Who Holds the Real Power in the AI Race?

Second Level Thinking | Picks and Shovels

DISCLAIMER & DISCLOSURE: The author may hold positions in one or more of the companies mentioned at the date of publication but that may change. The views expressed are those of the author and may change without notice. The author has no duty or obligation to update this information. Some content is sourced from third parties believed to be reliable, but accuracy is not guaranteed. Forward-looking statements involve assumptions, risks, and uncertainties, meaning actual outcomes may differ from those envisaged in this analysis. Past performance is not indicative of future results. All investments carry risk, including financial loss. This analysis is for educational purposes only and does not constitute investment advice or recommendations of any kind. Conduct your own research and seek professional advice before investing.

---

The Power and the Glory

The artificial intelligence boom is often framed as a race for faster chips and larger models, but that framing misses the real constraint that will define the next decade. AI chips are growing at roughly 170% year over year, data center footprints are doubling every three years and hyperscalers like Amazon (NYSE: AMZN) and Google (NYSE: GOOG) are committing hundreds of billions of dollars to AI infrastructure. Yet none of that capital matters if the servers can’t be powered. The true bottleneck isn’t silicon, it’s electricity, and this requires second-level thinking on the part of investors. The smart money is rotating into sectors where valuations actually reflect fundamentals.

The idea that AI data centers require enormous amounts of power, and that today’s energy infrastructure is not prepared for that demand, is not new. What is often underappreciated is the sheer scale of the numbers involved. By 2030, U.S. data centers alone are projected to consume roughly 1,200 terawatt hours of electricity annually, enough to power every American home twice over. Another way to think about it is this: meeting that demand would require the equivalent of building thirty Hoover Dams every single year just to keep the servers running.

Globally, the picture is even more stark. The International Energy Agency estimates that electricity demand will rise from roughly 415 terawatts today to nearly 1,000 terawatts by 2030, and that surge occurs before the next generation of even more compute-intensive AI models comes online. In that world, the defining competitive advantage of the AI era will not be who builds the fastest chip, but who controls the energy that keeps those chips alive.

IEA Data center electricity consumption forecasts by region (base case 2020 - 2013

Modern AI data centers are already pushing the grid to its limits. A single large AI training campus can draw hundreds of megawatts continuously. To put that in perspective, Meta is developing a site with approximately five gigawatts of capacity. That is more power than California’s Diablo Canyon nuclear plant, which produces about 2.2 gigawatts and remains the state’s last operating nuclear facility.

Once AI models are deployed, the power demand never turns off. Inference load, the constant electricity draw from AI embedded across phones, searches, cloud software, and vehicles, turns AI into a 24-hour baseload industry. Renewables can’t supply uninterrupted power at this scale, while coal is incompatible with the carbon targets that corporations have already committed to. That naturally leads many to ask whether natural gas is the answer.

The Permian Basin is often cited as a logical solution. It offers abundant trapped natural gas, vast stretches of contiguous land suitable for data centers, space for gas-fired generation, and room for solar as a supplementary source. Texas also brings relatively favorable regulatory conditions and low population density, which can significantly speed up construction timelines. This is not a theoretical argument. Chevron has announced plans to build a 2.5-gigawatt power generation facility in Reeves County, Texas. In December, Eric Schmidt, formerly of Google, launched Bolt Data & Energy with the goal of bringing ten gigawatts of data center capacity to the Permian Basin. Clearly, some of the most sophisticated players in technology and energy are placing their bets on natural gas.

Yet there is an inconvenient truth that often gets glossed over. Data centers generate immense heat and require constant cooling, which is precisely why colder climates have historically been preferred. Against that backdrop, it’s fair to ask whether Texas, with its extreme temperatures and rising climate volatility, is really the optimal long-term solution.

There is an alternative. A scalable, zero-carbon, always-on solution: nuclear energy. When it comes to power generation, reliable baseload power to meet the demands of AI data centers is essential. Nuclear power plants are highly dependable, producing power 93% of the time, on average, versus just 57% for natural gas and 40% for coal. sources like wind and solar are more variable, generating power 35% and 25% of the time, respectively. As a result, this is yet another reason why nuclear ought to be the number one choice.

Nuclear power is also being endorsed by the White House.

For investors it is important to understand that historically, while the initial capital costs for nuclear have been high, the “energy return on investment” (EROI), surpasses that of competing energy sources (i.e., coal and natural gas). But, as shall be explained later, the cost of deploying nuclear power is coming down with the advent of small modular reactors, so this effect will be even more pronounced in future.

If nuclear power captures even 30% of incremental AI-driven electricity demand, it opens a $60 to $90 billion annual market by 2035, effectively doubling the size of the entire nuclear industry.

There is something else that appears to be underappreciated. For decades, nuclear power has been synonymous with megaprojects, massive, bespoke construction efforts plagued by delays, cost overruns, and political friction. That legacy is real, but it’s no longer the whole story. A fundamental shift is underway as the industry pivots toward modular reactors, reshaping both the economics and the investment profile of nuclear energy.

The “on-site, off-grid” nuclear thesis sits at the intersection of three powerful forces. The first is the insatiable appetite of AI for reliable baseload power. The second is the global mandate for carbon neutrality. The third is a reimagining of nuclear hardware as a factory-produced commodity rather than a bespoke civil engineering feat.

At the core of this shift is the decoupling of power generation from the centralized grid. Historically, power plants were built far from demand centers, with electricity transmitted over aging infrastructure. Today, the largest sources of incremental demand, hyperscale data centers and dense industrial clusters, cannot afford to wait a decade for grid upgrades. They need reliable, carbon-free power exactly where the chips are humming. This is the “on-site” advantage. By shrinking the footprint of a reactor from something resembling a city suburb to something closer to a shipping container, a new generation of companies is turning energy into a plug-and-play industrial asset.

AI cannot scale another decade without a new energy backbone. The companies that build that backbone will shape how far the technology ultimately goes. That opportunity does not sit in a single stock or a single breakthrough, but across a four-tier nuclear stack that runs from uranium in the ground to electrons flowing directly into data centers. Understanding that stack, and how each layer fits into the AI energy equation, is where the real investment insight begins.

The four-tiers are:

• Uranium mining and feedstock

• Conversion and enrichment

• Nuclear reactors

• Operating utilities

Let’s explore each in turn.

Tier 1. Uranium mining and feedstock

As the "Renaissance" of nuclear takes hold, the demand for raw uranium remains a fundamental driver of the entire sector.

Commercialized nuclear energy produced within a conventional nuclear reactor is created through fission: the splitting of large uranium atoms into smaller atoms. The byproduct of this fission is heat that boils water to produce steam, which drives a turbine to create electricity.

Uranium is the naturally occurring element that serves as the primary fuel for nuclear energy generation through fission. Some estimates suggest that the Earth’s crust contains 35 trillion tons of this element, albeit not distributed evenly across the globe. Today, two-thirds of uranium production comes from three countries: Kazakhstan, Canada, and Australia (see chart below).

While supply is dominated by three countries, 99% of uranium mined is used in nuclear power generation (equating to roughly 10% of the world’s electricity generation), and the demand side of the market has a different geographic mix (see chart below).

Global uranium demand is approximately 180 million pounds per year, while supply is closer to 140 million pounds. The resulting 40-million-pound deficit is being filled by inventories that are rapidly shrinking. New production is no longer optional; it’s urgent.

The Uranium supply vs demand imbalance

Uranium producers, developers, explorers and other ways to play the market

There are many companies operating in the market and it is beyond the scope of this post to touch upon them all, so what follows is an overview highlighting some names of interest.

Uranium Energy Corp (NYSE: UEC) exemplifies this urgency with fully permitted, shovel-ready projects in Texas and Wyoming, many acquired from Rio Tinto and Uranium One Americas during periods of depressed prices. Its in-situ recovery mining method allows production to restart within months rather than years once prices justify it and the company already delivers uranium directly into U.S. government stockpile programs under Department of Energy contracts. In a market where domestic uranium supply is increasingly viewed as a national security priority, Uranium Energy Corp benefits from existing permits, infrastructure and political tailwinds.

NexGen Energy Ltd (TSX: NXE) represents a different kind of leverage. Its Arrow deposit in Saskatchewan’s Athabasca Basin contains some of the highest-grade uranium on Earth, averaging over 4% compared with a global average of roughly 0.2%. This translates into materially lower mining costs, smaller physical footprints and reduced waste. A completed feasibility study indicates production costs below $10 per pound and the company has secured agreements with local First Nations communities, a critical requirement for Canadian permitting. Once financed and constructed, Arrow could produce 29 million pounds annually, representing roughly 10% of global uranium mine supply from a single site.

For investors seeking exposure with reduced operational risk, Uranium Royalty Corp (TSX: URC) offers a different model. Rather than operating mines, it focuses exclusively on royalties and streaming agreements, providing exposure to uranium production without the regulatory and execution risks inherent in mining.

This structure allows the company to sidestep many regulatory hurdles, including the Nuclear Regulatory Commission’s historically onerous approval processes, where a single reactor design application once required 12,000 pages and cost $500 million. Uranium Royalty Corp’s diversified portfolio spans 19 projects globally, includes partnerships with firms such as Yellow Cake PLC (LSE: YCA) for physical uranium access and has already delivered tangible returns, with the stock rising over 160% in the past six months. [A link to my analysis of Yellow Cake for further reading.]

Energy Fuels Inc (TSX: EFR / NYSE: UUUU) occupies a strategically unique position within the U.S. supply chain as the owner of White Mesa Mill in Utah, the only conventional uranium mill currently operating in the country. The mill is licensed to process uranium from third-party mines and permitted to handle rare earth elements, making it a dual-use strategic asset. Energy Fuels has already sold uranium into the Department of Energy’s newly established uranium reserve program and is positioned as a primary U.S. converter for government and utility customers seeking to avoid Russian supply chains.

Anchoring the mining tier is Cameco Corp (TSX: CCO), one of the largest publicly traded uranium producers globally, operating in Canada and Kazakhstan. Its Cigar Lake mine is among the highest-grade uranium operations in the world and the company controls approximately 15% of global production. This is a "picks-and-shovels" play on the entire nuclear renaissance with the security of existing cash flows from current-generation reactors: all nuclear reactors need fuel.

Of particular note is that Westinghouse Electric Company is owned by a strategic partnership between Brookfield Corp (NYSE: BN) (51%) and Cameco (49%). This consortium combines Brookfield's energy investment expertise with Cameco's nuclear fuel cycle knowledge to position Westinghouse as a key player in the global clean energy transition.

Westinghouse Electric is a global leader in nuclear technology, providing nuclear fuel, services and advanced reactor designs (like the AP1000 and SMRs) for the entire lifecycle of nuclear power, from building new plants to operation and maintenance. On December 8, 2025, Brookfield and Cameco, through Westinghouse, have entered into a transformational partnership with the U.S. government to deliver $80 billion of new nuclear plants in the U.S. With a market capitalization exceeding $20 billion, Cameco is priced for stability rather than explosive growth, but it remains the blue-chip option for investors seeking uranium exposure with a lower risk profile.

Tier 2. Conversion and enrichment

Uranium must undergo several processes, called the nuclear fuel cycle, before it can be used within a nuclear reactor. The cycle begins with mining and milling, where uranium ore is extracted from the ground and processed into a concentrated powder called “yellowcake” (U₃O₈). This material is then converted into a gas called uranium hexafluoride, which is the form needed for enrichment.

The enrichment step is critical to the process because natural uranium ore contains only about 0.7% of the isotope U-235 (the isotope that sustains nuclear reactions). During enrichment, the concentration of U-235 is increased to about 3-5% for standard reactor fuel, known as low-enriched uranium (LEU), or up to 20% for high-assay low-enriched uranium (HALEU) used in advanced reactors.

Those concerned with enrichment and conversion make up the second tier of the stack, which is the tightest bottleneck in the nuclear supply chain.

Roughly 45% of the world’s enriched uranium supply still comes from Russian facilities, making fuel enrichment one of the most significant vulnerabilities in the nuclear thesis.

Control over enrichment capacity isn’t merely a supply issue, it’s a gatekeeping function.

Centrus Energy Corp (NYSE: LEU) operates the only U.S. facility currently licensed to produce HALEU, which can enable smaller reactor cores, higher fuel efficiency and longer operating cycles in many advanced reactor designs.

It should be noted that the aforementioned performance gains are design-dependent; HALEU enables these attributes in many advanced designs rather than guaranteeing them in all reactors that use it.

Centrus is, in effect, the only active U.S. commercial-scale HALEU production cascade in North America at the Department of Energy’s Piketon, Ohio facility. This currently makes them a ‘toll booth’ for the advance modular reactors that we’ll discuss in tier 3 below.

Other HALEU supply comes from overseas, with Russia’s Tenex facility being a primary supplier — not much use to most of the world given the ongoing sanctions against Russia.

However, the Department of Energy (DOE) has selected other companies, such as French nuclear giant Orano Group and General Matter, an American nuclear-fuel start up company, founded in 2024, to engage in HALEU-related programs, so Centrus should not be considered to have monopoly power.

Centrus does have a first mover advantage. It delivered its first commercial HALEU batch in late 2023 and has since scaled production beyond six metric tons annually, sufficient to fuel multiple SMRs.

Where Centrus enriches fuel, BWX Technologies (NYSE: BWXT) fabricates it into usable reactor cores. BWX produces nuclear components for both the U.S. military and commercial markets and manufactures TRISO fuel, in which each uranium kernel is encapsulated in carbon and ceramic layers capable of withstanding extreme heat without meltdown. BWX operates one of the few facilities in the world qualified to produce TRISO fuel, located in Lynchburg, Virginia, under contracts with both the DOE and the Department of Defense.

Defense programs fund much of the research and development, while commercial demand is just beginning, particularly for micro-reactors designed for AI data centers, industrial campuses and remote infrastructure. Together, Centrus and BWX form the critical bridge between uranium and electricity, protected by regulatory timelines that cannot be accelerated.

Tier 3. Nuclear reactors

Reactor innovation is where nuclear begins to resemble technology rather than heavy infrastructure. These companies carry the highest risk, but also the highest potential reward.

Companies are developing the next-generation Small Modular Reactors (SMRs) and Microreactors. Technological developments have seen a move from large nuclear power stations to small, modular, on-site, off-grid power supplies. Not only do these introduce much needed power supply where it is most needed, but they also by-pass the aging electric grid infrastructure which may not be able to handle the increasing capacity demands.

Traditional Nuclear Power Plants (NPP) and New Small Modular Reactors (SMR)

The investment narrative is no longer about government subsidies, but about the “Big Tech” backstop. When entities with the balance sheet strength of Meta, Microsoft and Amazon sign direct Power Purchase Agreements with modular developers, they effectively de-risk the commercialization phase.

Big tech partnerships with nuclear power providers with approximate uranium (U3O8) yellowcake requirement. Data for US only.

This private-sector validation attracts early-stage capital and creates a predictable revenue roadmap for investors. It means that these are not speculative science projects; they are becoming the essential infrastructure of the digital age.

So where are the investment opportunities?

Oklo Inc (NYSE: OKLO), backed by Sam Altman of OpenAI, is developing micro-reactors, known as Aurora, capable of producing roughly 15 megawatts of continuous power, sufficient for large AI data centers. Its Aurora design operates for up to ten years without refueling, uses no water-cooled core and generates minimal waste.

Rather than selling reactors, Oklo plans to own and operate them, offering nuclear power as a service (NPaaS) to private customers, industrial sites, defense installations and hyperscalers.

The company has secured a conditional site use permit from the DOE at Idaho National Laboratory and targets its first operational reactor by 2027. More interesting is that it has signed a massive 1.2 GW deal with Meta (NYSE: META) to power data centers directly.

NuScale Power (NYSE: SMR) - yes, their ticker SMR is the acronym for small modular reactor - represents the more established end of the innovation spectrum. In 2023, it became the first and only small modular reactor design fully approved by the U.S. Nuclear Regulatory Commission, creating a significant regulatory moat.

Each NuScale module generates 77 megawatts and configurations can scale to nearly one gigawatt.

Factory production allows deployment timelines under four years, compared with 10 to 15 years for traditional large scale water-cooled nuclear power plants.

NuScale’s NRC-approved SMR flagship Voyager (VOYGR) design modular reactor, developed with Utah Associated Municipal Power Systems and backed by the DOE, marked the first commercial deployment. It has partnerships with the Tennessee Valley Authority (TVA) and active projects in Eastern Europe (Romania), Their VOYGR modules are larger than microreactors but designed for "factory-built" modularity.

They are targeting a variety of customers, including industrial sites and data centers, for on-site power and heat.

While their initial flagship project (UAMPS) faced challenges, the design certification is a colossal regulatory moat. The stock has been heavily sold off after project setbacks and so this may offer an interesting entry opportunity.

Nano Nuclear Energy (Nasdaq: NNE) isn’t data center or AI focused, but for the sake of completeness, it is worth a mention. It’s a speculative play on highly portable microreactors (ZEUS and LOKI) designed for shipping containers. It targets remote mining and military applications where off-grid power is essential.

Most pure-plays (OKLO, SMR, NNE) are currently pre-revenue and lose money. Expect periodic share offerings to fund the years-long construction and licensing phases, which introduces risks of dilution.

If this is off putting for you, there are the blue chip ‘non-pure play’ businesses that are active in tier 3 to consider.

When discussing ‘conversion and enrichment’ in tier 2, we looked at BWX Technologies. It is a seasoned nuclear engineering firm having built over 400 nuclear reactors for the US Navy. Relevant to this tier 3 discussion, it is pivoting to commercial microreactors, developing the BWXT Advanced Nuclear Reactor (BANR), designed for remote mining, military and off-grid use. While not a pure play in tier 3, BWX Technologies is a critical enabler and developer with the benefit of having existing government/defense contracts providing a stable revenue floor. It is a critical manufacturer for almost every other SMR startup. However, for BWX the microreactor business is a future growth bet; its current valuation is driven by its legacy businesses (medical isotopes, government contracts).

Rolls-Royce Holdings (LSE: RR / ADR: RYCEY), the aircraft engine manufacturer, has invested heavily in the manufacture of factory produced modular nuclear reactors. It is actively developing a 470 MWe SMR through its unlisted subsidiary Rolls-Royce SMR Ltd. The UK government has provided direct design funding and the company targets both grid and industrial applications. It aims to be producing at scale by 2030 and the share price has been on a tear over the past few years based, at least in part, on excitement around this new revenue stream.

GE Vernova Inc (NYSE: GEV) was a spin-off from General Electric and houses GE Hitachi Nuclear Energy, which is developing the BWRX-300 SMR, one of the most advanced and actively marketed designs globally, with projects planned in Canada, Sweden, and the U.S. As a spin-off, GE Vernova is a pure-play energy company where the nuclear segment could become a significant growth driver.

Westinghouse Electric is developing the eVinci microreactor, explicitly marketed for remote communities, mines, and off-grid industrial sites. Exposure to Westinghouse, itself not publicly traded, may be had by investing in Brookfield Corporation or Cameco (discussed above in tier 1).

Risks and Other Things to Consider when investing in Tier 3

1. Regulatory Pathway: Where is the design in the licensing process (NRC in U.S., CNSC in Canada, etc.)? This is the single biggest hurdle.

2. Fuel Strategy: Does the reactor require traditional fuel or HALEU? HALEU supply is currently limited, creating a potential bottleneck.

3. First Customer/Market: Who is the “anchor tenant”? (e.g., data centers, remote mines, defense departments, utilities). A firm orderbook is a huge positive signal.

4. Balance Sheet & Funding: How much cash do they have to burn through the long development cycle? Are they reliant on dilutive equity raises, or do they have strategic/government funding?

5. Strategic Partnerships: Links with established players (like Fluor for NuScale, or Constellation for some GE-Hitachi projects) reduce execution risk.

Tier 4. Operating Utilities

The fourth and final tier in this nuclear stack is the operating utilities, where innovation meets execution. Two to consider are:

Dominion Energy (NYSE: D) operates 3 nuclear plants across Virginia and Connecticut, supplying zero-carbon baseload power to more than two million homes. Reactor licenses extend into the 2060s, providing decades of operating runway. More notably, Dominion has signed an agreement with Amazon to explore SMR deployment in Virginia, marking the first direct hyperscaler-utility nuclear partnership.

Duke Energy (NYSE: DUK) operates 11 nuclear units across six sites in the Carolinas, totaling more than 10,000 megawatts, or roughly 10% of the entire U.S. nuclear fleet. License extensions run into the 2050s and Duke’s 2025 resource plan explicitly includes new reactors and SMRs to meet data center and AI demand across the Southeast.

These utilities are cash-flow positive, dividend-paying and already executing at scale.

Conclusion

The shift toward small modular reactors is not just a technological evolution, it is a supply-chain shock. Advanced reactor designs increasingly rely on High-Assay Low-Enriched Uranium, a specialized fuel with higher energy density and far more stringent processing requirements. That immediately creates bottlenecks. Companies such as Centrus Energy sit in a strategic toll-booth position, controlling access to a critical input that few others are qualified to provide. At the same time, the extreme manufacturing precision required for modular nuclear components elevates firms like BWX Technologies from routine contractors to indispensable partners. Their nuclear-grade certifications and quality controls are not easily replicated, forming a moat that meaningfully limits new competition.

Any professional assessment of the sector, however, has to acknowledge its volatility. We are currently in what might best be described as the deployment gap, the uneasy phase where headline-grabbing contracts collide with regulatory scrutiny and the realities of first-of-a-kind engineering. The core investment thesis behind modular nuclear is the transition from bespoke to serial production. The prize is a world where the hundredth reactor is dramatically cheaper and faster to build than the first. For investors with a five-to-ten-year horizon, that transition offers a rare chance to own the physical foundation of the next industrial cycle, but it is not a path to quick or effortless returns.

From an investment standpoint, how one approaches nuclear energy should ultimately reflect individual risk tolerance. This is a high-risk, high-potential opportunity. Most pure-play companies are pre-revenue, pre-approval, and their share prices can swing violently on policy announcements, regulatory milestones, or shifts in investor sentiment. For many, the most sensible way to engage with the theme is to think of it as a venture-style allocation spread across the four tiers of the nuclear stack, embedded within a broader, diversified portfolio.

At one end of the spectrum sit innovation-focused reactor developers, offering extraordinary upside but carrying meaningful execution and technology risk. At the other end are operating utilities, which provide relative stability, predictable cash flows, and dividends, but far more modest growth. Between those poles lie fuel and enrichment, arguably the most acute bottleneck in the entire system, and uranium mining, which provides exposure to rising demand but remains subject to commodity price dynamics rather than pricing power. A balanced approach might overweight utilities for stability, lean heavily into fuel and enrichment for leverage, maintain measured exposure to miners for supply optionality, and reserve a smaller allocation for reactor developers as asymmetric upside.

Investors with a higher tolerance for risk, and a preference for concentration over diversification, may choose a more selective approach. A sniper rifle rather than a scattergun. Over a ten-year horizon, even if AI power demand grows at only half the pace currently projected, nuclear demand could still double by 2035. Fuel and enrichment capacity may need to expand three to five times. Successful reactor platforms could deliver five-to-ten-fold returns. Utilities could plausibly double or triple while paying dividends along the way, and uranium miners could see two-to-three-times appreciation driven by structurally higher prices.

This is not a short-term trade. It is a long-duration thesis grounded in physical constraints rather than financial engineering. A decade from now, every AI data center’s uptime will depend on a relatively small group of companies that control the journey from uranium ore to electrons delivered at the server rack.

Today, most investors remain fixated on chips and software, areas where risk has significantly risen. Valuations have expanded aggressively on the back of exuberant expectations, capital expenditure has reached unprecedented levels, and companies are increasingly forced to burn cash simply to remain competitive. Balance sheets are being stretched as debt grows faster than revenues, opaque circular capital flows obscure underlying fragility, and artificial intelligence itself is showing early signs of commoditization that threaten long-term profitability.

The more durable opportunity may sit deeper in the stack. Grid capacity and power generation are the true bottlenecks of the AI economy. The U.S. push to accelerate power auctions, combined with structurally higher electricity prices in data-center-heavy regions, points to the early stages of a grid supercycle. No matter who ultimately wins the AI race, one outcome is certain: baseload power will be consumed at levels the world has never seen before.

This analysis only scratches the surface, but it will have you thinking about nuclear power not as a legacy industry, but as a critical enabler of the AI age. It should serve as a solid starting point for considering investing in this ‘picks and shovels’ sector of the AI revolution.

• What do you think about this opportunity?

• Do you have any strong views on how best to play it, or which companies are likely to emerge as winners?

• Perhaps you have insights into the management of some of these companies or the unique qualities that they possess which provide them with a competitive advantage?

• Which companies did I not mention that ought to have been included?

Please leave your comments below. Let’s start a discussion.