How Effective Government Policy Can Further Extend The US Lead In Data Center Development

The Race to AI Infrastructure

"The U.S. is in a race to achieve global dominance in artificial intelligence (AI)." So begins the opening sentence of "Winning the Race: America's AI Action Plan," the AI strategy released by the White House in July 2025. So far, we are winning; of the 10GW AI-driven data center capacity installed by 2024, ~80% was concentrated in the U.S.¹ Our success thus far has been driven by our competitive advantage for hyperscaler deployments compared to alternative markets. The U.S. is the wealthiest free market in the world, with a clear investment thesis, scores highest on ease of doing business for hyperscale deployments, and has an abundance of land, access to cheap power, and an innovative talent pool. Attracted by these advantages, hyperscalers have chosen the U.S. as the primary hub of data center deployments, housing 40% of global installed base of hyperscale IT GWs. But the AI race challenges this paradigm, with every subsequent GW more difficult to deploy in power-constrained hubs, an increasing number of sovereigns committing to their own expansive AI roadmaps, and emerging global markets attracting hyperscalers with faster time to market and more available power. Keeping the U.S. ahead of the game has never been more challenging or more critical.

Where We Are Today

The business case for financing data center deployments is strong. Hyperscalers have each committed ~$100 billion² in just 2025 toward deployments, and third-party data center operators/developers are stepping up their investments as well. Together with PE, infra funds, and other investors, developers have raised over $130 billion³ in capital commitments toward U.S. deployments. While the sums committed are promising and astonishing, many of these projects are three to five (or more) years away from actualization due to various bottlenecks that are overwhelming the private sector's best efforts at fast deployment.

A&M Insights – Key Bottlenecks

Power generation bottlenecks: Data centers are power hungry, and the Tier 1 and 2 are running out of available capacity. Pockets of excess power generation exist, but they are 1) outside of typical deployment markets, 2) hard to identify, and 3) could have long interconnection queues.

Grid congestion: Even in markets where excess power capacity is available, new data centers must line up on the "interconnection queue" — sometimes behind stale or low priority projects — before getting connected to the grid, a process that can now take as long as five years or more.⁴

Still, our analysis points to pockets of excess power and interconnection opportunities existing, though they are hard to identify.

What about connectivity? Isn't that a critical bottleneck? There is a lot of discussion around fiber readiness to bolster data center infrastructure. While this is certainly important and necessary, we would argue that this does not contribute as much to the short-term data center deployment bottleneck. Our analysis⁵ points to existing fiber runs being available within less than five miles of approximately 80% of future desirable data center locations in the U.S.

This distance is well within most network operators' attractive payback distance bands, from an economics perspective. New infrastructure and connectivity corridors will indeed need to be purpose-built for the eventual facility specifications, but current solutions can serve as temporary "bridges" in the interim, so this is not as much of a gating criterion for delaying deployments outside of a few very remote locations.

Lack of ecosystem coordination: Beyond the daunting power bottlenecks, various ecosystem players could be collaborating more proactively and transparently to unlock faster times to market but are instead either unaware of their ability to benefit from this AI ecosystem participation, or are working in information/coordination silos.

Key AI Data Center Ecosystem Stakeholder Silos

A&M Insights – How the Public Sector Can Help

There are several ways the race to ready can be further accelerated by state and federal actions beyond the efforts already underway:

Federal

• Consider further reforms to the interconnection queue process:
  
  We would suggest considering:
  • Defining new priorities for RTO and ISO project prioritization criteria for new projects coming into the queue
  • Allowing credible AI developments already in the queue to jump the existing queue
  • Creating a system for automatically withdrawing stale or infeasible projects
• Creation of ecosystem portal: Consider working with ecosystem players to create a portal for municipalities and utility companies to interface with hyperscalers and developers and accelerate the "powered land" process. The portal could serve three purposes:
  • Educating local utilities and municipalities on the advantages and rules of considering data centers for their communities
  • Streamlining local/state and federal-level permitting
  • Creating (to the extent possible) a uniform and transparent process flow and communication platform for all parties involved
• Consider further transparency and private industry engagement in the federal land use selection process: The "Accelerating Federal Permitting of Data Center Infrastructure" executive order passed in July 2025 currently directs the DOD and DOE to identify top data center development locations and well as consider easements on federal land to enable utility infrastructure shortcuts. While this is a great first step, we would advocate for going a step further and allowing credible developers and utilities transparency into potentially developable land for purposes of either data center builds or utility infrastructure easements. Access to a geospatial, nationwide fact-base of possibly eligible federal lands allows developers to apply their sophisticated criteria know-how and to work handin-hand with utilities to explore the art of the possible and identify ideal development sites. The federal government can prescreen and filter out any lands that would fall into a definite "cannot be used" bucket and stipulate that any shortlist of sites piquing the interest of eligible private parties is subject to federal government approval and terms.
• Consider reforming the federal grid upgrade grants and funding process: While the federal government already issues grants for electrical grid resiliency and enhancement projects, the process is bottom-up and reactive. We would suggest considering a move toward a model where the government is proactively predicting evolving data center hot spots and incentivizing targeted grid investments for specific projects and markets. This approach enables the government to stay ahead of bottlenecks and apply funding toward where it may be most valuable from an AI bottleneck perspective, rather than letting utilities reactively apply for funding without priorities in mind after bottlenecks already form.

State

• Creating and communicating an AI infrastructure agenda and roadmap for your state, inclusive of attractive hyperscaler incentives: States may benefit from articulating a clear AI infrastructure action plan and agenda for all the participants in the data center ecosystem (municipalities, utilities, hyperscalers, developers, suppliers). In our work we've found that hyperscale ecosystems develop from pull as much as push factors, and certain seemingly unlikely locations are emerging as data center hubs due to state efforts to attract developers/hyperscalers. While hyperscalers may prefer to continue deploying in existing hubs, time to market is the key criterion in the AI race. Alternative markets with relatively cheap and available power, such as emerging markets of PA and LA can serve neighboring population centers within 3ms latency bands (~200 miles) and are rising up to meet hyperscaler needs. States can further incentivize deployments by creating attractive data center operating regimes, offering subsidies, and deregulating around nuclear developments. AI data center operations are expensive — from the cost of upfront builds, to expensive lease costs for outsourced facilities, to onerous electricity costs at $600+ million of annual electricity costs per GW deployed. Creating attractive total cost of ownership conditions that are comparable or better than alternative markets can weigh the scales toward deploying in undeveloped states.
• Consider streamlining the ISO grid interconnection process at the state level and providing funding toward strategic infrastructure projects: States also have significant leverage into the status of their interconnection queues. From providing funding and resources to accelerate project deployment, to reforming the process by which states evaluate and plan for large loads, states can enable utilities to evaluate and plan for projects quickly and fairly. Texas serves as a recent example of a state working to alleviate transmission grid bottlenecks and accelerate time to market.⁹
• Considering working with local governments and utilities to take an inventory of existing infrastructure, gauge community interest, and alleviate zoning and permitting delays: As part of their roadmap, interested states could work with local governments and utilities to: 1) educate them on the benefits of inviting AI data center into their communities, 2) inventory the current state of generation and transmission infrastructure and prioritize investments, 3) connect interested communities/utilities with hyperscalers and credible developers to signal interest, and 4) help local governments think through an accelerated local permitting and zoning roadmap to accelerate time to market.
• Consider opening state lands for data center deployments and/or infrastructure easements: Similarly to the federal efforts, state governments could consider opening up state-owned land to data center developments or infrastructure easements.
• Consider crafting private/public partnership models (P3s) that benefit all players: As part of their roadmaps, states could consider P3 investment models to help fund large-scale infrastructure projects for utilities upgrades, as well as share in some of the economic lease benefits of data center facility developments.
• Investing in behind-the-meter power "bridges": Solutions such as natural gas, SMRs,¹⁰ fuel cells, and BESS¹¹-enabled renewables can serve as behindthe-meter "bridges" for power in the shorter term, as data center sites await interconnection and utility infrastructure development. The state and federal governments can also consider fast tracking these "bridge" solutions with tax credits, deregulation, and supply chain enablers.

Private

Beyond state and federal efforts, the private sector (utilities, suppliers, developers, hyperscalers) can also help lubricate the coordination of AI infrastructure efforts across the ecosystem through the following efforts:

• Creating consortia to pool highly fragmented groups such as rural utilities or regional suppliers
• Convening around potential collective investment return/P3 models for government land, infrastructure development, and other potential partnerships
• Partnering with government to create a transparent and comprehensive stakeholder experience (through efforts such as a comprehensive ecosystem portal)
• Setting standards and processes for separating speculative and unfeasible developments from credible ones to help alleviate the project diligence congestion apparent at multiple levels of the ecosystem

Footnotes

1. A&M Global Data Center Demand Model. Please contact A&M directly for further details.

2. Amazon, Google, Microsoft, and Meta are committing to more than $320 billion in 2025, and OpenAI and Softbank have announced plans for investments of $500 billion over the next four years.

3. Vantage Data Centers, Switch, Equinix, Aligned Data Centers, Compass Datacenters, CyrusOne, Digital Realty, Stack, CloudHQ, DataBank, QTS Realty, Cologix, Flexential, Sabey Data Centers, CoreSite, and Evoque press releases; CBRE; Cushman and Wakefield.

4. Joseph Rand et al., "Queued Up: 2024 Edition, Characteristics of Power Plants Seeking Transmission Interconnection as of the End of 2023," Berkeley Lab, April 2024 https://emp.lbl.gov/publications/queued-2024-edition-characteristics.

5. A&M Research and Analysis. Please contact A&M directly for further details.

6. Regional Transmission Organization: An independent organization that manages the high-voltage electrical grid across a broad area.

7. Independent System Operator: An independent organization that manages the electrical grid, usually within a single state.

8. FERC Order No. 2023, issued in July 2023, aims to clear grid connection backlogs by prioritizing commercially ready projects and requiring them to be studied in clusters rather than in a simple queue. See Federal Energy Regulatory Commission, "Explainer on the Interconnection Final Rule," https://www.ferc.gov/explainer-interconnection-final-rule.

9. Case, Nathan, "Why Does It Take So Long to Connect a Data Center to the Grid?" Camus Energy, July 1, 2025, https://www.camus.energy/blog/why-doesit-take-so-long-to-connect-a-data-center-to-the-grid.

10. Small Modular Reactors: Advanced, factory-fabricated nuclear reactors that are much smaller in size and power output than traditional nuclear power plants.

11. Battery Energy Storage System: A technology that captures energy from the grid or a power source and stores it in rechargeable batteries for later use.

Originally published 14 October 2025

The content of this article is intended to provide a general guide to the subject matter. Specialist advice should be sought about your specific circumstances.