AI data center electricity consumption surges by 26%: Can nuclear energy and fuel cells become the next major investment focus?

By 2026, the global AI data center electricity demand is reshaping the energy industry landscape at an unprecedented pace. The latest Gartner data shows that worldwide data center electricity consumption is expected to reach 565 TWh, a 26% year-over-year increase, with AI-optimized servers' power usage soaring from 95 TWh in 2025 to 175 TWh, an 84% jump. This growth rate far exceeds the expansion capacity of traditional power grid infrastructure, making electricity "availability" the primary bottleneck constraining computing power deployment.

On the supply side, two technological paths are attracting high market attention: nuclear energy solutions represented by small modular reactors, and distributed power solutions exemplified by solid oxide fuel cells. Bloom Energy (NYSE: BE), as a global leader in SOFC commercialization, reported a 130.4% YoY revenue increase to $751 million in Q1 2026, achieving its first net profit of $70.6 million, sparking widespread market discussion on "AI electricity" investment themes.

Meanwhile, major tech giants are accelerating their nuclear energy deployments, marking the start of the second wave of nuclear power construction worldwide.

Data Center Power Consumption—From Computing Bottleneck to Power Bottleneck

Over the past two years, global capital markets have focused heavily on Nvidia’s GPU inventory and advanced packaging capacity for AI infrastructure. But by 2026, a deeper supply-side constraint is emerging: the core bottleneck in AI computing infrastructure is shifting from chip supply to power supply. Goldman Sachs has listed energy availability as the top constraint for AI infrastructure, even surpassing chip supply chain tensions.

Data indicates this assessment is well-founded. Gartner predicts global data center power demand will grow 26% in 2026, reaching 290 GW by 2030. More notably, the demand structure is changing—by 2027, AI-optimized servers’ power consumption will surpass that of traditional servers for the first time, indicating that AI-driven power demand increments have already exceeded conventional digitalization needs.

On the supply side, traditional power grid expansion is lagging severely behind data center construction speeds. Data centers can be built and operational in as little as 8 months, whereas substation and transmission line projects typically take 5 to 13 years. According to a report from Guojin Securities citing PJM regional data, the average project takes over 7 years to connect to the grid. This time lag is creating an unprecedented "power bottleneck" globally—not in terms of cost, but in terms of availability.

The U.S. Department of Energy’s analysis further underscores the severity: by 2030, the U.S. needs an additional 100 GW of peak power capacity, with 50 GW directly for data centers. Of the 104 GW of power plants announced to be shut down, 210 GW of new generation capacity will be built to replace them, but only 22 GW of this is from stable, round-the-clock power sources. The gap in reliable, dispatchable clean power is projected to reach 78 GW.

The core issue is that wind and solar, despite being zero-carbon, are intermittent due to natural conditions and cannot support the base load operation of AI data centers running 24/7. The zero-tolerance for downtime in data centers makes stable, dispatchable clean electricity a rigid demand.

Nuclear Power—Long-term Solution vs. Short-term Challenges

Nuclear power, with over 90% capacity factor and stable all-weather output, is beginning to occupy a unique position among power options for AI data centers. From 2024 to 2026, leading U.S. tech companies have made strategic shifts in power procurement—from predominantly wind and solar green power agreements to direct power purchase agreements (PPAs) with baseload-stable nuclear plants.

In Q1 2026, Meta signed three nuclear PPAs within a month: partnering with Oklo to develop a 1,200 MW advanced nuclear tech park; reaching a 2,609 MW power purchase agreement with Vistra; and investing in TerraPower’s 690 MW sodium-cooled fast reactor project. Microsoft signed a 20-year PPA with Constellation Energy, exclusively purchasing all 835 MW output from the Crane Clean Energy Center (formerly Three Mile Island), with a total investment of about $3 billion, supported by a $1 billion loan from the U.S. Department of Energy, aiming for grid connection by 2028. Amazon also signed a 1.92 GW PPA with Talen Energy and invested in advanced nuclear reactor developer X-energy, aiming to deploy up to 5 GW of small modular reactors (SMRs) in the U.S. by 2039. As of March 2026, U.S. tech giants have accumulated about $74.5 billion in nuclear power orders.

In China, the trend is similarly notable. By the end of 2025, China’s nuclear capacity reached 61 GW. In April 2025, the State Council approved 10 nuclear units simultaneously, the highest approval volume in over 15 years for the first half of the year. The China Nuclear Energy Industry Association expects an annual approval of 8–10 GW of nuclear capacity during the 14th Five-Year Plan, reaching 110 GW operational by 2030 and 200 GW by 2040.

In June 2026, news emerged that Alibaba had approached state nuclear enterprises about building small nuclear reactors for the Hangzhou Renhe data center. This aligns with the trend among U.S. tech giants, but domestic SMR deployment still faces practical constraints such as electricity pricing and supply modes.

However, scaling nuclear solutions faces significant time lags. SMRs typically have a single-unit capacity of no more than 300 MW, with factory pre-fabrication and modular deployment allowing completion within 12–24 months, but overall construction still takes 3–5 years, and large-scale grid connection has yet to be achieved globally. The nuclear industry has been dormant for over 30 years, facing aging equipment and a shortage of skilled workers; from 1990 to 2025, overseas nuclear capacity increased by only 108.1 GW, with a compound annual growth rate of just 0.7%.

This time lag means that before large-scale SMR deployment, data center operators will need to rely on other distributed power sources to fill short-term gaps.

Fuel Cells—Key Pathway to Short-term Power Gap Filling

Given the long grid expansion cycles and delayed nuclear grid integration, solid oxide fuel cells (SOFCs), with their modular design and rapid deployment capabilities, are gaining a competitive edge. SOFC systems can deliver 50 MW in 90 days and 100 MW in 120 days—practical deployments include Oracle’s 55-day project.

Technically, SOFCs achieve a pure power generation efficiency of up to 65%, with combined heat and power (CHP) efficiencies of 85–95%, surpassing traditional gas turbines. They produce 800V DC directly, eliminating multiple AC/DC conversion stages physically, saving approximately $1.35–1.5 billion in capital expenditure for 1 GW of AI data center power. SOFCs operate with zero water consumption, near-zero NOx emissions, and noise levels around 65 decibels, suitable for community deployment.

Industry developments in 2026 further validate this path. On June 11, Samsung Heavy Industries announced plans to commercialize a 50 MW floating data center powered by seawater-cooled SOFCs using LNG fuel, with approvals from American Bureau of Shipping and Lloyd’s Register. The platform can dock at ports and connect to the grid, and when external power is unavailable, the SOFC system supplies power independently.

In China, progress is also tangible. Qingneng Co. recently launched fuel cell units designed for data center backup and auxiliary power, with 100% higher power density than other PEM fuel cells. JH2 Technology’s fuel cell products have been used in Egypt’s first hydrogen-powered emergency data center, providing 2 hours of continuous power. Guojin Securities’ research highlights the promising outlook for the SOFC industry chain, indicating the sector has entered a “from 1 to 10” scaling phase.

Regarding subsidies, under the IRA framework, SOFCs can enjoy a 30% Investment Tax Credit (ITC), potentially up to 50% with local manufacturing and energy community conditions. Current SOFC system costs are about $2,075 per kW, with the U.S. Department of Energy aiming to reduce this below $900 per kW by 2030. Amid rising gas turbine prices due to supply shortages, post-subsidy generation costs are approaching parity.

Bloom Energy—Core Stock in the AI Power Investment Theme

Bloom Energy (NYSE: BE) is the most representative listed company in this trend. The company specializes in solid oxide fuel cell systems, mainly serving data centers, hospitals, manufacturing plants, and other high-reliability power needs.

Bloom Energy’s Q1 2026 financial report significantly exceeded market expectations. The company reported revenue of $751 million, up 130.4% YoY. Product revenue was $653 million, up 208.4%. Gross margin increased from 27.2% to 30.0%, with non-GAAP gross margin at 31.5%. Net profit attributable to shareholders was $70.6 million, reversing a loss of $23.8 million in the same period last year. Operating cash flow was $73.6 million, up $184.3 million YoY.

The company also raised its full-year guidance, with the revenue growth midpoint increased to about 80%, above the previous 60%. The Q1 backlog of products is about $6 billion (up 140% YoY), with service backlog around $14 billion. The company has prepared for a 5 GW manufacturing capacity.

In stock performance, since 2026, Bloom Energy’s share price has surged by over 198%. On June 9, trading volume reached $4.22B, a 92.2% increase from the previous day. Recently, the stock experienced short-term volatility: on June 10, BE fell about 10%, mainly due to news that Crusoe Wyoming’s data center project was paused. The project planned to deploy 900 MW of Bloom fuel cells, but Crusoe suspended development at the client’s request. Morgan Stanley maintained an “Overweight” rating with a $310 target, emphasizing that the pause does not alter the long-term narrative of AI power demand. RBC Capital reiterated an “Outperform” rating with a $335 target.

Market consensus on Bloom Energy is a “Moderate Buy,” with 9 buy and 9 hold ratings. The average target price is $266.56, about 12.47% above the current price. Analysts’ consensus for 2026 full-year EPS is $1.31, while the company’s guidance is between $1.85 and $2.25, reflecting differing views on the pace of AI power deployment.

Gate Stock Trading—USDT Direct Access to the AI Power Sector

Gate stock trading’s core mechanism allows users to directly use USDT in their accounts to trade stocks and ETFs listed on major U.S. exchanges like NYSE and Nasdaq within the platform, without currency exchange, cross-border remittance, or opening a separate brokerage account. As of June 2026, Gate supports over 10,000 U.S. stocks and ETFs across five major exchanges.

Cost-wise, trading fees are as low as 0.023%, with no platform fees, commissions, or hidden charges. Unlike traditional CFDs or perpetual contracts, Gate stock spot trading has zero holding costs—no funding rates, no rollover fees, no overnight charges. Dividends and dividends are automatically paid in USDT to user accounts.

From an asset allocation perspective, Gate stock trading enables crypto investors to diversify across markets within a single platform, bridging digital assets and traditional stocks. For the AI power theme discussed here, investors can search for related tickers like BE (Bloom Energy), CCJ (Cameco, uranium miner), CEG (Constellation Energy, nuclear operator), SMR (NuScale Power, SMR developer), and trade using USDT.

The operation involves four steps: holding or acquiring USDT in Gate, navigating to the “TradFi” section, selecting “Stocks,” transferring USDT to the stock account, and entering the target ticker to place buy orders during trading hours.

Conclusion

The power demand of AI data centers is shifting from an external computing bottleneck to a structural investment theme on the supply side. In 2026, data center electricity use is projected to grow 26%, while traditional grid expansion takes over a decade—creating a supply-demand mismatch that opens clear market space for nuclear and fuel cell solutions. Bloom Energy’s 2026 Q1 revenue growth of 130% and $6 billion backlog mark this business logic moving from concept to performance.

However, multiple uncertainties remain. Commercialization of SMRs faces hurdles in technology maturity, regulatory approval, and economic viability; scaling fuel cell production also involves risks; and the pace of data center deployment and power demand growth will directly influence the realization of these investment themes.

For crypto investors, Gate’s platform lowers the barrier to participating in global stock markets. Using USDT to directly access AI power-related stocks allows investors to grasp this structural trend without leaving the crypto ecosystem.