The future data center leasing market is projected to exceed $850 billion, signifying that AI computing power expansion has transitioned from a "capital expenditure-driven" phase to a "long-term contract lock-in" phase. Essentially, tech companies are no longer expanding on demand, but rather securing computing power supply for the next 3-7 years in advance.
Structurally, the incremental growth is concentrated and accelerating in 2025-2026: new commitments in a single quarter increased by 204% year-on-year and 31% quarter-on-quarter, indicating that this round of expansion is not linear growth, but rather a clear "pre-emptive and concentrated deployment." This type of behavior typically corresponds to two stages: first, the demand for AI training and inference has reached a visible inflection point in terms of scale; and second, the supply side (data centers, electricity, chips) is beginning to become a constraint variable.
Structurally, the accelerated new additions by Meta and Microsoft reflect a dual-track expansion of general-purpose models and enterprise AI deployments; Oracle's lead stems more from its role as an infrastructure intermediary, absorbing the spillover demand from large-scale model customers like OpenAI.
For the market, this structure means that the AI narrative is shifting from "growth expectations driving valuations" to "contract and capacity lock-in driving cash flow expectations." The risk factors have also changed: it is no longer whether the demand exists, but whether the electricity, computing power delivery, and capital recovery cycle are matched.
