America’s Data Center Capacity Shortfall
The United States is entering a critical period for digital infrastructure. Between 2023 and 2028, demand for data center capacity is projected to grow far faster than the ability to supply power, land, and physical facilities. Driven largely by artificial intelligence (AI), cloud computing, and hyperscaler expansion, this imbalance is creating a persistent and historically large shortfall in U.S. data center capacity.
Estimates from Goldman Sachs, visualized by Visual Capitalist, show that despite record capital spending by major technology firms, power and infrastructure constraints will keep supply consistently below demand through at least 2028.
Key Takeaways
- AI-driven demand surge: Rapid growth in power-hungry AI training and inferencing workloads is pushing U.S. data center demand beyond available capacity through 2028.
- A widening power gap: By 2028, the projected shortfall of 10 gigawatts (GW) is equivalent to the electricity required to power roughly 7.5 million U.S. homes for a year.
- Massive capital investment: Hyperscalers such as Google, Meta, and Amazon are expected to spend a combined $325 billion on capital expenditures in a single year, largely focused on data centers—yet physical constraints remain a bottleneck.
Supply vs. Demand: A Persistent Gap
Despite aggressive expansion plans, U.S. data center supply has been unable to keep pace with soaring demand. The imbalance is not a short-term anomaly but a multi-year structural issue.
U.S. Data Center Capacity Supply and Demand (GW)
| Year | Demand (GW) | Supply (GW) | Shortfall (GW) |
|---|---|---|---|
| 2023 | 24.0 | 14.2 | -9.8 |
| 2024 | 29.7 | 21.1 | -8.6 |
| 2025P | 38.1 | 26.7 | -11.4 |
| 2026P | 49.8 | 40.5 | -9.3 |
| 2027P | 62.7 | 53.0 | -9.7 |
| 2028P | 77.0 | 67.0 | -10.0 |
While supply rises steadily, demand accelerates even faster—particularly after 2025—leaving a consistent gap of roughly 9–11 GW every year.
For context, a single large-scale data center can consume as much electricity annually as 400,000 electric vehicles, underscoring how quickly demand compounds at scale.
What’s Driving the Shortage?
1. AI and “Inferencing” Demand
While early AI growth was dominated by energy-intensive model training, the next phase is inferencing—when AI models respond to user queries in real time. Inferencing workloads must be fast, reliable, and geographically close to users, increasing the need for dense, urban-adjacent data centers with high power availability.
2. Limited Land and Grid Capacity
Locating data centers closer to population centers improves latency, but suitable land with access to high-voltage transmission is increasingly scarce. In many major U.S. markets, power grids are already constrained, and utilities face long timelines to approve and build new capacity.
3. Power Generation Bottlenecks
Natural-gas turbines—often the preferred solution for reliable, on-site or near-site generation—can face multi-year wait times due to manufacturing backlogs and permitting delays. As a result, tech companies are turning to smaller, more readily available gas turbines or alternative energy solutions, often at higher cost.
4. Supply Chain and Equipment Constraints
Ongoing shortages in specialized chips, electrical equipment, and skilled labor continue to slow construction timelines. Even when funding is abundant, physical delivery remains a limiting factor.
Record-Low Vacancy Rates Signal Tight Conditions
Data center vacancy rates in the U.S. have fallen to around 3%, an all-time low. This indicates that nearly all existing capacity is already spoken for, leaving little flexibility to absorb sudden demand spikes—especially from AI-driven workloads that can scale rapidly.
Looking Ahead to 2028
Even with unprecedented investment, the U.S. data center market is expected to remain undersupplied through 2028. The shortfall is forecast to narrow slightly in some years but ultimately returns to around 10 GW by 2028, highlighting the structural nature of the challenge.
Closing this gap will require:
- Faster grid expansion and interconnection approvals
- New approaches to distributed and on-site power generation
- Continued innovation in energy efficiency and cooling technologies
- Strategic siting of data centers beyond traditional urban hubs
Until these constraints ease, data center capacity—and especially power availability—will remain one of the most critical chokepoints in the growth of AI and the digital economy in the United States.
Source: Goldman Sachs; visualization by Visual Capitalist. Figures rounded.