Data-Center Forecasts Point to 217 GW by 2030, 2.6x 2025’s

“Tracked forecasts put global data-center capacity at a median of 217 GW by 2030, up from a median of 83 GW in 2025,” Shanu Mathew, whose X profile lists AI power and data centers among his topics, posted on X on 23 October 2026. He says he attempted to tie every installed-capacity estimate he could find to a published number. By our arithmetic, 217 GW is about 2.6 times 83 GW.

His post adds that across 9 publishers, “2030 estimates run 173–301 GW,” and across 10, “2025 estimates run 70–116 GW.” He also warns that publishers do not disclose on the same basis, so readers should “take all of this with a grain of salt: it’s directional color.”

Business Pill · TOKENS PER WATT

A short Business Engineer explainer on tokens per watt: how much AI output a unit of power produces. It teaches the general idea only and says nothing about the estimates in this piece.

What the Estimates Show

The table below copies the medians and ranges from the first chart attached to his post. Each figure is his compilation of other publishers’ estimates, not a forecast of ours. A gigawatt (GW) is a measure of power: here, the electricity that data-center capacity is rated to draw.

Source: Shanu Mathew’s chart, “Capacity estimates by region,” posted on X on 23 October 2026. The chart notes that regional medians “describe different source populations and are not a reconciled regional split,” so the rows are not added together.

The chart leaves some numbers out of its statistics on purpose. It marks Gartner’s 290 GW for 2030 as a demand figure and excludes it, and it notes that “most broker figures do not state IT vs facility power.” In plain words, some publishers count only the power the servers use, while others count the whole building, including cooling.

For the United States, his post says 2025 estimates run 34–50 GW, against “33 GW of tracked live IT,” a figure the chart labels BNEF/DC Byte. That gap between estimated and tracked capacity is one reason he puts the weight on medians rather than any single number.

What Was Posted on X

Shanu Mathew’s post with the two charts this piece draws on. He writes that tracked forecasts put global data-center capacity at a median of 217 GW by 2030, up from 83 GW in 2025, that 2030 estimates across 9 publishers run 173 to 301 GW, and that the median increase for publishers with both years is 116 GW, about 23 GW a year. He calls the result directional, because publishers disclose on different bases.

How Fast Capacity Is Added

His second chart turns the stock figures into yearly additions. Its headline reads: “Published annual paths put 2027 additions at 15–51 GW; endpoint-implied averages run 18–44 GW a year.”

The straight-line averages simply spread each publisher’s 2025-to-2030 increase evenly over five years. On his chart, Wells Fargo gives the highest average, about 44 GW a year, and BloombergNEF’s NEO the lowest, about 18 GW. His post says that for the 9 publishers with both years, “the median increase is 116 GW, about 23 GW a year on a straight line.”

The chart itself warns that its two panels are “different measures, not comparable”, and that “electrical definitions vary by source.” The 116 GW median increase also differs from the 134 GW gap between the two medians, because, as the chart notes, the change in median “is the difference between cross-sectional medians, not a build forecast.”

Connecting the Coverage

Our reporting this week has followed the same build-out from the spending, the demand and the local side. Gartner forecasts spending on data center systems to rise 55.5% to $1,366 billion in 2027, our Gartner piece reported, about half of the year’s rise in IT spending by our arithmetic on its table.

On demand, Amazon chief executive Andy Jassy said “2027 is largely sold out” of AWS computing capacity, Reuters reported, as covered in our AI financing analysis. On the local side, Amazon withdrew its plan for four data center buildings on a $427 million campus in Ashburn, Virginia, our Loudoun piece reported.

The same piece carries S&P Global Ratings’ view that it “believes power constraints, permitting challenges, public pushback and regulatory responses could delay or constrain new data center development.” On water, Capital Brief questioned whether the site Anthropic leased in Queensland can access the nearly 27 million litres each building needs to fill its cooling systems, as our Queensland piece reported; Zerra DC said the buildings would be entirely air-cooled, according to ABC News.

Single companies now state gigawatt goals too. Mistral says its compute roadmap is “targeting 1 GW of capacity by 2030,” per the Update in our GTC Berlin wrap. TikTok’s U.S. entity is looking for four gigawatts of compute power by the early 2030s, according to one source cited by CNBC, as our TikTok piece reported.

The Structural Read

As we read it, the 217 GW figure says where published estimates cluster, as his post frames it with medians, rather than how much will be built: his post notes the publishers report on different bases.

As we read it, the United States carries the largest regional increase in his chart, from a 39 GW median to 115.5 GW, while the chart says regional medians are not a reconciled split of the global total.

As we read it, the local limits in our recent reporting, from S&P’s view on power and permitting to Amazon’s withdrawn Ashburn buildings, sit on the same build-out the estimates describe; we did not test any link between them.

Three Implications

FOR READERS OF FORECASTS His charts show 2030 estimates from 173 GW (BNEF NEO, IEA) to 301 GW (Wells Fargo), and note that most broker figures do not state IT vs facility power.

FOR GRID AND SITE PLANNERS S&P Global Ratings said it believes power constraints, permitting challenges, public pushback and regulatory responses could delay or constrain new data center development, as our Loudoun piece reported.

FOR AI COMPANIES Mistral says its compute roadmap targets 1 GW of capacity by 2030, and Reuters reported Jassy saying AWS capacity for 2027 is largely sold out.

The Business Engineer Lens

This story maps onto the Business Engineer framework The State of AI Data Centers.

The framework puts its thesis as “The structural bottleneck isn’t computing, it’s electrons.”

As we read it, the compilation gives that thesis a number: the median published estimate rises from 83 GW to 217 GW over five years, per Shanu Mathew. The framework was published in December 2025 and does not discuss this compilation.

What Is Not Established

These are not our forecasts. The compilation is the work of one analyst: the charts label it “author analysis,” and his X profile says his posts are personal views, not his employer’s, and not advice. We have not seen the register he says ties each value to a source, and figures read from his charts are rounded where the chart itself rounds them.

The publishers count on different bases: facility versus IT power, stocks versus flows, annual paths versus point estimates, as his post notes. The medians therefore describe where published estimates cluster, not a measured build plan, and the regional rows are not a reconciled split of the global total.

The Bottom Line

Shanu Mathew’s compilation puts the median published estimate for global data-center capacity at 217 GW for end-2030, against 83 GW for end-2025, about 2.6 times by our arithmetic. The 2030 estimates he gathered still span 173 GW to 301 GW, and he calls the result “directional color.”