Great Scott! The AI-fueled data center build-out could double next year: Chart of the Day

Doc Brown needed 1.21 gigawatts to send a DeLorean through time in the 1985 movie "Back to the Future." The data-center industry could add about 42 times that next year.

Research firm SemiAnalysis expects annual additions to data center IT capacity outside China to more than double next year, to about 51 gigawatts from 25 gigawatts this year.

In movie units, that's about 21 Doc Browns ballooning to 42. These are additions in a single year, not the whole fleet or the electricity it uses. By SemiAnalysis's count, that one year would roughly match the world's entire data center build-out through 2024.

About three-quarters of next year's projected additions come from projects already operational or under construction. The rest are still planned. SemiAnalysis points to more than 24 gigawatts across 40 campuses in advanced construction for expected delivery next year.

Not everyone buys it. At the All-In Summit in September, Altimeter Capital's Brad Gerstner called SemiAnalysis's forecast at the time, 43 gigawatts, too aggressive and guessed the industry would stand up closer to 25 gigawatts. That's only about 21 Doc Browns — no faster than this year.

How do you build that much, that quickly? Increasingly, in pieces.

Suppliers now build electrical rooms and cooling systems in factories while crews prepare the site. SemiAnalysis calls the results Lego data centers, with bricks that can weigh 50,000 pounds.

Amazon ( AMZN ) is rolling out its own modular design at large scale, and Meta Platforms ( META ) has put up tent-like halls to enclose buildings faster, according to SemiAnalysis.

A SemiAnalysis model of a 50-megawatt AI data hall puts conventional construction at 22 months and a fully modular approach at 14, excluding permitting. The biggest saving comes from electrical and mechanical installation. For the owner of the chips inside, SemiAnalysis values that eight-month head start at about $200 million.

That shift puts suppliers in the race alongside chipmakers such as Nvidia ( NVDA ). SemiAnalysis estimates content per megawatt for Vertiv ( VRT ) roughly doubles, to as much as $7 million, when it sells the full modular stack. Contractors Comfort Systems USA ( FIX ) and Sterling Infrastructure ( STRL ) assemble modules too.

But a finished building still needs power.

Microsoft ( MSFT ) confirmed Monday that it pulled out of a lease at a planned 50-megawatt Nscale data center outside London — the same size as the SemiAnalysis model — after grid officials said power might not arrive until the 2030s. Microsoft said the decision involved a single site and that its broader partnership with Nscale continues.