The global pursuit of artificial intelligence has encountered a significant obstacle: the ability to rapidly connect data centers to the power grid. The state of Texas in the US is a prime example of this issue, with over 470 GW of connection requests being examined, which is more than five times the record peak demand of the ERCOT grid. While not all of these requests will result in data centers, authorities are working to identify speculative or unrealizable demands.
This new bottleneck follows previous challenges in the AI sector, including the scarcity of Graphics Processing Units (GPUs) and the electricity required to power data centers. According to the International Energy Agency (IEA), over 2,500 GW of renewable production, storage, and large load projects, including data centers, are currently waiting in connection queues worldwide. This situation highlights that an abundance of electricity does not necessarily guarantee that a data center can quickly obtain the necessary power.
The capacity of power grids, connection queues, and lead times can now determine when and where new AI infrastructure will be built. The IEA estimates that grid constraints could delay around 20% of the global data center capacity expected by 2030. As a result, the localization of digital investments may increasingly depend on territories that can combine electricity, grid capacity, cooling systems, land availability, and necessary permits.
The previous experience of electro-intensive industries supports this hypothesis, as their geographical location tends to be influenced by the availability of energy infrastructure. However, computing loads are more easily relocatable than those of heavy industries. The scarcity of grid connections may not be permanent, as reforms and better utilization of existing grids could enable the connection of 1,200 to 1,600 GW of advanced projects currently in waiting.
The situation in Texas and other regions highlights the need for efficient grid management and planning to support the growth of AI and data centers. With the increasing demand for AI infrastructure, governments and grid operators must work together to address these challenges and ensure that the necessary power and grid capacity are available to support this growth.
The impact of these challenges on the development and localization of AI infrastructure is significant, as it may lead to a new geography of AI, with data centers and AI investments concentrated in areas with favorable energy infrastructure. This, in turn, may have implications for the global distribution of AI capabilities and the competitiveness of different regions.
To mitigate these challenges, policymakers and industry stakeholders must collaborate to develop and implement effective solutions, including grid reforms, more efficient connection processes, and better utilization of existing grid capacity. By addressing these bottlenecks, it may be possible to unlock the full potential of AI and ensure that its benefits are realized globally.
Key points
- Grid constraints could delay around 20% of the global data center capacity expected by 2030.
- Over 2,500 GW of renewable production, storage, and large load projects are currently waiting in connection queues worldwide.
- Reforms and better grid utilization could enable the connection of 1,200 to 1,600 GW of advanced projects currently in waiting.
