Firm Power 2026: Who Powers the Intelligence Economy

1. Introduction

Every article we have published this year has quietly depended on the same assumption: that the power would be there.

Defense autonomy needs compute. Physical AI needs compute. Orbital intelligence needs compute. And compute needs electricity — vast, continuous, uninterruptible quantities of it. For most of the last decade, that assumption held so reliably that investors treated energy as someone else's problem: a regulated utility question, a climate question, an ESG allocation. It was not a frontier technology question.

That has now decisively changed. Global data centre power demand is expected to rise 27% in 2026 to 132 gigawatts, up from 104 GW in 2025, and to reach 290 GW by 2030. [1] Gartner projects that 40% of AI data centres will be power-constrained by 2027. [2] In the United States, Goldman Sachs Research forecasts a structural data centre power shortfall of 9.3 GW in 2026, widening to 45 GW by 2028 — equivalent to the annual electricity needs of roughly 34 million households. [3]

The bottleneck has moved. It is no longer chips. It is watts.

For frontier allocators, this creates a specific and unusually legible investment thesis. The winners in this cycle will not be renewables in the abstract — solar and wind cannot deliver the 24/7 baseload that a hyperscaler requires. The winners will be whoever can deliver firm power: always-on, dispatchable, low-carbon electricity, sited close to load, at a price that makes sense. That is a narrower question than "the energy transition," and a far more investable one.

Picture 1. The Demand Shock: Global Data Centre Power Demand | Global data centre power demand rises 27% in 2026 to 132 GW, and is projected to reach 290 GW by 2030 — a near-tripling in five years that the existing grid cannot absorb. Source: [1]

2. The Capital Story

The demand signal is not speculative. It is contracted.

The five largest hyperscalers are projected to spend between $600 billion and $725 billion on capital expenditure in 2026, with roughly 75% of that — approximately $450 billion — going directly into AI infrastructure. [4] That capital has to land somewhere, and increasingly it lands wherever there is power to spare.

The constraint is physical, not financial. Roughly 2,300 gigawatts of generation and storage capacity currently sit in U.S. interconnection queues — more than the country's entire installed power base — with wait times in many regions now exceeding five years. [4] Transformers, the critical component for connecting new generation to the grid, carry lead times of two to four years. [5] Individual AI data centre sites are requesting between 100 and 750 MW each, loads that regional grids were never designed to deliver on short notice. [4]

The result is that hyperscalers have stopped waiting for the grid and started buying power directly. Microsoft signed an approximately $16 billion, 20-year power purchase agreement for the entire 835 MW output of the Three Mile Island Unit 1 restart. Meta has announced deals for up to 6.6 GW of nuclear capacity, including eight TerraPower Natrium plants (2.8 GW) and a 1.2 GW Oklo campus. Google signed the first U.S. corporate SMR fleet deal with Kairos Power. [6] Collectively, technology companies have now committed to more than 10 GW of new nuclear capacity for data centres. [6]

This is the structural change that matters for investors. The most creditworthy buyers in the global economy have begun signing 20-year offtake agreements for power that does not yet exist. That converts speculative energy technology into contracted infrastructure — and it is why capital is now moving into parts of the energy stack that were considered uninvestable five years ago.

Picture 2. The Hyperscaler Nuclear Land Grab | Technology companies have committed to more than 10 GW of new nuclear capacity for data centres, signing 20-year offtake agreements for power that does not yet exist — converting speculative energy technology into contracted infrastructure. Source: [6]

3. The Firm Power Stack

Like defense, physical AI, and space, energy is best understood not as a single sector but as a stack of layers — each with a different maturity, a different risk profile, and a different investability window.

Layer 1 — Nuclear restarts and existing fleet. The fastest path to firm power is the power that already exists. Restarting shuttered plants and extending the life of operating ones requires re-licensing and refurbishment rather than first-of-a-kind construction, which is why Microsoft's Three Mile Island deal targets 2027–28 rather than the 2030s. [7] This is the lowest-technology-risk layer, and it is dominated by incumbent utilities — Constellation, Vistra, Talen — which offer the most conservative nuclear exposure with existing cash flows plus AI-driven upside. [8] For frontier allocators, it is also the least interesting: the assets are public, large, and already repriced.

Layer 2 — SMRs and microreactors. This is where the private capital is concentrating. As of 2026, at least 66 companies across 15 countries are building small modular reactors and advanced nuclear technology, of which roughly 20 are publicly traded. [9] Major private rounds in 2025 included X-energy ($700M), Nano Nuclear ($400M), Radiant ($300M+), Newcleo ($125M), and Last Energy ($100M). [6] Total U.S. federal investment in new nuclear now exceeds $10 billion since 2020. [6]

The regulatory milestones are real but the timelines are long. Kairos Power holds the first NRC construction permit for a non-water-cooled reactor in over 50 years, and its Google agreement is the first corporate SMR purchase agreement in the world. [10] TerraPower has broken ground at Kemmerer, Wyoming, targeting 2030. [9] But no Western SMR has yet completed the full licensing gauntlet, and the realistic window for first data-centre-serving units is 2028–2030 at the earliest, with most 100+ MW projects landing in the 2030s. [7]

Layer 3 — Fusion. The most striking capital story of the past twelve months. The Fusion Industry Association's 2026 report, published this month, records a record $4.48 billion raised in the year to July 2026 — a 69% increase year-on-year — bringing cumulative private investment to $14.24 billion since 2021. [11] The survey now covers 56 companies, up from 23 in 2021. [11] The largest rounds: Commonwealth Fusion Systems ($863M Series B2), Proxima Fusion ($518M), Helion Energy ($465M Series G, led by Thrive Capital at a $15.5 billion post-money valuation), and Inertia Enterprises ($450M Series A). [11][12] Seventeen fusion startups have now raised more than $100 million each. [13]

Critically, fusion has begun to acquire the commercial architecture of a real industry: five companies have signed power purchase or offtake agreements — including Microsoft with Helion and Google with Commonwealth Fusion Systems — and six have secured sites for future facilities. [11] TAE Technologies and General Fusion are both entering the Nasdaq in 2026. [11] Seventy-one percent of surveyed companies expect first commercial electricity in the 2030s. [11]

Picture 3. Fusion's Inflection | Cumulative private fusion investment has grown from $1.8 billion in 2021 to $14.24 billion in 2026, with a record $4.48 billion raised in the twelve months to July 2026 alone. Source: [11]

Layer 4 — Long-duration energy storage. The most contrarian layer, and the one where the investment case is most contested. LDES installations rose 49% in 2025 to more than 15 GWh, but long-duration systems still accounted for only 6% of global storage installations. [14][15] More tellingly, global LDES funding fell 30% in 2025 and venture capital investment into the category dropped 72%, as high interest rates, competition from AI data centre capital, and falling lithium-ion prices squeezed the sector. [15]

That divergence — surging deployment, collapsing venture funding — is precisely what makes it interesting. The demand case is strengthening: Form Energy's 300 MW / 30 GWh iron-air system with Xcel Energy is the largest battery project by energy capacity announced globally, designed to discharge for up to 100 hours. [16] The DOE estimates the U.S. grid may need between 225 and 460 GW of LDES by 2050, requiring $330 billion in capital. [17] The LDES Council projects up to 8 TW of global capacity needed by 2040 — a fifty-fold acceleration from current deployment. [14]

Picture 4. The Firm Power Stack | Four layers, four investability profiles: restarts (lowest risk, already repriced), SMRs (where private capital is concentrating), fusion (record funding, 2030s timeline), and long-duration storage (deployment rising, venture funding collapsed). Sources: [6][9][11][14][15]

4. The Dual Driver: AI Demand Meets Energy Sovereignty

What makes firm power a durable thesis rather than a cyclical one is that two independent forces are pushing in the same direction.

The first is AI demand, described above. The second is energy sovereignty — and in Europe especially, it is arguably the stronger driver.

The European Commission's Nuclear Illustrative Programme estimates that nuclear energy will require approximately €241 billion in investment by 2050, covering both lifetime extensions and new large-scale construction, with additional investment needed for SMRs, advanced modular reactors, and fusion. [18] In March 2026, the Commission adopted its first dedicated SMR Strategy, targeting Europe's first SMR deployment by the early 2030s and creating a €200 million InvestEU guarantee to support private investment in initial commercial units. [19] Projections in the accompanying programme put potential EU SMR capacity at between 17 GW and 53 GW by 2050. [19]

The political shift is unambiguous. In a March 2026 address, Commission President Ursula von der Leyen said it had been a strategic mistake for Europe to turn its back on a reliable, affordable source of low-emissions power. [20] The France-led Nuclear Alliance, now spanning twelve-plus member states, is targeting 150 GW of European nuclear capacity by 2050. [20]

Investors are following the policy signal. European nuclear M&A reached a seven-year high in 2025, and deal value in 2026 had already reached $3 billion by early June — twice the full-year value recorded in 2025. [21]

Europe is nonetheless at an earlier point in the cycle than the United States, and that timing is itself the investment case. The EU hosts 19 of the 78 SMR designs under development globally — a substantial share — though as a cohort they are younger and earlier-stage than their U.S., Chinese, and Russian counterparts. [22] The first EU construction licences are still ahead rather than behind, with Romania expected to host the bloc's first SMR around 2030. [22][23] The €200 million InvestEU guarantee is an opening instrument against a €241 billion long-term need, which points to how much of the financing architecture is still being built.

For frontier allocators, that is a favourable set-up rather than a shortfall. It is the same asymmetry we identified in defense and in space: policy commitment and sovereign demand arriving ahead of the private capital base, which is precisely the window in which early positions are taken.

Picture 5. Europe's Firm Power Opportunity | Europe's policy commitment — €241 billion identified by 2050, a 150 GW capacity target, and a dedicated SMR Strategy — is arriving ahead of its private capital base, which is the window in which early positions are taken. Sources: [11][18][19][20][22][23]

5. The Convergence : Why Firm Power Sits Underneath Everything

Firm power is not a vertical alongside AI, defense, robotics, and space. It is the layer all four of them stand on.

Energy and AI is the most direct convergence, and it now runs in both directions. AI-optimised servers will account for 31% of data centre power consumption in 2026, and by 2027 their consumption will surpass that of conventional servers entirely. [1] But AI is also being deployed to solve the energy problem it created — from grid digital twins that simulate where new loads can safely be absorbed, to AI-driven optimisation of reactor operations and licensing documentation.

Energy and defense converge on the same sovereignty logic. Microreactors are being designed explicitly for military bases and forward operating environments; Radiant's Kaleidos is built to be mass-produced for U.S. military installations. [24] The customer that values firm, transportable, off-grid power most highly is the same customer driving the defense technology re-rating.

Energy and space converge on both power generation and materials. Antares designs microfission reactors for space missions alongside commercial and military energy resilience. [24] The same physics, supply chain, and regulatory expertise serves both.

The common thread — as in every stack we have mapped this year — is that the most valuable companies own a closed loop: contracted offtake from creditworthy buyers, proprietary technology that is difficult to replicate, and a position deep enough in the infrastructure that switching becomes structurally impossible.

6. What Remains Accessible for Frontier Allocators

The headline names have moved. Oklo trades at roughly $12.9 billion after a 700%+ rise over the past year; Helion is valued at $15.5 billion; Commonwealth Fusion Systems has raised approximately $3 billion. [8][12][13] These are no longer early-stage entries.

Four areas remain genuinely accessible.

The fuel bottleneck is the clearest picks-and-shovels opportunity. High-assay low-enriched uranium (HALEU) is required by nearly every advanced reactor design — TerraPower's Natrium, X-energy's Xe-100, Oklo's Aurora, Kairos's Hermes, Radiant's Kaleidos. Centrus is currently the only U.S. producer, operating a demonstration cascade producing roughly 900 kg per year against multi-ton demand from planned reactors. [6] The DOE has committed over $2.7 billion to domestic enrichment, and awarded Centrus a $900 million task order in January 2026 for a full-scale cascade targeting 6 metric tons per year. [6] The supply gap between current production and reactor demand is the single biggest risk to the advanced nuclear industry — and therefore the single clearest place where capital is needed. Standard Nuclear, the TRISO fuel manufacturer backed by Andreessen Horowitz and Chevron Technology Ventures, filed an S-1 targeting a $3.55 billion valuation. [25]

Long-duration storage is structurally mispriced. A category where deployment rose 49% while venture funding fell 72% is either broken or dislocated. [14][15] The EPRI and LDES Council benchmark projects roughly 37% cost reductions by 2030 for intraday electrochemical LDES. [14] Large offtakers — data centres, industrial users, utilities — can anchor early projects through long-term contracts, and the Google–Form Energy agreement suggests that transition is beginning. [14] For investors comfortable with infrastructure-style returns rather than venture multiples, this is the most contrarian entry point in the stack.

European fusion and SMR startups sit early relative to sovereign demand. Proxima Fusion's $518 million round in July 2026 is the largest European fusion raise on record, and the continent's 19 SMR designs represent a meaningful share of the global field at a stage where valuations have not yet compressed. [11][22] With the first EU licensing decisions still ahead, patient European capital enters a market where the policy tailwind is exceptional and competition for the best assets remains limited. [23]

The manufacturing and supply chain layer benefits regardless of which reactor wins. BWXT, the sole manufacturer of naval nuclear reactors for the U.S. Navy, is positioned as a potential "foundry for SMRs," supplying heavy components across multiple developers. [26] The same logic applies to transformers, turbines, and grid components — where two-to-four-year lead times are themselves the constraint. [5]

7. Conclusion

There is a pattern to the theses we have mapped this year. Defense, physical AI, space, and now energy have all followed the same arc: a category the market had written off, re-rated by a combination of geopolitical urgency and technological readiness, with the headline names repricing quickly and the durable returns forming one or two layers beneath them.

Firm power is the most consequential, because it sits underneath the other ones. There is no autonomous defense without compute. There is no physical AI without electricity. There is no orbital infrastructure without power. And there is no compute without watts.

The demand is now contracted rather than forecast: $450 billion of hyperscaler AI capex in 2026, 10 GW+ of committed corporate nuclear, 20-year PPAs signed with the most creditworthy buyers in the global economy. [4][6] The supply is not yet built. Between those two facts sits the investment opportunity.

The question for frontier allocators is not whether firm power is investable. The record capital flowing into fusion, the queue of hyperscalers signing nuclear offtake, and Europe's €241 billion long-term commitment answer that. The question is which layer remains early enough, and which bottleneck is severe enough, to generate returns that justify the wait. The fuel supply chain, the storage dislocation, and the European early-stage pipeline are where that answer is most likely to be found.

The future is digital. The returns are becoming physical.