The global edge data center market is estimated at $40.0 billion in 2026 and projected to reach $105.8 billion by 2033, a 14.9% compound annual growth rate, according to Grand View Research. That number sits on top of a bigger wave: JLL forecasts 100 GW of new data center capacity between 2026 and 2030, doubling global capacity to 200 GW, with total spending approaching $3 trillion once tenant fit-out is included.
The shift behind both figures is the same. AI workloads were about a quarter of all data center demand in 2025, most of it training. JLL expects inference to overtake training as the dominant AI workload by 2027. Training can run in a few massive campuses. Inference has to sit close to users, which pushes compute out to regional and edge sites.
That is the nut of it: the next five years of infrastructure spending will not just scale up existing campuses, it will spread out. If JLL is right that inference becomes the primary driver, the winners will be operators who can place power and compute within milliseconds of end users, not just those who can pour concrete fastest in Virginia or Texas.
The money follows the workload
JLL puts global data center construction cost at $11.3 million per MW in 2026, up 6% from 2025, after a 7% CAGR from 2020 to 2025 when costs rose from $7.7 million to $10.7 million per MW. AI fit-out can cost tenants another $25 million per MW on top of the shell. The sector needs roughly $870 billion in new debt financing to fund the build-out through 2030.
Those costs explain why the market is consolidating. JLL notes that larger, more expensive projects push out smaller developers and accelerate viable projects backed by credible companies. Debt markets stay open for those groups, while speculative entries face a harder time.
The Americas remain the largest region at about 50% of global capacity, growing at a 17% supply CAGR through 2030, with the U.S. accounting for roughly 90% of Americas capacity. APAC grows from 32 GW to 57 GW at a 12% CAGR, with colocation growing 19%. EMEA adds 13 GW at a 10% CAGR, partly driven by sovereign AI cloud demand tied to data privacy rules.
Power is the real bottleneck
Grid connection wait times in primary data center markets now exceed four years, per JLL. Operators are responding with behind-the-meter power and colocated battery storage. Natural gas is projected to play a major role in the U.S. for both bridge power and permanent on-site generation, though some of the largest tenants reject it on sustainability grounds. In EMEA, projects combining renewables with private wire transmission can cut tenant power costs by 40% compared with grid power.
Some markets, including Ireland and Texas, have implemented bring-your-own-power mandates. That pushes operators to fund their own generation rather than wait on utilities. It also favors sites where land, permits, and fuel access come together quickly.
Edge sites face a sharper version of the same problem. Grand View Research flags that edge facilities often sit in urban or semi-urban locations where local grids are already stressed. High-density GPU servers, cooling, and networking can exceed regional capacity, and operators may face restrictions on expansion or costly upgrades. Battery storage and renewable integration are becoming standard responses.
On-site generation is not cheap, and it is not fast. Turbine lead times have stretched, and battery storage adds cost per MW that tenants must absorb or pass on. JLL's $11.3 million per MW figure covers shell and core only, so the true delivered cost of a powered, cooled, ready-to-rack megawatt in 2026 is materially higher.
Cooling follows density
Deloitte projects next-generation AI racks could reach 370 kW in 2026, making liquid cooling and advanced power architectures essential rather than optional. Direct-to-chip and rear-door heat exchangers are leading the scalable options. Prime Group, Microsoft, and Hanwha announced in April 2026 a nationwide U.S. network of edge data centers integrated with battery energy storage, built around real-time AI inference using existing real estate and energy assets.
Liquid cooling is not a nice-to-have at these densities. Air cooling fails above a certain rack density, and retrofitting after lease-up is expensive. Build.inc notes that developers need a plan for air, hybrid, or liquid-ready halls before lease-up, because retrofitting cooling after tenant requirements are known is far costlier than designing for it upfront.
Edge growth has a geography problem
Grand View Research puts North America at 35.5% of edge revenue in 2025, with Asia Pacific as the fastest-growing region. The U.S. held the largest country share. That matches the broader data center picture where the Americas lead on capacity and growth rate.
The inference shift is what drives the geography. JLL is explicit that inference demand requires geographical distribution to reduce latency and serve users effectively, and that this will drive regional deployments and embedded systems at the edge. Every AI model deployment creates sustained inference demand that grows with user adoption, but that growth depends on inference applications that do not yet exist at scale.
That caveat matters. Grand View Research frames the edge market as driven by IoT, 5G, autonomous systems, and real-time analytics as much as by AI. The AI inference story is the headline, but the underlying demand is broader and older. Operators building edge capacity on AI inference alone are taking a timing bet.
What to watch
Three things will test the forecasts. First, whether inference applications actually arrive at scale by 2027, since JLL ties the workload shift to applications that do not yet exist. Second, whether grid interconnection delays ease or worsen as 100 GW of new capacity competes for the same utility queues. Third, whether liquid cooling and on-site power costs come down fast enough to keep edge projects economical outside the top tier of operators.
For more on the cloud and edge landscape, see our cloud and edge coverage. For capacity and pricing detail, Reuters has been tracking data center power constraints at reuters.com.
The firms that win will be the ones that treat power as a product, not a utility bill. Everyone else is building shells that cannot be turned on.