Skip to content

Invest3 publishers3 min readPublished

Goldman's 'largely unchanged' data-center forecast cuts 2027 capacity additions by more than a quarter

Goldman Sachs calls its US data-center outlook through 2027 largely unchanged, yet its figures cut 2027 capacity additions to 26 GW from 36 GW. The bank says local opposition has had limited effect on growth so far, a judgement the November 3 midterms will test.

The Investor · Invest desk

Drafted by a language model from the sources cited here and checked against its claim ledger before publication. How we use AISend a correction

Illustration accompanying Goldman's 'largely unchanged' data-center forecast cuts 2027 capacity additions by more than a quarter
Generated illustration

What happened

  • Goldman projects US data-center power demand rising 38% in both 2026 and 2027, an extra 12 GW and then 17 GW.
  • Its May forecast had demand climbing from 31 GW in 2025 to 41 GW in 2026 and then to 66 GW in 2027.
  • Only 14% of Americans in a June Reuters/Ipsos poll would support a data center built in their community for big tech firms' AI projects.

Compiled by The InvestorSomething wrong?How this is made

Why it matters

  • contradiction Goldman's 'largely unchanged' holds only for the end-2027 total, about 5% below May's 95 GW, while the path underneath it now loads more of the build into 2026.
  • decision Anyone who sized 2027 power purchases or equipment orders off May's steep path is now working to a 2027 demand step of 17 GW, 8 GW less than the 25 GW May implied.
  • exposure Goldman's case against the opposition rests on a supply pipeline running through 2027, so the evidence it used to discount local resistance does not cover projects dated after that.

Before the October note [1], Goldman's figures put US data-center capacity at 59 GW at the end of 2026 and roughly 95 GW a year later [14][4]. Moving those to 64 GW and 90 GW [3] shrinks the capacity added during 2027 from 36 GW to 26 GW, a cut of about 28% to a single year's build [15]. The bank described the same numbers this way. "While US data center development faces increasing political and community opposition, project data and our updated capacity estimates suggest that the US data center growth outlook through 2027 remains largely unchanged," Goldman said [6].

Demand was reshaped the same way, into two even 38% years [2]. The May path of 31 GW, 41 GW and 66 GW [5] meant growth of about 32% in 2026 and 61% in 2027 [17]. The new 12 GW for 2026 is 38% of about 31.6 GW, so the 2025 base barely moved. Adding 12 GW and then 17 GW to that base gives roughly 61 GW of demand in 2027, about 5 GW below May [18].

Projects arriving early would explain part of the capacity move. Pure timing would have left the end-2027 total at 95 GW, though, so it covers at most 5 GW of the 10 GW cut, and the other 5 GW drops out of the two-year forecast [16]. Another possibility is that opposition is starting to reach later-dated projects. In Crypto Briefing's account, Goldman concludes that resistance has had limited effect on growth so far [12], and a cut to the later year is where an early effect would show first. Or the cause is something else. The reports do not say what drove the cut, and they do not identify electricity supply as the limit on the build.

Goldman's argument, as Crypto Briefing summarised the note, is that AI infrastructure spending by large technology firms will likely offset permitting and political barriers, with a near-term supply pipeline behind it [11]. The evidence the bank cites is project data [6]. "Public sentiment has also turned more negative," Goldman said [7]. About a third of Americans approve of the pace of construction, according to a June Reuters/Ipsos poll [8], and the objections center on electricity demand, utility costs and other local impacts [13]. The midterms are on November 3, and CNA's report calls the backlash a political challenge for Republicans [10].

I'd expect the lost 5 GW to be where opposition shows up first, because it falls in the year furthest from the near-term pipeline Goldman leans on [16][11]. The counter-case is the bank's own: its project data show the outlook holding, and it expects spending to outrun the barriers [6][11]. A later Goldman revision that restores the 2027 figure, or attributes the cut to equipment or power, would show that expectation to be wrong.

What to watch

  • The November 3 midterms and any local moratoria or permit refusals that land on hyperscaler projects scheduled for 2027.
  • Whether US capacity actually reaches Goldman's 64 GW by end-2026; a shortfall would mean the 2026 raise was not projects arriving early.
Loading claim ledger
Loading source directory links
Loading share composer