Powering the AI Frontier: Why Energy Infrastructure and Power Purchase Agreements Are the New Battleground for Innovation

As artificial intelligence reshapes the technological landscape, a quiet but monumental shift is occurring beneath the surface: the race to secure enough power to support it. Behind every breakthrough in AI—whether it’s a frontier model or a real-time inference engine—lies an insatiable demand for electricity. And not just any electricity. We're talking about multi-gigawatt-scale facilities that require predictable, efficient, and often clean power contracts. In short, power purchase agreements (PPAs) and grid readiness are now at the heart of America’s AI leadership.

A recent policy blueprint from Anthropic, titled Build AI in America v3, underscores just how critical the energy dimension has become. The report lays out a national roadmap for building 50+ gigawatts (GW) of AI-related capacity by 2028, signaling that the intersection of energy, infrastructure, and data centers is no longer a side conversation—it’s the main event. Let’s explore what this means for power developers, AI companies, and investors alike.

AI is Driving Unprecedented Energy Demand

By 2028, the U.S. AI sector alone is expected to require at least 50 GW of electric capacity to stay competitive globally. Of that, between 20 and 25 GW will be needed for training the most advanced frontier models. For context, that's roughly equivalent to the peak demand of the entire state of New York in summer. The rest will power AI inference workloads, which, while lighter individually, add up quickly as AI applications spread across industries and geographies.

Training vs. Inference: Different Workloads, Different Grids

AI training requires massive, localized power delivery—think single data centers needing up to 5 GW, rivaling the output of entire power plants. In contrast, inference is more distributed, relying on smaller clusters of edge or regional data centers. Each use case has implications for how PPAs are structured, which transmission corridors are prioritized, and what types of generation are most viable.

Behind-the-Meter Is Not Enough

While co-locating generation with data centers (behind-the-meter power) can be appealing, the scale and reliability requirements of AI workloads make grid interconnection essential. That’s why structured PPAs, grid-enhancing technologies (GETs), and utility-scale interconnections are now prerequisites for large-scale AI infrastructure.

Federal Land Leasing Could Reshape the Power Market

Anthropic’s report encourages opening DOD and DOE lands for large-scale AI infrastructure development. This would allow developers to bypass local zoning barriers and expedite environmental permitting via streamlined NEPA reviews. For energy companies, this creates new opportunities for federal-site-linked PPAs, particularly in the western U.S. where solar and geothermal are abundant.

DOE Transmission Authorities May Override State Delays

One of the most compelling policy options involves using DOE’s Section 1222 and 40106 authorities to fast-track transmission lines that would otherwise face years of state-level delays. This is significant for developers and investors structuring PPAs with offtake from remote resources (like geothermal or large-scale solar), as it clears a critical bottleneck to commercialization.

Interconnection Is the New Bottleneck

On average, it takes 4–6 years for new power projects to receive interconnection approval. Anthropic proposes a series of utility-side reforms—including auctioning queue slots, allowing temporary demand caps, and AI-assisted resilience testing—to bring that down to 2–3 years. Projects tied to AI training centers may even receive “national security” prioritization.

PPAs Could Be Federally Mandated for National Security Projects

If necessary, the executive branch could invoke the Defense Production Act or Section 202(c) of the Federal Power Act to compel utilities to interconnect power projects tied to AI infrastructure. That opens the door to federally-backed PPAs with national security guarantees, offering a new class of bankable energy contracts.

“All of the Above” Strategy Encouraged for Power Sources

Rather than mandating specific generation types, the report supports a market-based, developer-led approach to procurement. Solar, batteries, natural gas, geothermal, and potentially advanced nuclear are all viable depending on region, economics, and buildout speed. This flexibility is critical for structuring PPAs that reflect real-world cost and timeline constraints.

Strategic Corridors and Grid Zones May Streamline PPAs

The creation of National Interest Electric Transmission Corridors (NIETCs) could provide pre-cleared pathways for transmission, enabling quicker PPA execution by reducing regulatory and permitting risks. These zones will likely align with emerging data center clusters—especially on federal land.

Transmission and Grid Financing Tools Are Growing

DOE already has $5.75 billion in lending authority for transmission and grid projects under Sections 402 and 40106. These tools can be coupled with PPAs to improve financing for developers. There's even a proposal for the DOE to sell debt instruments on its balance sheet to third parties, allowing for more lending without new congressional approvals.

The Bottom Line: Energy Strategy is Now AI Strategy

As Anthropic rightly notes, "Building the infrastructure required for AI leadership is no simple task." It requires more than semiconductors and software—it demands strategic energy planning, transmission capacity, and flexible PPAs. The companies that succeed in this space will be those who can bridge the gap between AI ambitions and energy execution.

Sources:

Anthropic, Build AI in America v3, July 2025.