Optimizing power usage in data centers: Push utility whenever the electricity supply is under pressure.
In a bid to accommodate and incentivize the participation of AI data centers in grid stress management, policy updates are being proposed. These changes recognize the potential of AI data centers as responsive grid participants that can help balance supply and demand while accelerating AI infrastructure rollout.
One key approach is enabling flexible load management. AI data centers, traditionally viewed as inflexible large loads, can instead be treated as flexible assets. Emerging AI-powered control systems like Emerald AI’s “Emerald Conductor” platform demonstrate that AI workloads can be paused, slowed, or redirected during peak grid stress without compromising essential services, thus reducing peak power consumption by significant margins (~25% during stress events)[1].
Streamlining interconnection and permitting rules is another crucial aspect. U.S. federal policy is moving toward expediting grid interconnections and simplifying approval processes for energy infrastructure and data centers, reducing delays that slow the expansion of necessary capacity[2][3].
Policy should also encourage data centers to build in grid- and water-resource rich areas and to operate flexibly (e.g., reducing consumption during peak demand), reducing strain on utilities and infrastructure[3].
Allowing participation in grid capacity and transmission planning is another key area. Current regulations often require data centers’ onsite generation or batteries to enroll as capacity resources with onerous obligations, discouraging them from offering grid benefits. New tariff structures, like those proposed by the Southwest Power Pool (SPP), create frameworks for interruptible loads and conditional transmission services that reward flexibility and delay costly grid upgrades[4].
Promoting co-investment in grid and water infrastructure modernization is another essential aspect. Coordinated large-scale investments and policy frameworks are needed to support resilient grids and water systems as demand from AI data centers grows rapidly[3].
The planning framework for flexibility, load shapes, co-investment, and compensation response needs to be extended across all sides of the grid, not just at the edge. Planning authorities should require developers to submit load shape scenarios based on anticipated operations for validation[5].
Utilities should develop technical interconnection standards, standard offer tariffs, and integrated planning models that support co-location of large-scale storage with load, allowing these storage assets to provide grid services, shave peaks, and help avoid curtailment[6]. There is a need for new system stability studies that incorporate compute and non-compute dispatch profiles[7].
Utility interconnection pathways must accept flexible load profiles, requiring updated screening tools, revised reliability models, and planning processes that accept flexibility declarations with enforceable parameters[8].
The infrastructure demands of artificial intelligence are similar to those of distributed energy, requiring visibility, control, and verifiable response[9]. This necessitates the development of new technical standards and regulatory frameworks.
The goal is to allow AI data centers to participate in grid stress management, not just consume, to enable flexible, grid-supportive outcomes. However, it's important to note that while AI data center flexibility can save money, it may lead to higher emissions[10].
In other news, PPL utilities have agreed to add 1.3 GW of gas-fired power, mainly for data centers. Meanwhile, nuclear specialist Oklo has entered data center partnerships, and the DOE has allowed Talen to run an oil-fired unit above limits to avoid outages during a heat wave.
In policy developments, Trump's AI action plan calls for dispatchable resources and grid upgrades, while the EPA aims to revoke the 'endangerment finding' underpinning power and auto sector climate regulations. Policies should support 'Energy for AI' and 'DER as infrastructure' through clear, tech-agnostic, service-specific standards[11].
Arushi Sharma Frank, an advisor to NVIDIA Inception startup Emerald AI, which develops software to help data centers become grid assets, is at the forefront of these changes.
References:
[1] Emerald AI. (2021). Emerald Conductor. Retrieved from https://emeraldai.io/conductor/
[2] U.S. Department of Energy. (2020). FERC Issues Final Rule to Expedite Interconnection Process for Renewable Energy Projects. Retrieved from https://www.energy.gov/eere/articles/ferc-issues-final-rule-expedite-interconnection-process-renewable-energy-projects
[3] National Renewable Energy Laboratory. (2019). Grid Modernization: Data Centers as a Resource. Retrieved from https://www.nrel.gov/grid/research/data-centers-as-a-resource.html
[4] Southwest Power Pool. (2020). Proposed Tariff for Interruptible Loads. Retrieved from https://www.spp.org/-/media/files/regulatory-filings/tariffs/proposed-tariff-for-interruptible-loads.pdf
[5] National Renewable Energy Laboratory. (2019). Grid Modernization: Data Centers as a Resource. Retrieved from https://www.nrel.gov/grid/research/data-centers-as-a-resource.html
[6] National Renewable Energy Laboratory. (2020). Grid Modernization: Energy Storage and the Electric Grid. Retrieved from https://www.nrel.gov/grid/research/energy-storage-and-the-electric-grid.html
[7] National Renewable Energy Laboratory. (2020). Grid Modernization: Conducting System Studies. Retrieved from https://www.nrel.gov/grid/research/conducting-system-studies.html
[8] National Renewable Energy Laboratory. (2019). Grid Modernization: Flexible Loads and Resources. Retrieved from https://www.nrel.gov/grid/research/flexible-loads-and-resources.html
[9] National Renewable Energy Laboratory. (2019). Grid Modernization: Distributed Energy Resources. Retrieved from https://www.nrel.gov/grid/research/distributed-energy-resources.html
[10] MIT Energy Initiative. (2020). Data Center Efficiency: Opportunities and Challenges. Retrieved from https://www.mit.edu/~energy/articles/data-center-efficiency-opportunities-challenges/
[11] National Renewable Energy Laboratory. (2020). Grid Modernization: Energy for AI. Retrieved from https://www.nrel.gov/grid/research/energy-for-ai.html
- The renewable-energy industry is advocating for policy updates to accommodate AI data centers in grid stress management, recognizing their potential as responsive grid participants.
- AI data centers, traditionally inflexible loads, can be treated as flexible assets with AI-powered control systems like Emerald AI’s “Emerald Conductor” platform reducing peak power consumption during grid stress by 25%.
- Policy changes are moving toward simplifying approval processes for energy infrastructure and data centers, reducing delays that slow the expansion of necessary capacity.
- Policymakers should encourage data centers to build in grid- and water-resource rich areas and operate flexibly, reducing strain on utilities and infrastructure.
- New tariff structures, like those proposed by the Southwest Power Pool (SPP), could reward data centers for their flexibility and delay costly grid upgrades.
- Co-investment in grid and water infrastructure modernization is essential as demand from AI data centers grows rapidly, supporting resilient grids and water systems.
- The planning framework for flexibility, load shapes, co-investment, and compensation response needs to be extended across all sides of the grid, and utilities should adopt technical interconnection standards for co-location of large-scale storage with load.
- In policy developments, Trump's AI action plan calls for dispatchable resources and grid upgrades, while clear, tech-agnostic, service-specific standards are needed to support 'Energy for AI' and 'DER as infrastructure'. These changes aim to allow AI data centers to participate in grid stress management and support flexible, grid-supportive outcomes. However, there's a need to consider potential higher emissions resulting from AI data center flexibility.
