Track Hyper | Who Powers AI? Amazon Bets on Small Modular Nuclear Power

Author: Zhou Yuan / Wall Street Insight

As the training and inference demands of generative artificial intelligence drive up global data center (IDC: Internet Data Center) electricity consumption, the question of "who will meet the supply" becomes a real issue.

In late August, American nuclear technology company X-energy announced a strategic cooperation with Amazon (AWS), Korea Hydro & Nuclear Power (KHNP), and Doosan Energy, planning to deploy over 5GW of small modular reactors (SMR) in the U.S. market, with AWS as the anchor customer.

This cooperation mobilizes up to $50 billion in capital, aiming not only to power AI and data centers but also to reshape a new value path in the capital market, supply chain, and energy landscape.

Nuclear Energy and AI: Demand-Driven Logic

For a long time, financial markets have explained the growth in data center electricity consumption through renewable energy, natural gas, and energy storage.

However, in the past two years, the electricity consumption growth from artificial intelligence training has exceeded expectations.

According to the International Energy Agency (IEA) report "Energy and Artificial Intelligence" released on April 10 this year, global data center electricity consumption could nearly double to 945TWh by 2030, with the U.S. being the primary load growth point.

In this context, relying solely on wind or solar power is insufficient to provide round-the-clock, low-carbon, and predictable electricity, prompting capital to reassess nuclear power.

Amazon's choice carries a clear financial logic: rather than binding power plants solely through power purchase agreements (PPA), it directly enters upstream, collaborating with reactor developers and equipment manufacturers to build projects.

This means that capital investment is not limited to electricity purchase costs but extends to equipment manufacturing, fuel supply chains, and long-term financial arrangements, forming a structure similar to "vertical integration."

If this model proves successful, it is expected to transform the uncontrollable risks of electricity procurement into calculable returns on capital investment.

The real challenge of small modular reactors lies not in the technical principles but in financing discipline.

The heavy lessons from past U.S. nuclear projects—Georgia's Vogtle third-generation units being over budget and delayed, and NuScale's UAMPS small reactor being canceled due to cost issues—have taught the capital market that without replicable financing and delivery models, projects are prone to the "increasingly expensive" trap.

The cooperation between X-energy and AWS is attempting to establish a new order in two dimensions: financing and project.

First, on the financing side: the mobilization capital framework of up to $50 billion aims to support multiple projects in bulk rather than financing a single demonstration power station. Capital is deployed in phases and through scaled procurement to reduce project uncertainty.

Next, on the project side: based on an 80MW single module, an initial construction of a 320MW power station is planned, with subsequent expansions. This model reduces initial cash flow pressure and allows investors to reassess midway.

From a financial perspective, this means that capital is closer to a portfolio investment logic: a group of projects is launched in batches and gradually expanded, with returns and risks averaged over different time periods.

The greatest concern for the capital market lies in supply chain uncertainty SMR generally requires high-assay low-enriched uranium (HALEU) fuel, but currently, the production capacity in the United States is limited.

The first commercial high-assay low-enriched uranium fuel manufacturing plant in the U.S., TRISO-X, has been accepted by regulators, but it will take several years to ramp up production capacity. This directly affects the discount rate of capital for projects: if there is a gap in the fuel supply, even if the project is under construction, commercial operation may be delayed.

This is the motivation behind Amazon's introduction of Korean partners.

KHNP and Doosan have mature experience and cost control capabilities in the nuclear power manufacturing sector, and can provide replicable manufacturing capabilities for core components such as pressure vessels, graphite components, and heat exchange systems.

For investors, this is equivalent to using supply chain stability to hedge project risks, thereby making the interest rates and capital costs of the financing structure more controllable.

This approach is similar to "risk sharing" in infrastructure investment: the manufacturing and construction phases are secured by experienced international vendors, while the fuel supply relies on the domestic expansion capabilities of the U.S. Department of Energy (DOE) and TRISO-X (a wholly-owned subsidiary of X-energy focused on the commercialization of TRISO fuel).

Different risk segments are clearly delineated and allocated to capable entities, forming a financing model recognized by the capital markets.

The Anchoring Effect of IDC and Financial Markets

In traditional energy financing, power plants often rely on utilities or power companies as buyers. However, in the AI-driven energy landscape, data center (IDC) companies have directly become anchor customers. This shift significantly changes the capital structure.

As one of the largest cloud service providers globally, AWS has long-term and rigid electricity demand. The capital market favors demand-side entities with high credit backing. Financial investors are more willing to finance a group of power plants directly tied to AWS, as their cash flows are predictable and the default risk is low.

This is also a structural change observed in the capital markets: future nuclear power financing may rely more on the credit of large electricity consumers rather than solely on electricity price markets or government subsidies. This makes nuclear power more like a "corporate energy debt instrument": the guarantees behind power generation projects come from stable large customers rather than policy guidance.

Amazon and X-energy have previously launched a demonstration project in Washington State, with an initial four modules (320MW), and reserved capacity for expansion to 960MW.

The significance of such demonstration power stations lies not only in power supply but also in forming a financially replicable model: once the project can be delivered on time and within budget, the capital market can use actual experience to calibrate the risk model for future 5GW deployments.

At the same time, the industrial thermal energy project in collaboration with Dow Chemical in Texas provides another financial perspective: nuclear energy can not only power data centers but also form long-term heat and power bundling contracts with industries such as chemicals and manufacturing. This diversification of scenarios reduces reliance on a single source of income, increasing the acceptability of debt and equity capital.

The pace of regulation directly affects capital confidence.

The U.S. Nuclear Regulatory Commission (NRC) has announced an 18-month construction permit review timeline, significantly speeding up compared to the past This is a positive signal for the financial market: policy risks are reduced; however, on the other hand, implicit costs such as community acceptance, waste management, and cooling water resources still need to be incorporated into the financial model.

When assessing projects, capital markets typically hedge these uncertainties by increasing the risk premium, which may also become a variable affecting the scale and interest rates of financing.

The collaboration between Amazon and X-energy is not just a single energy project, but an experiment in capital models.

Whether this experiment can power AI depends not only on technological maturity but also on financing discipline, supply chain resilience, customer credit, and policy coordination.

If the first plants in Washington State and Texas can complete their construction schedules, costs, and fuel guarantees by around 2030, the capital market will quickly replicate this model. At that time, the 5GW deployment plan will no longer be a slogan but may become a realistic path for a set of structured financing projects.

In other words, this is not merely an energy transition but a financial reconstruction driven by AI demand: nuclear power is being redefined as a "long-term asset allocation" for data centers in the interplay between electricity supply, capital operations, and customer credit