top of page
  • Writer's pictureDevika Kumar

Powering the Future: Investment and Infrastructure Strategies for the AI and Data Center Boom

Note this post is repost and originally appeared as a LinkedIn Article here.

Increased AI adoption and investments, driven by Gen AI, are expected to significantly increase electricity demand for data centers, from ~2.5% of total U.S. electricity consumption in 2022 to ~7.5% by 2030.

Boston Consulting Group (BCG) and Keyframe recently hosted a panel to discuss the opportunities and path forward.

Access to reliable and affordable energy is the biggest challenge for data centers. Given that, we discussed 5 areas of opportunities (including infrastructure and investment solutions):

1) Opportunities within data centers to optimize location and timing of energy loads

Data center owners are trying to geographically diversify their locations. This helps manage the interconnection queues and grid interconnection challenges for fastest grid connectivity as well as helps balance overall energy footprint and operational efficiency given regional differences.

Potential to optimize energy use by separating workloads between model training (more energy intensive and less frequent) and inference functions. Model training functions can be scheduled across different locations based on energy availability, shifting workloads to regions with lower electricity demand or access to green energy.

Using different data center locations and approaches for internal and external workloads (e.g., Google's AI tools vs client services) can further optimize energy use. Internal AI model training can be scheduled during non-peak hours, reducing energy demand and infrastructure requirements without affecting client service reliability.

2) Opportunities for utilities to improve utilization of existing infrastructure and add cleaner baseload energy sources to satisfy load growth

Interconnection queue in the US reached ~2.6 TW last year, double the size of the existing grid. At the same time, average T&D infrastructure capacity utilization is ~30%. Edge AI technology can improve energy demand predictions and enhance grid operators' visibility into energy needs, enabling more efficient use of existing grid infrastructure (above the standard 30%) and freeing up untapped capacity.

Today data centers rely on gas power systems for reliable electricity supply, however investments in new sources of baseload generation such as battery storage, geothermal, and nuclear SMR have potential to reduce the energy footprint of data centers.

3) Opportunities for regulators to incentivize faster grid interconnections and demand side flexibility through policy

Regulators have an opportunity to manage the capital outlay for new infrastructure and grid interconnection with utilities’ risk appetite. Recently, AEP Ohio proposed to change their tariff structure to require that large data centers assume risk for new transmission infrastructure upgrades by agreeing to 10-year service contracts. This helps balance meeting the data center energy requirements while ensuring that costs of new infrastructure are shared fairly among utilities and data center owners with limited impact on other rate payers.

Regulators can incentivize grid flexibility with policies that value demand-side flexibility equally with generation. This will encourage utilities and data centers to invest in solutions for optimizing energy demand.

4) Potential technical innovations to reduce the energy footprint of data centers

Exciting developments in chip infrastructure, including photonics and optical networks, show significant potential for increased energy efficiency within data centers’ digital infrastructure.

Advances in energy management systems are driving improvements in data center building performance, as are upgrades from HVAC to advanced liquid cooling systems.

AI-powered optimization of energy assets such as solar panels and battery storage systems can also help optimize energy consumption and share resources more efficiently.

5) Large-scale financial commitments and partnerships required to support this growth

Significant longer-term financial commitments are needed to fund growth and help de-risk uncertainties within energy markets- i.e., energy prices and grid reliability. Data center owners will need to lead by example by entering into longer financial contracts related to infrastructure investments and green energy access (e.g., PPAs).

Large-scale commitments such as Microsoft’s $10 billion agreement with Brookfield Asset Management to develop 10.5 GW of renewable generation assets between 2026 and 2030 is one such example. Higher collaboration and partnership across the value chain expected in the future.

Thanks to our panelists and organizers Matthew Sundberg John Rapaport Marissa Hummon Rajeev Oak Athanasios Caramanolis, Khushboo Goel and Devika Kumar for their insights and participation.


kf block.png
bottom of page