AI's Energy Appetite: Big Tech Turns to Nuclear and Alternative Power Sources

Big Tech's AI Ambitions Could Supercharge These Energy Stocks

Companies that make fuel cells, mobile natural gas turbines, and small modular nuclear reactors are some of the potential beneficiaries of this next phase of the AI infrastructure buildout. AI data centers are expected to consume a massive amount of energy in the coming years, and meeting that need could be a boon to some investments, according to Morgan Stanley. Morgan Stanley forecasts U.S. data centers will consume 65 gigawatts of power between 2025 and 2028, but available capacity could fall short by about 45 GW. To make up the difference, "all potential 'de-bottlenecking' solutions will need to be drawn upon," the analysts wrote in a note on Tuesday. Possible solutions, they say, include converting crypto mining operations into data centers, building data centers at large nuclear power plants, and constructing new natural gas-fired power plants. But all of that is easier said than done. First, the rising price of bitcoin could discourage miners from converting their mining facilities or selling excess power to data centers. Second, concerns about stressing regional power grids could compel regulators to mandate that new data centers not come online until additional power sources are connected to the grid. That's why Morgan Stanley expects to see hyperscalers and data-center owners adopt a "bridge" approach, "in which temporary, mobile generation is deployed" to address the regulatory and economic hurdles to quickly ramping power capacity. Small modular nuclear reactors are one solution that gives companies the flexibility they'll need. SMRs have the added benefit of providing reliable carbon-free energy that aligns with Big Tech's emission-reduction goals. However, small reactors are a nascent, "next decade technology," Morgan Stanley analysts said. For that reason, cloud hyperscalers like Microsoft (MSFT), Alphabet (GOOG), Amazon (AMZN), and Meta Platforms (META) increasingly have turned to existing nuclear infrastructure during the AI buildout of the past few years. Meta on Tuesday signed a 20-year deal with Constellation Energy (CEG), America's largest nuclear power provider, to sustain its AI. Constellation and Microsoft last year agreed to bring back online a reactor at Pennsylvania's Three Mile Island. With SMRs still a ways off, new data centers are likely to rely on small, mobile natural gas generators from the likes of GE Vernova (GEV) and Caterpillar (CAT). Hyperscalers may also buy from fuel cell manufacturers like Bloom Energy (BE), whose electricity servers are a low-carbon way to convert natural gas, biofuel, or hydrogen into power. These fuel cells, the analysts said, offer the benefit of short lead times, reliable equipment, the ability to add redundant capacity in the event of a unit failure, and exceptional flexibility in terms of power output. "We believe [Bloom Energy] could quickly increase manufacturing capacity to ~3 GW per year, with the potential for further increases in output if demand grows," the analysts wrote. "Bloom Energy is in our view one of the under-appreciated beneficiaries of the rapid growth in data center power demand globally."

Nuclear Meets AI: Which Companies Will Reflect Gains First? | Investing.com UK

To make use of resources, the primary requirement is intelligence. The capacity to perceive reality and organize its assets determines the level of civilizational development. In this light, artificial intelligence is a historical breakthrough in which automated intelligence can be integrated into every aspect of civilizational development, serving as an amplifying force. Although current AI models are far from robust, it is already clear that AI should not be viewed as a bubble just waiting to burst. Rather, AI should be viewed as a membrane on top of existing digital infrastructure, connecting disparate nodes with AI agents. Case in point, within Google's existing ecosystem of apps, Agent2Agent (A2A) protocol allows AI to roam the infrastructure nodes to deliver a new level of usefulness. And there is no ceiling for AI integration and interoperability. If a thought can be had by a human, AI could interpret it. If a task can be envisioned, AI could pursue it. Ultimately, AI agents will join robotic platforms to complete tasks in the physical world. But the underlying requirement for the AI civilizational layer is power. For investors, this translates to exposure in key energy sectors. Although nuclear power took PR hits from the Chernobyl and Fukushima Daiichi disasters, alongside the Three Mile Island incident, it remains the most energy-dense, cost-effective, safe and reliable source of power generation. Having a more efficient government unbothered by PR, China has gone the farthest in harnessing nuclear power by developing the world's first 4th gen nuclear power plant in Shandong Province in late 2023. This is on top of the world's first-ever thorium-based reactor in the Gobi Desert, utilizing thorium, a more abundant radioactive element than uranium. At present, China is tied with France in the number of nuclear reactors, at 58 (56.9 GW capacity) vs 57 respectively, and still lagging behind 94 nuclear reactors in the US. However, by 2030, China is headed to the nuclear top with at least 30 new reactors under construction, adding 34.4 GW of capacity, per WNA figures. Consequently, just as the USG intervened in China's development by implementing strict export controls on AI chips, so will the federal government have to build up its power base for the AI layer. In Washington, there is already a bipartisan consensus that America has to win the AI race, expressed by the Treasury Secretary Scott Bessent at the Milken Institute. President Trump is completely aligned with this sentiment, having hosted Larry Ellison's Stargate project shortly after his inauguration, in addition to issuing executive orders (EOs) toward "Unleashing American Energy". The overall plan is to quadruple the United States' nuclear capacity over the next 25 years. "We've got enough electricity to win the AI arms race with China... What we do in the next five years related to electricity is going to determine the next 50 because it's the first time in history where electricity can be translated into intelligence." -- Interior Secretary Doug Burgum at the Oval Office on May 23rd, following a new series of President Trump's EOs. According to the International Energy Agency (IEA) report in April, the world will need 30% more energy on a yearly basis for accelerated servers (AI workloads). This sector accounts for half of the net increase in global data centers' energy consumption. And as main geopolitical players, China and the US are projected to account for 80% of that growth up to 2030. In early 2024, Amazon (NASDAQ:AMZN) was the first to set the trend, having tagged Talen Energy for its Susquehanna nuclear plant to deliver up to 960 MW of power. However, this plan was foiled by FERC, capping the plan by 3x to only 300 MW. The possibility still exists for Talen Energy to successfully appeal FERC's ruling, potentially meeting Amazon's immediate energy needs. Microsoft (NASDAQ:MSFT) had greater regulatory success, being the first to sign a 20-year deal with Constellation in September 2024. In fact, this involved restarting the aforementioned Three Mile Island's nuclear reactor in Unit 1. Constellation Energy (NASDAQ:CEG) is the primary beneficiary of AI demand, having the largest fleet of nuclear reactors. In October 2024, Alphabet's Google (NASDAQ:GOOGL) made a deal for small modular reactors (SMRs) provided by Kairos Power, still not available for public trading. Representing advanced technology also adopted by the EU, SMRs are to add 500 MW worth of power to the U.S. electricity grid. After the dampened Talen plans, in the same month, Amazon also launched its SMR ambitions with Energy Northwest, a publicly owned municipal corporation not available for public trading. Amazon delegated the responsibility of delivering up to 5 GW of power by 2039 to privately held X-Energy Reactor Company. Interestingly, Ken Griffin's Citadel, notable for its role in GameStop (NYSE:GME) short squeeze, helped finance X-Energy's SMR push with a $500 million round, as a part of Amazon's Climate Pledge Fund. Most recently on Tuesday, Constellation Energy gained Meta Platforms (NASDAQ:META) as another AI data center client, having inked a 20-year deal just like Microsoft. In short, the Big Three hyperscalers, together with Meta, will be tied to the American energy surge. Accordingly, investors should keep an eye on their investor relations' news section. And just like in the EU, it is clear that the future lies in scalable and compact SMRs. *** Neither the author, Tim Fries, nor this website, The Tokenist, provide financial advice. Please consult our website policy prior to making financial decisions.