The end of AI is power supply! The options popularity of Bloom (BE.US) surpasses SpaceX, and fuel cells are fully benefiting from the super cycle dividend of AI computing power.

Bloom Energy Corp. (BE.US), a leader in fuel cells in the United States, saw its stock price surge more than 30% in the past week. Since the company announced its financial results in late July, the cumulative increase has approached 70%, and the stock has skyrocketed an astonishing 220% year-to-date. The S&P Dow Jones Indices announced last weekend that the company will join the S&P 500 Index on September 21. Benefiting from the news of this upcoming inclusion, Bloom Energy's stock price rose nearly 10% as of Tuesday's close in the U.S. stock market. Senior analyst Manav Gupta from UBS Group AG stated that this will be the first energy stock to be included in this benchmark stock index since 2022. From an options perspective, Bloom will stand out significantly among S&P energy stocks and in the broader U.S. stock market. The activity in Bloom Energy's options market has even far exceeded that of SpaceX, with total premiums nearing $500 million by Tuesday afternoon. This surpasses the $350 million in options volume for SpaceX and greatly exceeds the $25 million in options volume for the most volatile S&P 500 energy stock, Valero Energy Corporation, highlighting a concentrated bullish trading logic from global capital benefiting from the data center power supply bottleneck, compounded by the catalyst of index inclusion. The recently strengthened investment appeal of Bloom Energy primarily stems from the ongoing expansion of AI data center power demand, robust earnings growth, and the influx of funds anticipated from its inclusion in the S&P 500. The company's revenue in the second quarter reached $1.0654 billion, a 165.5% year-on-year increase, with product revenue soaring 215.4% year-on-year. The full-year revenue guidance has been raised to $3.9 billion to $4.2 billion, representing an approximate 100% year-on-year growth in terms of the median. This has led the markets pricing trajectory for Bloom to become increasingly focused on the delivery scale and profit expansion of power supply equipment for large AI data centers. From a technical standpoint, the key benefit to Bloom is the urgent need for reliable power to support new AI workloads. The OpenAI Astra model, heralded by NVIDIA Corporation's CEO Jensen Huang as ushering in the "AGI era," drives long-duration deep reasoning, concurrent intelligent agents, and parallel experiments involving recursive self-improvement (the so-called RSI), which will significantly increase the operational duration and deployment scale of clusters. When the growth of workloads exceeds the efficiency improvements of individual tasks, the total power demand of data centers will increase drastically. For example, a 100-megawatt IT load at a Power Usage Effectiveness (PUE) of 1.2 requires about 120 megawatts of facility power capacity. Bloom's core product, the Energy Server, utilizes solid oxide fuel cells (SOFC) to convert natural gas, biogas, or hydrogen into electricity through electrochemical reactions, providing on-site power generation for data centers, semiconductor factories, and other commercial and industrial enterprises. Its commercial value lies in modular expansion, continuous power supply, and reduced waiting time for grid access. However, it is important to note that fuel cells do not equate to zero carbon emissions and energy storage; using natural gas still generates carbon dioxide, and fuel cells are fundamentally different from battery storage for AI data centers. Fuel cells (SOFC) focus on generating electricity directly through electrochemical reactions, making them suitable for modular and phased deployment, with advantages such as lower noise, less water usage, and faster on-site installation. In contrast, large gas power plants burn natural gas to drive gas turbines and, in combined cycle operations, re-utilize waste heat to drive steam turbines, with advanced systems achieving power generation efficiencies exceeding 60%. However, technology giants are actively building natural gas power plants to directly power data centers, primarily because future, larger-scale gigawatt-level parks will require long-term and extensive power supply, where the economies of scale of large units, mature operational systems, and fuel utilization efficiency remain attractive. Bloom can help customers obtain power more quickly, but this does not mean it will have long-term advantages in the lifetime cost of electricity across all projects; these two routes are complementary but ultimately depend on delivery cycles, construction costs, fuel supply, and long-term reliability requirements. The end of AI is power supply: Bloom is transitioning from a fuel cell manufacturer to a key element in the explosive expansion of AI computing infrastructure. Order levels have already translated the value of the on-site power supply technology led by Bloom into quantifiable demand. Bloom has a deal with Oracle Corporation to supply up to 2.8 gigawatts of fuel cell systems, with the first 1.2 gigawatts already contracted and deployment initiated. The company also disclosed that it previously delivered and commissioned a fuel cell system for Oracle within 55 days, over a month ahead of the originally scheduled 90 days. The Power Connect launched in August reduces on-site installation time by more than 40% through pre-wiring, pre-integration, and testing at the factory. For the high-value AI GPU cluster power systems, earlier power availability and modular rapid on-site power supply equipment mean that saleable computational power can be formed sooner, making "the actual delivery speed of the power supply system" an important purchasing criterion for customers. The launch of Astra has heightened market expectations for Artificial General Intelligence (AGI), thereby creating a stronger trajectory for power demand, especially with the new growth model of "paying for results" likely to generate even stronger overall computational power requirements. Recent direct evidence of computational power demand comes from the AI R&D process itselfspecifically, AIs initiation of building AI through the recursive self-improvement (RSI) research trajectory. OpenAI recently disclosed that as of mid-August, for every human workday invested by its research department, about 3.1 agent operation workdays corresponded. According to API pricing, researchers using the median agent amount consumed over $600 in inference services daily. These data indicate, to some extent, that AI is simultaneously expanding the demand for commercial application inference and internal R&D inference, providing a new source of demand for continuous training, evaluation, and experimentation. However, the company has made it clear that human decisions still dictate research directions and system deployment, and the complete RSI has not yet been realized; this progress cannot be directly equated with RSI having taken over global training. The capacity acquisitions of leading laboratories are also expanding. According to media reports, Anthropic has signed a $35 billion cloud computing agreement with Lambda, involving approximately 350 megawatts of capacity in Texas; prior to that, it reached a six-year, $45 billion computational power leasing agreement with Nscale, corresponding to about 460 megawatts of capacity. The two agreements total $80 billion and approximately 810 megawatts, reflecting the intensity of leading laboratories locking down resources for future training and inference. These amounts and corresponding capacities under multi-year contracts indicate that future construction, equipment delivery, and power assurance demands are continuing to show a strong expansion trajectory. Signs of a shift from chips to supporting infrastructure are also emerging in the stock market. On September 7, Samsung Electronics rose 5.68%, and SK Hynix rose 8.26%. The Korea composite stock price index, known as the "AI computational power barometer," significantly increased by 4.61% to 6,995.39 points that day, rebounding approximately 25.06% from the July 30 level of 5,593.56 points, exceeding the technical bull market threshold typically employed. The Philadelphia Semiconductor Index, which encompasses semiconductor giants like NVIDIA Corporation, AMD, Micron, and Broadcom Inc., also rebounded more than 20% from its July 29 low during intraday trading on August 13, entering a new round of technical bull market; after subsequent fluctuations, it rose again on September 8 by 1.30% to 11,887.87 points. The narrative around electricity is shifting towards fuel cells! The options market for Bloom Energy has even surpassed that of SpaceX: AI on-site power supply has become the focal point for global capital. With crude oil prices back above $90 per barrel, the S&P 500 Index energy sub-sector has risen over 40% this year, and stocks in this sector are approaching new 52-week highs. Although energy stocks are the best-performing sector in the S&P 500 Index for 2026, options traders seem uncertain about the sustainability of the strength in traditional energy assets. They are flocking instead to Bloom Energy, a highly popular alternative energy stock that is even more favored in the options market than SpaceX, the space exploration and AI giant founded and led by Elon Musk. As of Tuesday's close, this fuel power supplier, with a market capitalization of approximately $82 billion, mainly sells on-site fuel cell systems related to the power demand of AI data centers and the AI industrial chain. Bloom's stock has risen 1,800% over the past three years. According to data from Cboe LiveVol and SpotGamma, the stock rose more than 10% on Tuesday, with options trading volume surging to more than 2.5 times the average over the past 30 days. As of Tuesday afternoon in the U.S. stock market, the total premiums reached nearly $500 million, exceeding the $350 million in options volume for SpaceX and far exceeding the $25 million in options volume for Valero Energy Corporation, the most volatile energy stock in the S&P 500. In the past week, Bloom's stock price increased by more than 30%, and since the company announced its earnings report in late July, it has cumulatively risen nearly 70%. The S&P Dow Jones Indices announced last weekend that the company will join the S&P 500 Index on September 21. According to ThinkOrSwim data, Bloom's current implied volatility exceeds 90%, significantly higher than that of any energy stock in the S&P 500. Within the sector, Valero Energy Corporation, a powerhouse in the American refining industry, has the highest volatility at 50%, while the largest oil and gas giant, Exxon Mobil Corporation, is around 30%. On Tuesday, the most actively traded options for Bloom were call options expiring Friday with a strike price of $300, trading at $4.65, requiring the stock price to rise another 8% to break even. In contrast, the options flow for the U.S. Oil Fund (ETF code: USO) and the SPDR Select Sector Fund Energy ETF (ETF code: XLE) shows a mixed sentiment. USO's options trading volume is 50% above the average over the past 30 days, with a turnover of approximately $90 million. According to SpotGamma's data, the number of purchased put options is nearly equal to that of call options, and the three most traded contracts are all puts. In terms of XLE, about 48,000 call options are estimated to have been purchased, compared to about 34,000 puts; of the $37 million in total premiums, $31 million is related to call options. According to SpotGamma's data, three of the five most traded contracts for XLE are puts. AMD, a powerful competitor to NVIDIA Corporation in the GPU space, reiterated at the Citigroup Tech Conference on September 8 that the market scale associated with AI data center accelerated computing has expanded to $2 trillion by 2030, noting that the demand for AI inference is now the primary growth source for AI computing resources, with AI agents focusing on task-oriented AI workflows driving demand for both GPUs and server CPUs. AMD stated at the conference that three of its core customers in the Helios framework, including Meta and two other AI labs, have projected future procurement demand that exceeds initial expectations set during the establishment of their strategic partnership; the company expects its server CPU business to grow over 80% year-on-year in the second half of this year and over 70% next year. This latest forecast undoubtedly strongly supports the continued expansion of computing power demand, but because it focuses on significantly upward revisions of customer demand expectations, it cannot all be viewed as irrevocable orders for computing infrastructure that have already been placed. Helios is AMD's rack-level AI computing system, and Meta, the parent company of Facebook, is one of its core customers. Wall Street financial giant Morgan Stanleys recent judgments and key trends surrounding Astra suggest that the significant enhancement of capabilities in AI large models has made more workloads economically viable, thereby strengthening constraints on the supply of electricity, substrates, and storage manufacturing. In their scenario estimates, the power capacity corresponding to the computational power deployment of ultra-large-scale cloud providers is expected to grow from approximately 35 gigawatts in 2025 to about 145 gigawatts in 2028, reaching about 4.1 times the original; the U.S. faces a projected power shortfall of around 38 gigawatts. With estimates of about $3 billion in required Behind-the-Meter investment for every additional gigawatt of power, this corresponds to approximately $114 billion in potential additional project investment. Morgan Stanleys latest expectations indicate that as demands for AI computing and data center power chains continue to grow, the construction of AI computing infrastructure is gradually entering an initial phase of digesting and monetizing early investments. The market's enthusiasm for Bloom, as a stock that can alleviate power supply bottlenecks and accelerate project production, reflects the value of sooner power delivery, sooner formation of sellable computing power in this process; for instance, in Oracle Corporation's procurement framework of up to 2.8 gigawatts, the first 1.2 gigawatts has already been contracted and accelerated deployment processes initiated. Predictive data from Morgan Stanley anticipates that enterprise AI spending will reach $812 billion by 2027, and the proportion of S&P 500 companies that can quantify the benefits of GenAI (Generative AI) is expected to rise from 14% a year ago to 25% by the second quarter of 2026. Morgan Stanley foresees the formation of two investment tracks: a short-term focus on companies like Bloom that can relieve power supply bottlenecks and accelerate project deployment; and a mid-term focus on how models, cloud platforms, and applications can convert new computing power into actual profits. Bloom's options heat surpassing SpaceX reflects a concentrated trading logic from global capital benefitting from this power supply bottleneck, compounded by the catalyst of index inclusion. However, nearly $500 million represents the total premium volume, which does not equate to a large-scale influx of net bullish capital; a call option with a strike price of $300 and a premium of $4.65 has a breakeven price of $304.65, suggesting that investors in the stock market already anticipate an increase in the stock price, raising transaction costs. The active trading of bloom's short-term bullish options, along with an implied volatility above 90%, indicates that the market is paying a high premium for potential large fluctuations, meaning that those chasing the stock price need significant increases to cover their costs.