procurement and pass-through energy costs.

What the capacity costs in the US and Europe

AI-ready colocation and wholesale capacity is available across major US markets, including Northern Virginia, Phoenix, Chicago, New York and New Jersey, Atlanta, and Silicon Valley. European availability includes London, Frankfurt, Amsterdam, Paris, Dublin and other major FLAPD+ markets, though capacity is tight in core hubs.

For context, CBRE’s Q1 2025 market data cited an example Phoenix rate of about $190 per kW per month, while its global weighted average was $217.30 per kW per month. London was cited at $180–$215 per kW per month. These are market benchmarks, not the rates attached to the leases in the $1.09 trillion figure.

Alternative infrastructure models show how different the economics can be:

• Digital Realty reported new-lease pricing of $256 per kW per month for Americas contracts of 0–1 MW and $174 per kW per month for contracts above 1 MW.
• Retail colocation through providers such as Equinix is commonly quoted at roughly $1,500–$4,500 per full rack per month in major Tier-1 metros, with power-first pricing usually negotiated.
• A Google Cloud a2-highgpu-8g virtual machine with eight A100 GPUs is listed by price aggregators at roughly $29–$30 per hour in major US and European regions.

Those options are not interchangeable: a cloud GPU instance is billed hourly, while wholesale and colocation contracts reserve physical power and space over years. Still, the comparison makes the scale of Big Tech’s commitments clearer. The companies are locking in long-duration capacity at a time when, as recent Big Tech earnings showed, AI spending is already colliding with infrastructure bottlenecks and uncertain enterprise demand.

The core risk is therefore not simply the size of the reservation. It is the gap between when these facilities begin operating, when lease payments are recognized, and whether customer demand arrives on the same schedule. Oracle’s disclosed $260 billion commitment, with many sites not expected until fiscal 2027–2029, is the clearest example of that exposure.