Powered by What? AI, Capital, and the Clean Energy Question

It is hard to ignore conversations around power and data centers if you’re someone interested in the energy sector. Even if you’re not living and breathing energy infrastructure, you have probably come across the recent wave of articles on data centers being built to support AI development in the US.  

There is a real surge in power demand, and the grid is not ready for it. That led me to reflect on the long-term consequences and political choices shaping them. Capital committed today determines the grid carbon intensity of AI infrastructure for a generation. And 2026–2028 is exactly when that capital is being deployed, at speed and at scale. 

My immediate reaction, because I’m human and I have political bias, was that the administration is causing irreparable damage and setting back the energy transition in ways that will take years to unwind. But that’s too simple. Let’s dig deeper.  

The scale of the rapid power demand surge 

US data centers drew roughly 40 GW in 2024. By 2035, that figure is projected to reach 106 GW — nearly three times today's demand, driven almost entirely by AI workloads. For context: a large offshore wind farm generates around 1 GW. We're talking about the equivalent of 60 to 70 of them, planned, permitted, financed, and built within a decade. 

The grid is not ready for this. The Department of Energy estimated in July 2025 that 100 GW of new peak generation capacity will be needed by 2030, half of it directly attributable to data centers. The 47 largest US investor-owned utilities have responded by committing $1.1 trillion in capital spending between 2025 and 2029, matching the entire prior decade of investment in a single five-year window, according to the Edison Electric Institute. 

The estimated gap we just highlighted—106 GW by 2035 minus 40 GW today—means roughly 66 GW of new generation capacity needs to be planned, permitted, financed, and built within a decade. The real question is who finances that buildout, under what structure, and whether it ends up being gas or clean energy.  

To answer that, it helps to start with the companies driving the demand and the gap between what they have promised and what they are actually building. 

The stated commitments vs. reality gap 

The companies at the center of this story are known as hyperscalers, the handful of technology giants, Microsoft, Google, Amazon and Meta, whose cloud and AI infrastructure now operates at a scale that rivals national grid systems. Microsoft has committed to being carbon negative by 2030. Google has pledged to operate on 24/7 carbon-free energy across every grid where it operates by the same date. Amazon targets net-zero by 2040. Taken together, these are among the most ambitious corporate energy commitments ever made, and they come from companies with the balance sheets to back them up. 

Yet natural gas currently accounts for 26% of the electricity consumed by data centers globally, while renewables sit barely ahead at 27%. Whether that gap widens or closes depends almost entirely on decisions being made right now. Meta and OpenAI have plans to co-locate gas plants on-site. NextEra, the largest renewable energy developer in the US, announced plans with ExxonMobil on a 1.2 GW gas-powered data center campus. The same companies proclaiming clean energy leadership are signing gas deals. 

Calling it hypocrisy might be too simple. It is actually a rational response to a set of constraints that make gas the path of least resistance today. To understand that, we have to look at the policy and grid constraints around those decisions. 

Why the current policy environment tips the balance 

On his first day in office, President Trump froze IRA disbursements, withdrew from the Paris Agreement, and halted offshore wind permitting. Those were clear signals later reinforced with the OBBBA signed on July 4, 2025, which sharply limited wind and solar projects’ eligibility for the credits: only wind and solar projects that begin construction by July 4, 2026, or are placed in service by December 31, 2027 are eligible to claim section 45Y and 48E tax credits. Credits for nuclear, storage, geothermal and hydro remain intact through 2033.  

Renewables now face a double headwind: a pre-existing grid bottleneck that no policy has yet solved, and growing uncertainty over the tax credits that underpin their project economics. Gas, meanwhile, benefits from accelerated federal permitting and structural grid access advantages that bypass years-long interconnection queues.

To sum up, the policy environment now structurally favors firm, dispatchable power over variable renewables, precisely when data centers need to make decade-long energy decisions. And that policy asymmetry shows up directly in financing terms, not just in technology choices. 

Why hyperscalers are choosing gas 

To understand why policy conditions translate so directly into energy mix decisions, we need to understand how these projects actually get financed. The major hyperscalers are not constrained by capital in the way independent developers are. For these companies, choosing gas is not primarily a financing decision. It is an operational one.

Gas co-location behind the meter bypasses grid interconnection entirely, avoiding a process that now takes an average of eight years in key US markets. A gas plant near a pipeline corridor can be operational in 18 to 24 months. A renewable project, even with a signed power purchase agreement, still requires grid interconnection and transmission upgrades before drawing a single megawatt. When the AI capacity race is measured in quarters, that timeline difference is decisive. For hyperscalers, this is less ideology than timing: gas solves for speed in a way clean projects often still can’t. 

Why the project finance case for clean energy has never been harder 

But hyperscalers are not the only players building generation capacity to serve data centers. Behind them is a larger ecosystem of independent developers and utilities who rely on project finance – funding long-lived assets through non-recourse debt secured against future cash flows. This is how the majority of large-scale energy infrastructure gets built.

And this is where policy conditions hit hardest. A typical utility-scale solar or wind project relied on a specific capital structure: roughly 60 to 65% senior debt, and 35 to 40% tax equity – capital provided by institutional investors in exchange for federal tax credits generated by the project. Remove those credits, and you remove the tranche that made the economics work. The OBBBA did exactly that for wind and solar, imposing deadlines that most new projects cannot physically meet. Without tax equity, sponsors must bring significantly more of their own equity, which compresses returns to the point where many projects don't get financed. Gas requires no tax equity. The capital structure is cleaner, the timeline shorter, and the regulatory environment is now an explicit tailwind. 

Here is the real irony. The data center boom created something genuinely rare in infrastructure finance: investment-grade offtakers willing to sign 15 to 20-year power agreements at scale. Long-dated, creditworthy offtake is the foundation on which you build leverage, attract institutional capital, and reduce cost of capital. The clean energy industry had spent a decade working toward exactly this kind of anchor customer. The hyperscalers arrived, and the financing conditions that would have allowed clean energy project finance to meet them were dismantled at precisely the same moment. Storage, nuclear, and geothermal retain their credit runway, but wind and solar, the technologies that could have been deployed at the speed and scale this moment demands, do not. 

A missed moment, not an irreversible one 

The structural obstacles were real before this administration arrived. But acknowledging structural complexity is not the same as absolving policy choices of consequence. The OBBBA didn't create the problem, it foreclosed the solution. Tax equity for wind and solar, long-term PPAs anchored by investment-grade offtakers, a financing ecosystem that had spent a decade maturing toward exactly this moment — all of it was available, and most of it has been impaired during the specific window when it was most needed. That is not a coincidence. It is a policy outcome. 

The more useful question is not whether the administration is responsible, but what it would have taken to do this differently. The answer is less dramatic than the political debate suggests: faster interconnection reform, stable tax credit windows long enough to support project finance timelines, and permitting parity between gas and clean energy. None of those require choosing between AI growth and decarbonization. The tragedy of the current moment is not that the choice was hard. It is that, for a brief window, it may not have needed to be a choice at all. 

The AI buildout will be powered by something. The decisions being made today, by hyperscalers, utilities, lenders, and governments, will determine what that something is for the next two decades. For those of us who finance, structure, and advise on energy infrastructure, that is not an abstract question. That is now the financing problem on the table. 

Sources

• Data center demand and grid capacity projections: BloombergNEF (Dec. 2025) on U.S. data center power demand to 2035; U.S. Department of Energy / Lawrence Berkeley National Laboratory, 2024 United States Data Center Energy Usage Report (Dec. 2024 / Jan. 2025 release coverage); International Energy Agency, Energy and AI (Apr. 2025).  

• Grid investment commitments: Edison Electric Institute, Industry Capital Expenditures—2025 Projections; Deloitte, Can U.S. Infrastructure Keep Up with the AI Economy? (June 2025).  

• Hyperscaler sustainability commitments: Microsoft, Google, and Amazon public sustainability disclosures; McKinsey, How Hyperscalers Are Fueling the Race for 24/7 Clean Power (Dec. 2024).

• Energy mix and gas co-location: IEA, Energy and AI (Apr. 2025); Duke Nicholas Institute, Hyperscaler Data Center Buildout: A Sustainability Bane, Boon, or Both? (Aug. 2025); public reporting on NextEra/ExxonMobil data center campus announcements.  

• Policy and regulatory changes: One Big Beautiful Bill Act (OBBBA, signed July 4, 2025) and legal analyses of Sections 45Y/48E changes (e.g., Williams Mullen, Jones Walker/Jones Day-style summaries, Novogradac commentary).  

• Project finance and interconnection: Chambers and Partners, Project Finance 2025 USA; RMI, PJM’s Speed to Power Problem and How to Fix It (Nov. 2025); Deloitte, 2026 Renewable Energy Industry Outlook (Dec. 2025). 

• Image credit: Google Data Centers photo gallery, “Daytime view of the New Albany data center campus in central Ohio” (Google).