For equities, Goldman’s read favours the bottleneck rather than the end user: companies supplying power generation, transmission, grid modernisation and high-voltage equipment hold pricing power while capacity stays scarce. Multi-year order books give these names unusually long revenue visibility, though execution slippage is now the main risk to earnings delivery. A broadening industrial recovery argues against treating the sector purely as an AI proxy, while construction exposure tied to commercial and residential markets still looks the weaker side of the trade. In defence, European manufacturers able to lift output are best placed, while US contractors face more budget and rate-driven uncertainty.
—
Goldman Sachs says nobody is questioning AI power demand anymore; the real test is whether the US can build the power plants, grid and data centres fast enough to meet it.
Summary:
- Goldman Sachs’ Fundamental Equity investors toured US industrial and data centre power conferences and concluded AI’s main constraint is now delivery, not demand
- Manufacturers are booking orders into the early 2030s, while labour shortages and stretched equipment lead times are slowing execution
- Development windows for the biggest data centre campuses have lengthened from five to seven years to seven to 10 years
- Power generation, transmission, grid modernisation and high-voltage equipment are seen as the choke points, and the opportunity
- Industrial demand is improving beyond AI, while construction is split between data centre and strategic mega-projects and a sluggish commercial and residential market
- Defence orders remain strong, especially in Europe, where the test is converting backlogs into deliveries; US spending faces budget, rate and political headwinds
The challenge facing the artificial intelligence buildout in the US has shifted from whether the demand is real to whether the infrastructure can be delivered fast enough, according to Goldman Sachs, after its Fundamental Equity investors attended a series of industrial and data centre power conferences across the country.
The bank’s investors said manufacturers are now taking orders that stretch into the early 2030s, turning AI-related power demand from an investment thesis into a multi-year backlog. With demand no longer seen as the open question, the pressure has moved to execution. Labour shortages and lengthening equipment lead times are acting as bottlenecks, and development timelines for the largest data centre campuses have extended from five to seven years to between seven and 10 years.
Goldman’s team framed that constraint as an opportunity as well as a risk. Investors identified power generation, transmission, grid modernisation and high-voltage equipment as the choke points most likely to determine the speed of the buildout.
Beyond AI, the broader industrial picture also appears to be improving. Company management teams reported healthy demand and customers newly willing to commit to longer-cycle projects. That raised the question of whether the US is seeing a genuine industrial recovery or simply a spillover from AI spending. Goldman’s view is that it is both, with strength showing up across manufacturing end markets well outside the data centre supply chain.
Construction remains a market of two speeds, the bank said. Mega-projects linked to data centres, semiconductors, life sciences and liquefied natural gas continue to support non-residential activity, while wider commercial and residential construction remains subdued. Capital is concentrating in strategic infrastructure while more traditional segments lag.
Defence demand is also robust, particularly in Europe, where spending trends continue to improve. In the US, however, investors are watching fiscal pressures, higher interest rates and a shifting political backdrop that could limit future spending growth. In Europe, the key issue is whether manufacturers can raise production quickly enough to turn growing backlogs into deliveries.
Goldman’s overall conclusion was less about a new theme than about stronger conviction in an existing one: AI is driving a multi-year buildout of power, grid and data centre infrastructure, and the harder part from here is building it.
This article was written by Eamonn Sheridan at investinglive.com.