Many AI applications are free to use (enterprises are the main consumers right now). If users balk at paying for them, the monetization rate of these offerings may weaken more than expected and demand for data centers could decline. Factoring in a 20% falloff in AI demand between 2025 and 2030, occupancy would run 8 percentage points less than the base case. This means there would be excess supply in data centers, potentially forcing operators to lower their lease rates.
While AI’s growth is making headlines, most of the capacity in data centers still supports run-of-the-mill cloud computing (cloud and traditional workloads currently make up about 85% of data center demand). In a scenario where corporate spending on these basic services declines, occupancy in data centers may fall 4 percentage points from the base case even though AI demand remains steady. Schneider says companies are also steadily trying to optimize their use of cloud services, which can also reduce demand.
Source: 451 Research - part of S&P Global Market Intelligence, Goldman Sachs Research
Source: 451 Research - part of S&P Global Market Intelligence, Goldman Sachs Research
As more data centers come online, Goldman Sachs Research’s base case is that the balance between supply and demand is set to narrow significantly over the next 18 months. Occupancy is predicted to remain at peak levels through 2026. Although visibility beyond 2027 is limited, our analysts expect supply constraints to ease beyond this point, with occupancy rates gradually falling back to around 90% by 2028 and then leveling out.
Source: 451 Research - part of S&P Global Market Intelligence, Goldman Sachs Research
AI videos, which are rolling out now, promise a much richer experience. Demand for computing capacity to support these data-hungry applications could jump significantly, Goldman Sachs Research forecasts.
A new generation of AI chips—graphics processing units, or GPUs—may also consume more power than expected. "The power efficiency is improving, but power demand is outstripping what a lot of people thought," Schneider says.
In this scenario, data center occupancy rates exceed 100% in peak regions by 2030 and come in 17 percentage points higher than the base case.
Source: 451 Research - part of S&P Global Market Intelligence, Goldman Sachs Research
Scenario 2: AI adoption falls short, resulting in excess supply of data centers
Scenario 3: Cloud computing drops in weakening economy
Scenario 1: AI surge overwhelms the supply of data centers
Base case: Demand for data centers comes close to exceeding supply
↓
↓
↓
↓
“What we’ve seen over the last nine months is that both supply and demand forecasts have gone up, but the demand side is increasing a bit more than the supply forecast.”
Jim Schneider
Goldman Sachs Research
Jim Schneider
Goldman Sachs Research
“If we get more multimedia models, that could significantly tighten the market.”
"If we see slowing demand for AI from end users, AI will have fewer monetization opportunities and likely less data center demand across the industry as a result."
Jim Schneider
Goldman Sachs Research
“If the economy slows a bit and companies decide they need to be a little tighter with the usage of cloud services and they optimize, then growth rates could decelerate.”
Jim Schneider
Goldman Sachs Research
There is a lot on the line. Five of the top US cloud computing giants are poised to almost triple their capital expenditures to an estimated $644 billion by 2027.
Goldman Sachs Research
"The time needed to start a data center could depend on whether you have to build a new power generation asset for it and the type of energy powering it. New solar and battery facilities can be up roughly in a year, while it may take five years for gas-turbine power."
Carly Davenport
Goldman Sachs Research
