The Power Bottleneck of AI Computing Power: Investment Opportunities in Nuclear Energy, Geothermal, and Energy Infrastructure

In 2026, the global artificial intelligence industry is facing an unprecedented structural contradiction: computing power is growing exponentially, while the growth curve of electricity supply is lagging far behind.

According to the latest data released by Gartner in June 2026, global data center electricity consumption is expected to jump from 447 TWh in 2025 to 565 TWh in 2026, representing a 26% year-on-year increase. In the same period, global data center power demand is expected to rise from 104 GW to 132 GW, an increase of 27%. More importantly, the long-term trend is even more concerning. Gartner predicts that by 2030, data center power demand will exceed 290 GW; electricity consumption is expected to surpass 1200 TWh, at which point the grid will be unable to meet future data center construction demand.

Goldman Sachs’s data also corroborates this trend. According to Goldman Sachs forecasts, U.S. data center power demand will increase from 31 GW in 2025 to 41 GW in 2026, and then further rise to 66 GW in 2027—nearly doubling. This forecast is based on a reverse analysis of the construction progress of hyperscaler data centers in Silicon Valley. In 2026, the estimated new power-capacity added to U.S. compute centers is expected to reach 13.6 GW, and in 2027 it is expected to reach 36.3 GW.

The direct engine driving this growth is AI-optimized servers. Gartner’s data shows that the electricity consumption of AI-optimized servers will increase from 95 TWh in 2025 to 175 TWh in 2026, an 84% increase. By 2026, AI-optimized servers will account for about 31% of total data center power consumption; by 2027, their power consumption will officially surpass that of traditional servers.

However, the constraint on electricity is not only about overall quantity—it is also about the fragility of power distribution across both space and time. In May 2026, the Equatorial Central Pacific officially entered an El Niño state, which is expected to develop into moderate or stronger El Niño events during the summer and autumn. The average spring temperature across the 48 U.S. states reached 13.22°C, the second-highest spring average temperature recorded in 132 years of meteorological data. During summer, air-conditioning load can drive regional peak demand to surge by 20% to 30%, and the near-constant full-load operation of AI data centers will amplify this pressure. On May 18 of this year, the U.S. Department of Energy issued an emergency order authorizing the PJM grid to dispatch data center on-site backup generators in extreme cases to prevent residents from experiencing load shedding and power rationing.

Electricity is evolving from the “supporting infrastructure” behind computing expansion into a “decisive bottleneck.”

Three Technical Routes for Supplying Power to AI Data Centers

In the face of this bottleneck, the global energy and technology industries are racing along three technological routes.

Nuclear Energy: A Stable Option for Base-Load Power

Thanks to its stable base-load electricity output, nuclear power has become an important candidate for powering AI data centers. Small modular reactors (SMRs), with their advantages of smaller capacity, flexible deployment, and inherent safety, are especially well-suited to data center needs. Estimates show that between 2024 and 2030, China’s data center electricity consumption will reach 405.1 billion to 530.1 billion kilowatt-hours, meaning that the power constraint is evolving from an industry topic into an infrastructure bottleneck. Against this backdrop, technology companies have started engaging with nuclear power state-owned enterprises to explore supplying power to compute facilities via direct connection with small reactors. However, nuclear power solutions face practical constraints such as electricity pricing models, regulatory approvals, and construction timelines, making large-scale deployment in the short term uncertain.

Renewable Energy: Scale Expansion and Intermittency Challenges

Wind and solar power have already become significantly cost-competitive, and globally, project models combining “wind power + compute” and “solar + compute” are accelerating. For example, the Soya Green Data Center in Japan’s Hokkaido plans to build a 3MW grid-connection power intake capacity, directly connecting to a wind farm via dedicated transmission lines. In VivaTech 2026, VisionTech announced its “Mission Gobi” plan, aiming to deploy 5GW of green AI data center capacity in deserts and arid regions before 2030. However, the intermittency of renewable energy output fundamentally conflicts with the 7×24-hour constant load of data centers. Stable power supply requires pairing with large-scale energy storage facilities.

Geothermal: The Overlooked Clean Base-Load Energy

The unique value of geothermal energy lies in the fact that it combines the attributes of clean energy with base-load power capability—without being affected by weather, day and night, or seasons—so it can provide uninterrupted, stable electricity output around the clock. Unlike solar and wind power, geothermal generation does not have “zero-output” periods. This makes it naturally advantageous in AI data center power supply scenarios. For a long time, geothermal development has been constrained by the high cost and risk of underground resource exploration. Drilling depths can reach 10,000 feet underground, and underground rock temperatures can reach 555°F. Traditional geologic modeling methods are time-consuming and have limited accuracy, making it difficult to support large-scale development.

EGS-Twin: When AI Starts “Mining” Geothermal

On June 22, 2026, a collaboration that could potentially change the paradigm of geothermal energy development was officially announced.

Fervo Energy, a U.S. next-generation geothermal energy company (Nasdaq code: FRVO), NVIDIA, a global leader in AI compute, and the U.S. Pacific Northwest National Laboratory (PNNL) have reached an agreement to jointly develop the next-generation digital twin platform for enhanced geothermal systems (EGS)—EGS-Twin.

The core design goal of EGS-Twin is to integrate high-resolution on-site data, physics-based simulation modeling, and AI-driven prediction to provide real-time insights into underground reservoir behavior and operational performance. Under the collaboration framework, PNNL researchers will use Fervo’s industry expertise and on-site data to train scalable AI models on NVIDIA AI infrastructure. The trained models will then be integrated into the NVIDIA Omniverse library. PNNL will also develop workflows and data pipelines, running large-scale simulations using high-performance computing facilities, including supercomputing resources from the U.S. Department of Energy.

The project will start by using Fervo’s existing proprietary data from project sites in Nevada and Utah to train models, and will continuously improve as production data increases. The platform is expected to go live in 2029.

Jack Norbeck, Chief Technology Officer and co-founder of Fervo Energy, said: “Combining high-fidelity physical models with AI-driven prediction is expected to reshape reservoir management, improve thermal recovery, and enhance system reliability.”

The technical logic behind this collaboration is straightforward: the core bottleneck in geothermal development is “invisibility.” Underground fracture networks, hydrothermal fluid flow, and rock mechanics cannot be directly observed. Traditional modeling relies on limited seismic data and geological inference, which is time-consuming and has high uncertainty. The value of an AI digital twin lies in training with large volumes of on-site data to build underground models that can be updated in real time—helping operators more quickly identify and respond to underground changes, optimize power generation efficiency, and improve the scalability of EGS systems.

It is worth noting that EGS-Twin is not Fervo Energy’s only recent milestone. The company’s flagship geothermal project, Cape Station, located in Beaver County, Utah, is progressing through commissioning. The first batch of power from Phase 1 GeoBlock 1 is still planned to connect to the grid in Q4 2026, while GeoBlocks 2 and 3 are expected to follow in Q1 2027. After Cape Station reaches full load, it will have about 100MW of operating capacity (early 2027), with long-term plans to expand to 500MW. In March 2026, Fervo also completed a $421 million non-recourse project financing.

The capital markets responded immediately to this collaboration. Although Fervo’s first earnings report released on June 22, 2026 showed both revenue and earnings per share below analysts’ expectations (Non-GAAP EPS of -3.72 dollars, expected -0.07 dollars; revenue of $61,000, expected $489,600), the stock still rose more than 8% in pre-market trading on Monday ahead of the news driven by the EGS-Twin collaboration. Fervo completed a $2.2 billion IPO on Nasdaq in May 2026, issuing 80.5 million shares priced at $27.

NVIDIA’s Energy Infrastructure Deployment

Fervo’s partnership is just the tip of NVIDIA’s strategic iceberg in the AI energy infrastructure domain.

In May 2026, NVIDIA and IREN Limited announced a strategic collaboration, planning to support the deployment of up to 5GW of NVIDIA DSX architecture AI infrastructure across IREN’s global data center pipeline. As part of the deal, IREN issued NVIDIA five-year warrants that can be used to purchase up to 30 million common shares at an exercise price of $70, representing investment rights of up to $2.1 billion. Both sides expect future deployment to focus on IREN’s 2GW Sweetwater campus in Texas.

NVIDIA founder and CEO Jensen Huang said: “AI factories are becoming the foundational infrastructure of the global economy. Large-scale deployment of these systems requires deep integration across the full stack of compute, networking, software, power, and operations.”

In June 2026, NVIDIA further announced an agreement with South Korea’s SK Hynix, Naver, and Doosan Group to jointly build AI data centers. Among them, Naver and NVIDIA will jointly establish gigawatt-scale AI factories, with plans to start operations next year and an initial capacity of 55MW.

NVIDIA’s strategic path is clear: as a core supplier of global AI compute, the sustainability of its business model depends heavily on whether downstream data centers can secure sufficient power supply. By deeply engaging in energy infrastructure—whether through geothermal digital twins, large-scale AI factory collaborations, or regional data center alliances—NVIDIA is turning electricity from an “external variable” into a “controllable factor.”

Gate Stock Trading: Capturing Investment Opportunities in AI Energy Infrastructure

For investors who want to participate in this wave of AI energy infrastructure investment, the Gate platform offers a differentiated trading pathway.

On June 1, 2026, Gate officially launched real-stock trading services, becoming one of the first exchanges in the industry to directly connect to the U.S. stock market within a crypto platform. As of June 2026, Gate TradFi has listed more than 12,500 real stocks and ETFs, fully covering the five major mainstream exchanges: NYSE, Nasdaq, NYSE Arca, NYSE American, and BATS.

The core advantages of Gate stock trading are reflected in three dimensions:

First, extremely low participation barriers. Fractional-share trading starts at as low as 0.01 shares, and you can begin investing in U.S. stocks with just $1. This means investors can build portfolios that include AI energy infrastructure concept stocks such as NVIDIA (NVDA), Fervo Energy (FRVO), and more—without needing large capital.

Second, direct USDT settlement. Users do not need to go through the cumbersome process of “selling crypto → withdrawing fiat → cross-border remittance → broker deposits.” They can complete trades directly using the USDT in their Gate account. This mechanism eliminates the friction costs for crypto investors to participate in traditional stock markets.

Third, compliance and security safeguards. All stock trades are executed by the compliant broker-dealer Alpaca, which holds U.S. Broker-Dealer licenses and clearing qualifications. Behind it are real assets independently held in custody through the DTC system, and customers enjoy full SIPC protection.

In addition, Gate has launched 7×24-hour stock trading. Investors are no longer limited to the traditional trading window in U.S. Eastern Time from 9:30 to 16:00. When AI energy infrastructure-related news (such as announcements about Fervo and NVIDIA collaborations) is released outside trading hours, investors can respond immediately. Gate’s stock products have been fully integrated into the platform’s VIP tier system. Users only need to hold $2000 in positions to upgrade to VIP and enjoy an exclusive fee rate as low as 0.023%.

Conclusion

In 2026, the rapid surge in electricity consumption for AI data centers has pushed the “computing power–electricity” contradiction to a critical point. Gartner’s forecast of annual power consumption of 565TWh and Goldman Sachs’s prediction of 41GW in U.S. data center power demand—behind these numbers lies a clear industrial logic: the next phase of AI competition is not only about chip computing power, but also about the competition for energy supply.

Nuclear power’s stability, renewable energy’s scale efficiencies, and geothermal’s clean base-load characteristics each have their pros and cons. But geothermal’s unique value lies in simultaneously meeting the two rigid requirements of AI data centers: “clean” and “uninterrupted.” The EGS-Twin collaboration between Fervo Energy, NVIDIA, and PNNL applies an AI methodology to solve energy problems: reducing uncertainty in geothermal development with digital twins, shortening exploration cycles with accelerated computing, and optimizing power generation efficiency with data-driven approaches.

For investors, AI energy infrastructure is becoming an investment direction that cannot be ignored. From NVIDIA’s leading position in AI compute, to Fervo Energy’s geothermal technology breakthrough, to energy infrastructure providers such as IREN that supply AI factory-level infrastructure, structural opportunities exist across every segment of the industrial chain. Through USDT direct settlement for U.S. stock trading, extremely low fractional-share thresholds, and a 7×24-hour trading mechanism, the Gate platform provides a compliant and convenient channel for users in the crypto ecosystem to participate in this trend.

Electricity is the “new oil” of the AI era. Whoever holds the keys to power supply will control the initiative in the next industrial cycle.