Is There Enough Data Center Capacity for AI?

As excitement about artificial intelligence (AI) mounts, it’s been tough getting a handle on how fast the technology is growing and whether it will fulfill expectations. “A lot of investors have struggled with the hype and quantifying what this all means,” says Jim Schneider, a senior equity analyst in Goldman Sachs Research.

How much power are data centers forecast to consume?

Goldman Sachs Research has projected what may come next for this burgeoning industry. Our analysts forecast that AI’s slice of the overall data center market will double to 30% over the next two years, taking share from conventional and cloud workloads. By 2030, their base case is that overall power consumption from data centers will jump 175% from 2023 levels (our analysts' previous forecast was for an increase of 165%). 

Are companies building enough data center capacity? 

Occupancy, or “supply sufficiency,” shows how much of a data center’s computing capacity is being tapped. “It’s very important,” Schneider says. “Occupancy is tied to margins, and margins are tied to earnings, and earnings are ultimately tied to the stocks of players in the space.” 

Goldman Sachs Research modeled a series of scenarios—from AI overwhelming the supply of data centers to a slowing economy and weaker monetization—to show how the data center boom may play out: 

The outlook for data center capacity

Goldman Sachs Research's base case is that the industry builds enough data center capacity to absorb increasing demand, with occupancy peaking at around 93% next year. 

Should the occupancy rate slacken, whether from a slowing economy or an inability to monetize AI models, it will get harder for data center operators to generate the expected returns on their capital investments. Data center capacity drops to around 80% by the end of the decade in Goldman Sachs Research's "AI Downside" scenario.