Cloud region launches have dropped from more than 40 in 2019 to 16 in 2025, according to TeleGeography data cited by analysts Jon Hjembo and Patrick Christian. The same data puts 2024 at 13 new regions globally. That is a sharp reversal from the last five years, when providers raced to plant flags in emerging markets.
The cause is a capital shift, not a demand collapse. Hyperscalers are moving money away from geographic expansion in Africa, the Middle East, and other developing markets and into massive AI compute clusters in the United States and Europe. TeleGeography's read is that the AI boom has not yet produced a spike in long-haul or cross-border bandwidth demand. Traffic growth is concentrated inside campuses and metro dark fiber networks where the training clusters sit.
The stakes land in November 2026 because the pipeline underneath this shift has never been larger. TeleGeography tracks roughly 650 data center sites in the immediate pipeline, more than double the historical volume. About 35 percent of all new builds are in the United States, and that figure rises past 60 percent when isolating AI and hyperscale facilities. Export controls on Nvidia chips are part of the reason: operators are scaling up in the U.S. or partner nations such as Saudi Arabia rather than distributing infrastructure evenly worldwide.
Power, Not Proximity, Is the New Bottleneck
The binding constraint has moved. In Northern Virginia, the world's largest data center market, reserve power capacity is projected to fall from 4 percent to 1 percent in the near future. TeleGeography estimates wholesale electricity prices in that market could rise by roughly 11 times by 2027, with brownout risk increasing alongside the price.
Developers are responding by moving to non-traditional markets chosen for power generation and land rather than latency or population density. West Texas and Columbus, Ohio are two examples. In Columbus, planned data centers account for roughly 13 percent of all planned power. The tradeoff is straightforward: cheaper power farther from users, with longer fiber runs and new network investment to close the gap.
AI Concentration Reshapes the Map
The geographic pull of AI compute is measurable in the build mix. Hyperscale and AI facilities are concentrated in the U.S. at more than 60 percent of new builds, while commercial data center construction is spread more evenly around the world. That split matters for buyers of cloud services in Asia, Latin America, and Africa, who are watching capacity arrive late or not at all while the core markets get the clusters.
It's not clear this concentration will hold. TeleGeography flags "bubble-like" circularity in the economy behind it. Microsoft, OpenAI, Nvidia, and Oracle are investing in one another to guarantee pipelines and perceived demand, with a tech giant sometimes investing in a startup that then buys chips from a partner or rents cloud space back from the investor. Meta's $27 billion debt financing for a campus in Louisiana shows how aggressively these projects are now funded. The structure makes true organic demand hard to gauge.
What Buyers Should Watch
For enterprises, the practical effect is that cloud capacity in core markets keeps growing while expansion to the edge slows. Latency-sensitive workloads that assumed a new region in their geography may need to wait, or to run on edge nodes and CDN capacity instead. TeleGeography's Market Connectivity Score, which ranks cities on a 0-100 scale for connectedness and growth, is one tool buyers can use to track where infrastructure is actually landing.
The shift also changes how network planners think about interconnection. When cloud regions were launching in new countries every quarter, carriers built long-haul capacity ahead of demand. Now that launches have slowed to 13 in 2024 and 16 in 2025, that speculative build has pulled back too. The result is that bandwidth demand growth is concentrated inside metro and campus networks where the training clusters reside, rather than across borders. For a company running inference workloads close to users, this means edge capacity matters more than it did two years ago, and the supply of that capacity is not growing as fast as the marketing suggests.
Financing and the Bubble Question
The money behind the boom is worth watching on its own terms. A small group of companies, Microsoft, OpenAI, Nvidia, and Oracle, are investing in one another to guarantee pipelines and perceived demand. A tech giant invests in a startup, which then uses that capital to buy chips from a partner or rent cloud space back from the investor. The loop makes it hard to tell how much demand is organic.
Debt is filling the gap. Meta's $27 billion debt financing for a campus in Louisiana is one example of how these projects are funded now, a shift from the equity-heavy model of earlier cloud buildouts. TeleGeography describes the economy as exhibiting "bubble-like" circularity, and the comparison to a gold rush where prospectors have stopped buying maps to new territories and instead build refineries next to the biggest power plants is apt. Banks funding the expedition are, in effect, the prospectors lending money to each other.
What This Means for 2026
The forward-looking question is whether the 650-site pipeline converts into built capacity or stalls on power permits and financing. With Northern Virginia reserve capacity heading toward 1 percent and Columbus planning data centers at 13 percent of projected power, the answer depends on grids, not on cloud demand. TeleGeography's data center research service tracks sites by operator, location, and networks connected, which is where the next quarter's reality will show up first.
For cloud buyers, the practical takeaway is to stop assuming a region will arrive in their geography on schedule. Plan for core-market capacity with edge fallback, and treat vendor roadmaps as intentions rather than commitments. The retraction to the core is not a temporary pause; it is a reallocation of capital toward AI compute, and it will shape where low-latency infrastructure exists for years.
Read more at TeleGeography's cloud and data center outlook.
Related coverage: cloud and edge computing.