The Constraint Nobody Talks About: Energy Is the Binding Limit on AI Growth

AI has an energy problem. Every GPU cluster, every inference call, every foundation model training run requires power that terrestrial grids are struggling to supply fast enough. Between 30% and 50% of planned US data centre projects are now delayed or cancelled due to grid connection and infrastructure bottlenecks. The growth of artificial intelligence is constrained not by the availability of chips but by the structural inability of the power grid to deliver electricity. Two private companies are building around that constraint from opposite ends of the technology spectrum, and together they illustrate where the most interesting energy infrastructure bets currently sit.

The Demand Is Already There

Before getting to who is building the supply, we dive into how concrete the demand has become.

As of May 2026, every major tech hyperscaler has signed at least one nuclear power deal for AI data centre capacity. Across 13 announced projects, Google, Microsoft, Amazon, and Meta have committed to over 9.8 GW of nuclear capacity. Microsoft locked in 835 MW via a 20-year power purchase agreement for the Three Mile Island Unit 1 restart. Google committed 500 MW from Kairos Power. Amazon invested $700 million in X-energy for up to 12 small modular reactors. Meta leads with up to 6.6 GW across TerraPower, Oklo, Vistra, and Constellation.

In March 2026, multiple hyperscalers signed a White House pledge to directly fund the generation and grid upgrades their projects require, bypassing the public utility model entirely. These are not renewable energy purchase agreements driven by ESG commitments. These are direct infrastructure bets, driven by compute demand.

Global grid investment is expected to reach $550 billion in 2026, but transformer and substation supply chains still face multi-year backlogs. Even fully funded data centre projects can wait years for grid connections. The hyperscalers have read that timeline and are funding private power solutions directly. The private companies building those solutions are the ones worth watching.

Panthalassa: Power Where Nobody Has Built Before

Panthalassa starts with a simple premise: the ocean covers 70% of the planet, receives abundant wave energy 24 hours a day, and nobody has seriously tried to build compute infrastructure there. The company's autonomous nodes (roughly 300 feet long, shaped like a lollipop, with a buoyant head and a long submerged tube) generate electricity from the relative motion between the node and surrounding waves, power AI inference servers directly on-site, cool those servers with cold seawater, and transmit data to shore via low-Earth-orbit satellites. The nodes require no grid connection, no transmission lines, and no permitting battles with local communities.

CEO Garth Sheldon-Coulson frames it as a simple energy geography argument: "There are three sources of energy on the planet with tens of terawatts of new capacity potential: solar, nuclear, and the open ocean."

The Series B closed in May 2026 at $140 million, led by Peter Thiel, with John Doerr, Marc Benioff's TIME Ventures, Max Levchin's SciFi Ventures, Hanwha, Fortescue Ventures, Super Micro Computer, Founders Fund, and Lowercarbon Capital all participating. The funding will complete the company's pilot manufacturing facility near Portland, Oregon, and accelerate deployment of its Ocean-3 series of nodes, with pilot deployments in the northern Pacific planned for 2026 before commercial operations in 2027.

The technology has been validated at sea across three prototype generations: Ocean-1 (2021), Ocean-2 (2024), and the Wavehopper series, confirming wave-powered electricity generation, self-propulsion, station-keeping, seawater cooling for servers, and satellite connectivity. The Ocean-3 series is the first generation designed for commercial manufacturing scale. The team draws from SpaceX, Blue Origin, NASA, Boeing, and naval architecture, with approximately 120 people focused on hardware, autonomy, and deployment in harsh marine environments.

The key investor question is not whether the technology works (prototype validation is confirmed) but whether it can scale economically to compete with land-based alternatives. The company is pre-revenue, transitioning from R&D to initial commercial deployments, and the $140 million Series B is the capital that funds that transition.

Valar Atomics: Nuclear Reactors, Deployed Like Hardware

Valar Atomics is approaching the same problem from a different direction: small, prefabricated high-temperature gas-cooled nuclear reactors designed to be manufactured at scale, deployed rapidly, and operated as dedicated power infrastructure for AI data centres and industrial facilities.

The company's NOVA Core achieved zero-power criticality at Los Alamos National Laboratory's National Criticality Experiments Research Centre in November 2025, making it the first company to reach that milestone under the US Department of Energy's Nuclear Reactor Pilot Programme. In February 2026, its Ward250 reactor was airlifted from California to Utah aboard three C-17 Globemaster military cargo aircraft, a logistics exercise that also served as a proof of concept for rapid reactor deployment.

The company raised $450 million at a $2 billion valuation in April 2026, including $340 million in equity and $110 million in debt, backed by Palmer Luckey (Anduril Industries), Shyam Sankar (Palantir CTO), Snowpoint Ventures, and Lockheed Martin board member John Donovan. This week, Valar is reported to be in talks to raise at a $6 billion valuation, three times its April mark, with Sequoia Capital expected to lead and a new partnership with Nvidia to explore nuclear power for AI data centres.

The regulatory picture is complicated. Valar has filed suit against the Nuclear Regulatory Commission, arguing that it applies the same protracted licensing process to small test reactors as to large commercial facilities, a legal dispute that remains unresolved and represents the primary timeline risk for commercial deployment.

Why the Timing Matters

The hyperscaler nuclear deals confirm something important for investors evaluating the private energy infrastructure opportunity: the demand is real, the willingness to pay is there, and the largest technology companies in the world are already funding supply directly rather than waiting for utilities to act.

What those deals also confirm is that the public market solutions are years away. Microsoft's Three Mile Island restart delivers first power in 2027. Google's Kairos Power commitment targets first power around 2030. Amazon's X-energy reactors have no confirmed commercial date. The hyperscalers are signing 20-year agreements precisely because they know the public infrastructure timeline cannot meet their near-term compute demand.

Private companies building alternative energy infrastructure now are not racing against utility-scale public projects. They are filling a gap that those projects cannot fill on the timescales AI compute growth requires. Panthalassa's Ocean-3 commercial deployments target 2027. Valar's Ward250 reactor targets operational status this year under the DOE pilot programme. Both are earlier than most public market nuclear alternatives.

As we covered in our piece on AI and energy demand and our analysis of Anthropic's $21.6 billion Australian data centre commitment, the energy layer underneath AI is where some of the most consequential private market positions are being built right now. The companies that solve the power constraint first, at commercial scale and acceptable cost, are not competing for a future market. They are competing for a market that already exists and is already constrained.

NonPublic Pty Ltd (ABN 49 607 216 928) holds Australian Financial Services Licence #482668. Investments are available to wholesale and sophisticated investors as defined under the Corporations Act 2001. This content is general in nature and does not constitute financial product advice. It does not take into account your objectives, financial situation, or needs. Investing in private markets involves significant risk, including the potential loss of your entire investment. Past performance is not a reliable indicator of future results. You should obtain independent financial advice before making any investment decision.