Big Tech's $1.09 Trillion AI Lease Bet Faces a Power Grid Wall

Big Tech's AI buildout is running into a delivery bottleneck

The core risk is no longer just whether AI demand exists. Big Tech has already committed roughly $1.09 trillion in future lease payments for AI facilities that have not yet started, while hyperscalers are on track to spend around $800 billion on AI infrastructure in 2026. At the same time, nearly half of the US data centers planned for 2026 are facing delays or cancellations. That points to a capacity and execution problem: can all that spending actually turn into usable compute on schedule?

The balance sheet understates the timing risk

These commitments are not hidden, but they are easy to understate. Signed leases generally are not recorded as liabilities until a facility is available for use, so a large part of the AI buildout sits outside reported lease debt until the assets are ready. That can make leverage look lighter than the underlying obligations imply. If AI demand holds, these facilities could support the next phase of cloud growth. If delivery keeps slipping, companies could be locked into paying for expensive, long-lived capacity before it is useful.

Deliverable capacity matters more than booked commitments

For now, execution matters more than the headline spend. A lease commitment is not the same as revenue, and a delayed site is not the same as capacity. If power and grid bottlenecks keep pushing timelines out, the market will have to decide whether these promises are creating future supply or merely financing delays. That is the real near-term rerating lever.

Why this could turn into a winners-take-most market

The next question is whether spending is converting into leased, power-ready square footage.

Supply, not demand, is driving the debate

The scale of the push is hard to ignore. The five largest US cloud and AI infrastructure providers have committed to $660 billion to $690 billion in 2026 capex. That does not prove every project will ship on time, but it does suggest that the biggest buyers still see physical constraints as the main obstacle. In that setup, the advantage goes to the assets that can clear grid and equipment bottlenecks first.

The Stargate project also shows how far ambitions have stretched. OpenAI, SoftBank, and Oracle are backing a $500 billion infrastructure ambition by 2029. That sounds extreme, but it also suggests that at least some buyers expect scarcity to persist. If that is right, the beneficiaries will not be every company claiming an AI real-estate exposure. They will be the projects with the clearest power paths and the shortest routes to operation.

Scarcity can reward only a narrow set of developers

Applied Digital's recent deal is a useful clue. The company said it signed a $7.5 billion, 15-year lease for 300 megawatts at Delta Forge 1, and that the agreement lifted total contracted lease revenue to more than $23 billion. Reuters also reported that Delta Forge 1 was built to handle large-scale AI operations, with the site designed for high-power AI workloads. In a market where the main bottlenecks are power availability and grid interconnection, that profile should matter more than bare land or concrete alone.

That does not mean every developer with a data-center story will win. It does suggest that operators with credible power access, realistic timelines, and hyperscaler-backed demand should be in a stronger position than the broader buildout crowd.

What investors should watch next

The next earnings cycle is the first real pressure test. Investors will want to see whether huge lease commitments and 2026 spending plans are translating into usable capacity, or simply reflecting longer build timelines. One key tension is that capex expected to rise by roughly $534 billion while operating cash flow is expected to increase by about $340 billion more by 2027. That gap does not prove the thesis is wrong, but it does mean timing matters. The market will likely pay closer attention to evidence of deliverable supply than to the size of the plan alone.