Investing in the AI Infrastructure Buildout

Key Concepts:

• AI Infrastructure Investment
• Value Chain Approach
• Data Centers
• Power Access Bottleneck
• Large-Scale Energy Campuses
• Power Generation Capacity
• Renewable Power Sources (Solar, Storage)
• Natural Gas as a Solution
• Integrated City/Campus Model
• Reliability (99.999%)
• Capital Investment
• Demand for Compute Power
• Cost-Effectiveness
• Timely Delivery

AI Infrastructure Investment Strategy

The speaker, representing Carlyle as an infrastructure investor, outlines a "piece approach" to investing in AI infrastructure, emphasizing a long-term time horizon. They view AI infrastructure as a significant investment opportunity.

Comprehensive Value Chain Approach

Carlyle's strategy is not to develop AI infrastructure assets in isolation. Instead, they adopt a much more comprehensive view, examining the entire value chain for AI infrastructure. This involves developing data centers and simultaneously addressing critical bottlenecks, such as access to power.

Addressing the Power Bottleneck: Large-Scale Energy Campuses

A major challenge in data center development is access to power, a point highlighted by Elon Musk's comments on AI and energy. Carlyle's approach to developing AI infrastructure involves building large-scale energy campuses. These campuses integrate power generation capacity directly alongside data center capacity.

Copia Power: A Platform for Integrated Campuses

Copia Power, a company formed by Carlyle, is central to this strategy. Launched in 2021, Copia is building out a platform of these integrated campuses. The scale of these projects is described as extraordinary.

Energy Mix: Beyond Renewables

While the initial focus in 2021 was on sustainable infrastructure and renewable power sources, the current reality necessitates a more diversified energy approach. The speaker emphasizes an "all of the above" strategy for energy. This includes solar and storage, but also recognizes natural gas as a crucial part of the solution.

The Campus Model Explained

A campus, as defined by Copia, is akin to a "mini city" with multiple large-scale facilities. This includes gas and storage capacity connected to the grid. Adjacent to this power generation capacity are hyperscale data centers, also connected to the grid. The goal is not to build isolated "islands" but rather a fully integrated campus.

Benefits of the Integrated Campus Model

This integrated approach ensures that data centers gain access to power in a more timely manner, which is critical given the importance of timing in this sector. Furthermore, the power generated is cost-effective, as cost and economics are significant factors. Crucially, this model also provides more reliable power, essential for achieving the desired 99.999% reliability for long-lived infrastructure assets.

Case Study: Arizona Campus

As an example, Copia is building a site in Arizona that is approximately twice the size of Manhattan. Once fully developed, this project represents a $20 billion capital investment, encompassing both power generation assets and data center assets. Copia has plans for five additional campuses in the West.

Key Considerations for Campus Location

The strategic placement of these campuses is vital. They need to be located close to areas with high demand for compute power. Additionally, the projects must be cost-effective, delivered on time, and ensure reliable power supply.

Conclusion

Carlyle's investment strategy in AI infrastructure is characterized by a long-term, comprehensive approach that addresses critical bottlenecks like power access. Through Copia Power, they are developing large-scale, integrated energy campuses that combine diverse power generation sources (including renewables and natural gas) with hyperscale data centers. This model aims to deliver timely, cost-effective, and highly reliable power, exemplified by the significant capital investment in projects like the Arizona campus. The focus is on building integrated ecosystems rather than isolated assets to meet the growing demand for compute power.