How Many Data Centers Were Built Since 2020 (Through End-2025) - And What’s Planned by 2030?
People ask this like there’s a single global registry.
There isn’t.
“Data center” can mean anything from a small enterprise server room to a 1-gigawatt AI campus. So the cleanest way to answer is:
Count hyperscale facilities (the big cloud/AI operators) as a reliable, tracked subset
Use capacity forecasts (GW) to describe what’s “planned” toward 2030
Describe typical sizes (MW + footprint + rack density), because “how many” alone is misleading
Numbers at a glance
A) Hyperscale facilities built/added since 2020 (best tracked global count)
Synergy Research Group’s hyperscale facility count (large data centers run by hyperscale cloud/internet companies):
End of 2020: 597 facilities
End of 2023: 992 facilities
End of 2024: 1,136 facilities
Q3 2025: 1,297 facilities (late-2025 snapshot)
What that implies: from end-2020 (597) to late-2025 (~1,300), the world added roughly ~700 hyperscale data centers (net new openings).
Important: This is not “all data centers.” It’s the subset that’s driving most of the new power + AI buildout, and it’s the subset we can count with decent confidence.
B) Capacity planned by 2030 (the number that actually matters)
Multiple forecasts converge on roughly “double global capacity by 2030,” but they differ on the exact ceiling:
JLL’s 2026 outlook: +97 GW added from 2025 → 2030, reaching ~200 GW by 2030
McKinsey (global demand): around ~220 GW by 2030
So when someone says “how many are planned by 2030,” the most honest translation is:
Planned buildout = ~+100 GW of new capacity (2025–2030) (give or take, depending on forecast and delays).
Why “how many data centers” is a trick question
Because the industry is shifting from:
many smaller rooms (enterprise/on-prem)
tofewer but massively larger sites (hyperscale + colocation + AI campuses)
Synergy explicitly notes that new facilities are getting bigger, so capacity grows faster than site count.
That’s why “we built 200 new data centers” could mean:
200 small 2–5 MW sites (modest impact), or
200 hyperscale/AI sites (gigawatt-scale grid impact)
What sizes are we talking about? (MW, footprint, and density)
1) “Large-scale” traditional data centers (still common)
Often ~30–60 MW per site is a standard “big” build class discussed in industry planning.
2) Hyperscale sites (cloud-scale)
Commonly tens of MW up toward ~100 MW+, with the trend moving upward
A 100 MW hyperscale site can occupy on the order of ~10 acres (campus varies widely)
3) AI campuses (the new category that changes everything)
This is the “AI boom” signature: campuses measured in hundreds of MW:
Industry discussions now regularly talk about adding 100, 200, 500 MW — even ~1 GW at a single campus
This is also why power availability has become the #1 constraint.
4) Rack power density (why AI data centers look different)
AI training/inference pushes extreme density:
Training workloads can require ~100–200+ kW per rack, driving liquid cooling and new electrical designs
This density shift is a key reason “a newer AI data center” can deliver far more compute than an older building of the same size.
“How many are planned by 2030?” — two practical answers
Answer 1 (capacity-first, best for investors/strategy)
By 2030, the world is likely adding ~100 GW of net new data center capacity versus 2025, roughly doubling total capacity.
Answer 2 (facility pipeline, best for “site count” thinking)
Synergy’s tracked hyperscale pipeline has been reported in the hundreds of facilities—one late-2025 report cited a known pipeline of ~770 hyperscale facilities at various stages (planned/designed/under construction).
Caveat: “pipeline” does not guarantee completion by 2030; some slip, some get redesigned, some get canceled, and some are phased.
What to watch between now and 2030 (the real bottlenecks)
Power delivery + grid interconnect timelines (often slower than building shells)
Cooling and water constraints as rack density rises
Campus-style builds (100–1000 MW) becoming the norm for frontier AI workloads