Goldman Sachs: AI infrastructure delivery now the main challenge

On equities, the preference from Goldman is for the choke point over the consumer: firms offering power generation, transmission, grid updates and high-voltage gear maintain pricing leverage as capacity remains tight. Extended order books provide these companies with exceptional revenue clarity, yet delivery delays now pose the greatest threat to profit realisation. An expanding industrial upturn suggests the sector should not be viewed solely as an AI-related play, whereas construction links to commercial and residential property continue to appear the less favourable part of the trade. In the defence space, European producers that can boost output are in the strongest position, while US contractors contend with more budget- and rate-driven challenges.

---

According to Goldman Sachs, AI power demand is no longer in doubt; the true challenge is whether the US can construct power stations, grid infrastructure and data centres at a pace sufficient to match it.

In summary:

• Goldman Sachs' Fundamental Equity investors attended US industrial and data centre power conferences and determined that AI's primary constraint has shifted to delivery from demand.
• Manufacturers are booking orders stretching into the early 2030s, while labour shortfalls and extended equipment lead times are impeding execution.
• Development timelines for the largest data centre campuses have expanded from five-to-seven years to seven-to-10 years.
• Power generation, transmission, grid modernisation and high-voltage equipment are considered the chokepoints, and also the opportunity.
• Industrial demand is improving beyond the AI sector, while construction is divided between data centre and strategic mega-projects and a slow commercial and residential market.
• Defence orders stay strong, particularly in Europe, where the challenge lies in converting backlogs into deliveries; US spending faces budget, interest rate and political obstacles.

The AI infrastructure buildout in the US has moved from questioning the authenticity of demand to whether the infrastructure can be supplied at sufficient pace, Goldman Sachs reports, following its Fundamental Equity investors' attendance at multiple industrial and data centre power conferences nationwide.

Goldman's investors noted that manufacturers are accepting orders extending into the early 2030s, converting AI-related power demand from a concept for investment into a multi-year order book. With demand no longer the unresolved issue, the focus has shifted to execution. Worker shortages and growing equipment lead times are creating bottlenecks, while development periods for the biggest data centre campuses have increased from five to seven years to seven to 10 years.

The Goldman team characterised that limitation as both an opportunity and a risk. Power generation, transmission, grid upgrades and high-voltage gear were singled out by investors as the chokepoints that will most likely set the pace of the construction.

Outside the AI sphere, the wider industrial landscape also seems to be brightening. Corporate management teams indicated solid demand and a new willingness among clients to engage in lengthier project cycles. This prompted the question of whether the US is experiencing a real industrial turnaround or merely a knock-on effect from AI expenditure. Goldman's assessment is that it is a mix of both, with vigour appearing across manufacturing end markets far beyond the data centre supply chain.

The construction sector is a two-speed market, according to the bank. Large-scale projects tied to data centres, semiconductors, life sciences and LNG are still bolstering non-residential activity, while broader commercial and residential construction stays weak. Investment is clustering in strategic infrastructure as more conventional areas fall behind.

Defence demand is also strong, especially in Europe, where expenditure patterns keep improving. In the US, though, investors are monitoring fiscal constraints, elevated interest rates and a changing political environment that might restrict future budget increases. In Europe, the central question is if manufacturers can ramp up output fast enough to convert expanding backlogs into actual shipments.

Goldman's final takeaway was not so much a fresh idea as a firmer belief in an established one: AI is fuelling a multi-year expansion of power, grid and data centre infrastructure, and the more difficult task moving forward is its construction.