The AI Data Center Boom Might Be Cooling Off: Goldman Sachs Sounds the Alarm

The AI data center gold rush might be peaking sooner than expected. Recent signals—from DeepSeek’s ultra-low-cost ChatGPT competitor to Microsoft scaling back AI data center projects, and Alibaba’s Joseph Tsai warning of an AI infrastructure bubble—suggest the frenzy is slowing.

Now, Goldman Sachs has adjusted its forecasts, moving up the expected peak in global data center utilization from late 2026 to 2025, warning that the boom may be coming to an end. What’s driving this shift?

The DeepSeek Effect: "Do More With Less"

One major factor is the rise of highly efficient large language models (LLMs) like DeepSeek. As ZeroHedge pointed out, the market is shifting toward “doing more with less”—smaller, optimized models that reduce the need for massive compute power.

The Shift in Demand and Supply

On the demand side, Goldman has revised downward its forecast for AI-related data center demand, particularly for 2025 and 2026. The slowdown in AI training demand, along with the broader adoption of AI inference workloads, will likely slow things down. However, Goldman still expects AI infrastructure demand to remain strong in the long run, with peak demand expected in 2027 and beyond.

On the supply side, Goldman updated its models to reflect the actual supply coming online in late 2024 and smaller, previously untracked data center operators. These adjustments have led to a 2GW upward revision in supply, with long-term supply expected to rise 8% by 2030 due to confirmed projects.

Key Risks and Uncertainties Ahead

Goldman flagged three major risks to the AI data center market:

Weak monetization of consumer-facing AI services: The ability of AI services targeting consumers to generate substantial revenue remains uncertain.
Oversupply of AI infrastructure: Large-scale AI infrastructure projects (like OpenAI’s Stargate) could lead to an oversupply of capacity, resulting in underutilized data centers.
Efficiency gains from smaller LLMs: The rise of smaller language models for enterprise use could reduce the need for large, resource-hungry data centers.

These factors, combined with the recent slowdown in AI training server shipments, suggest that the pace of growth for data centers might not be as explosive as previously anticipated.

For Investors

Despite the adjustments, Goldman remains optimistic about companies like Digital Realty (DLR) and Equinix (EQIX), which they believe will continue to perform well even as AI demand stabilizes. These companies are well-positioned to weather the cooling trend, thanks to their diversified customer base and solid operational frameworks.

In a quick survey conducted by Goldman Sachs, 25% of respondents identified "efficiency gains" as the biggest challenge for AI themes in 2025. This underscores the growing importance of efficiency in the tech industry. Investors should be mindful of the shifting dynamics and the potential for overcapacity in the data center market. While the tech sector is known for its rapid changes, the current trend suggests a need for a more measured approach to investments in AI infrastructure.

My Take

The AI data center boom was bound to slow. With rising efficiency, unclear monetization, and ballooning supply, the industry was overdue for a correction. The AI revolution isn’t over, but the "build at all costs" mentality may be.

What’s the bottom line? AI-driven data center growth isn’t gone, but it’s unlikely to continue at the breakneck pace of the past few years. With more efficient models coming into play and demand leveling out, the market is recalibrating. AI infrastructure investment isn't the "easy win" it once seemed, but there are still opportunities for those willing to adjust expectations.

For long-term investors, understanding the evolving demand for AI services and staying on top of efficiency trends will be key. It’s a good time to reassess positions—what was once high-growth, high-risk could now turn into more stable, moderate-return opportunities.

What do you think? #goldmansachs