US fuel cell leader Bloom Energy Corp. (BE.US) has surged over 30% in the past week, with shares climbing nearly 70% since the company reported earnings in late July, and an eye-popping 220% gain year-to-date. S&P Dow Jones Indices announced over the weekend that the company will join the S&P 500 Index on September 21, and on Tuesday's close, Bloom Energy shares jumped nearly 10% following the index inclusion news. Senior analyst Manav Gupta of UBS noted this will be the first energy stock added to the benchmark index since 2022.
From an options perspective, Bloom stands out not just among S&P 500 energy names but across the entire US equity market. Bloom Energy's options market activity has even surpassed that of SpaceX. As of Tuesday's midday trading session, total premium traded reached nearly $500 million, exceeding SpaceX's $350 million options volume and far outpacing Valero Energy's $25 million options turnover—the most volatile stock in the S&P 500 energy sector. This underscores the concentrated bullish positioning by global capital around the data center power bottleneck theme, amplified by the index inclusion catalyst.
Bloom Energy's recent investment appeal stems from the continued expansion of AI data center electricity demand, robust earnings growth, and the S&P 500 inclusion catalyst. The company's second-quarter revenue reached $1.0654 billion, up 165.5% year-over-year, with product revenue surging 215.4%. Full-year revenue guidance was raised to between $3.9 billion and $4.2 billion, implying roughly 100% year-over-year growth at the midpoint. This has shifted the market's pricing trajectory for Bloom around delivery scale and profit expansion for large AI data center power equipment.
Technologically, Bloom's key advantage is that new AI workloads require reliable power quickly. OpenAI's Astra model—dubbed by Nvidia CEO Jensen Huang as ushering in the "AGI era"—drives longer deep reasoning sessions, concurrent agents, and recursive self-improvement (RSI) involving parallel experiments, which increase cluster runtimes and deployment scales. As workloads grow faster than unit-task energy efficiency improvements, total data center power demand rises dramatically. For example, a 100-megawatt IT load requires approximately 120 megawatts of facility power capacity at a power usage effectiveness (PUE) of 1.2.
Bloom's core product, the Energy Server, uses solid oxide fuel cell (SOFC) technology to convert natural gas, biogas, or hydrogen into electricity through electrochemical reactions, providing on-site power for data centers, semiconductor factories, and other commercial facilities. Its commercial value lies in modular scalability, continuous power supply, and reduced grid interconnection wait times. However, it's important to note that fuel cells are not equivalent to zero-carbon emissions or energy storage—using natural gas still produces CO2, and fuel cells fundamentally differ from batteries used for AI data center energy storage.
Fuel cells focus on direct electricity generation through electrochemical reactions, suitable for modular, phased deployment, with advantages including lower noise, reduced water usage, and faster on-site installation. Large gas power plants burn natural gas to drive turbines, and in combined-cycle configuration, utilize waste heat to power steam turbines, achieving efficiency exceeding 60% in advanced systems. Tech giants continue building gas-fired plants for direct data center power supply, primarily because future gigawatt-scale mega-campuses require long-term, large-scale power, and the economies of scale, mature maintenance systems, and fuel efficiency of large plants remain attractive.
Bloom helps customers access electricity faster, but that doesn't mean it holds a long-term levelized cost of electricity advantage across all projects. The two approaches are complementary, with the final choice depending on delivery timelines, construction costs, fuel supply, and long-term reliability requirements. The end of AI is power supply: Bloom's transformation from fuel cell manufacturer to a key link in AI computing infrastructure's explosive expansion is converting this on-site power value into quantifiable demand at the order level.
Bloom's agreement with Oracle supports procurement of up to 2.8 gigawatts of fuel cell systems, with the initial 1.2 gigawatts already contracted and deployment underway. The company also disclosed it previously delivered and commissioned fuel cell systems for Oracle in just 55 days, more than a month ahead of the original 90-day schedule. The Power Connect product launched in August reduces on-site installation time by over 40% through factory pre-wiring, pre-integration, and testing. For high-value AI GPU cluster systems, earlier power availability and modular rapid power deployment mean monetizable computing capacity sooner, making "actual speed of power system delivery" a critical procurement criterion for customers.
The Astra release has heightened market expectations for artificial general intelligence (AGI), driving stronger computing demand trajectories. Notably, the new "pay-for-results" growth model could generate even more robust total computing demand, while more immediate evidence of computing demand comes from AI R&D itself—the "recursive self-improvement" trajectory where AI begins building AI. OpenAI recently disclosed that as of mid-August, every human workday in its research division corresponded to approximately 3.1 agent workdays running. At API prices, a median researcher consumes over $600 in inference services daily. These data points suggest AI is simultaneously expanding both commercial application inference and internal R&D inference demand, creating new demand sources for continuous training, evaluation, and experimentation. However, Bloom continues to emphasize that humans determine research direction and system deployment, and full RSI has not yet been achieved—this progress cannot be directly equated to RSI dominating global training.
Frontier labs are also expanding capacity procurement. Reports indicate Anthropic signed a $35 billion cloud computing agreement with Lambda involving approximately 350 megawatts of capacity in Texas, following a six-year, $45 billion computing rental arrangement with Nscale covering roughly 460 megawatts. Together, these agreements total $80 billion and approximately 810 megawatts, reflecting the intensity with which leading labs are securing resources for future training and inference. These are multi-year contract values and associated capacity figures, meaning future construction, equipment delivery, and power assurance needs continue showing strong expansion trajectories.
Stock markets are also showing signs of diffusion from chips to supporting infrastructure. On September 7, Samsung Electronics rose 5.68%, SK Hynix gained 8.26%, and Korea's KOSPI index—dubbed the "AI computing barometer"—surged 4.61% to 6995.39 points, a cumulative rebound of about 25.06% from its July 30 low of 5593.56, surpassing the typical technical bull market threshold. On the US side, the Philadelphia Semiconductor Index—covering Nvidia, AMD, Micron, Broadcom, and other semiconductor giants—rebounded over 20% from its July 29 low on August 13, entering a new technical bull market. After subsequent volatility, it rose another 1.30% to 11887.87 on September 8.
The power narrative's main axis is shifting toward fuel cells. Bloom Energy's options market heat even exceeds SpaceX: AI on-site power supply has become the focus of global capital. Crude oil has climbed back above $90 per barrel, and the S&P 500 energy sub-sector is up over 40% year-to-date, with the sector's stocks approaching new 52-week highs. Despite energy stocks being the S&P 500's best-performing sector in 2026, options traders seem unconvinced that traditional energy strength will persist. Instead, they're flooding into Bloom Energy, this ultra-hot alternative energy stock that's even more popular in the options pit than SpaceX—the space exploration and AI giant founded and led by Elon Musk. This fuel power provider, with a market cap of approximately $82 billion as of Tuesday's close, primarily sells on-site fuel cell systems tied to AI data center and AI industry chain power needs.
Bloom's stock has risen 1,800% over the past three years. According to Cboe LiveVol and SpotGamma data, the stock rose over 10% on Tuesday, with options volume surging to more than 2.5 times the 30-day average. Total premium traded approached $500 million by midday Tuesday, exceeding SpaceX's $350 million options volume and far surpassing Valero Energy's $25 million options turnover—the most volatile S&P 500 energy stock. Bloom shares are up over 30% in the past week and nearly 70% since the company reported earnings in late July. S&P Dow Jones Indices announced over the weekend that the company will join the S&P 500 on September 21.
According to ThinkOrSwim data, Bloom's current implied volatility exceeds 90%, significantly higher than any energy stock in the S&P 500. Valero Energy, the US refining giant, has the highest volatility in the sector at 50%, while ExxonMobil, America's largest oil and gas company, sits around 30%. On Tuesday, the most actively traded Bloom options contract was the Friday-expiring $300 strike call, trading at $4.65, requiring an additional 8% stock gain to reach breakeven. In contrast, options flows for the United States Oil Fund (USO) ETF and the Energy Select Sector SPDR Fund (XLE) ETF show mixed long-short positioning. USO options volume is 50% above the 30-day average, with turnover around $90 million. According to SpotGamma data, put and call contracts purchased are nearly identical in number, with the three highest-volume contracts all being puts. For XLE, an estimated 48,000 calls were purchased versus 34,000 puts; of the $37 million in total premium traded, $31 million involved calls. According to SpotGamma, three of the five highest-volume XLE contracts are puts.
At Citi's technology conference on September 8, AMD—Nvidia's strongest GPU competitor—reaffirmed that the addressable market for AI data center accelerated computing has expanded to $2 trillion by 2030, highlighting AI inference demand as the primary growth driver for AI computing resources. AI agents focused on agentic AI workflows simultaneously boost both GPUs and server CPUs. AMD also stated that its three Helios core customers—Meta and two other AI labs—have future procurement demand forecasts exceeding initial strategic partnership expectations. The company expects server CPU revenue to grow over 80% year-over-year in the second half of this year and over 70% next year. This latest projection combination strongly supports continued computing demand expansion, but it centers on significantly raised customer demand forecasts and shouldn't be entirely viewed as irrevocable computing infrastructure orders. Helios is AMD's rack-scale AI computing system, with Facebook parent Meta being one of its core customers.
Morgan Stanley's latest key trend analysis around Astra suggests that AI models' significantly enhanced capabilities make more workloads economically viable, thereby strengthening supply constraints in power, substrates, and memory manufacturing. In their scenario analysis, hyperscale cloud providers' computing deployment power capacity expands from approximately 35 gigawatts in 2025 to about 145 gigawatts by 2028—a 4.1x increase—with the US facing an approximately 38-gigawatt power gap. At an estimated $3 billion per gigawatt in behind-the-meter investment, this implies roughly $114 billion in potential additional project spending. Morgan Stanley's latest expectations indicate that as AI computing and data center power chain demand continues growing, AI computing infrastructure construction is entering the initial digestion and monetization phase of early investments. Bloom's stock market rally reflects the market trading the value of "powering up early and monetizing computing capacity sooner"—for instance, of Oracle's up to 2.8-gigawatt procurement framework, the initial 1.2 gigawatts are already contracted and accelerating deployment.
Morgan Stanley projects enterprise AI spending reaching $812 billion by 2027, with the share of S&P 500 companies able to quantify GenAI benefits expected to rise to 25% by Q2 2026 from 14% a year ago. This creates two investment themes: near-term focus on companies like Bloom that can unblock power bottlenecks and accelerate project commissioning; medium-term focus on how models, cloud platforms, and applications convert new computing capacity into actual profits. Bloom's options heat exceeding SpaceX reflects concentrated global capital trading on this power bottleneck beneficiary theme, amplified by the index inclusion catalyst. However, the nearly $500 million in premium traded doesn't equate to a net bullish inflow of equal magnitude. The $300 strike call with $4.65 premium and $304.65 breakeven price strongly indicates that stock market investors' upside expectations have already driven up trading costs. With active short-dated call trading and implied volatility above 90%, the market is paying substantial premiums for potential sharp moves, requiring even larger stock gains for momentum chasers to cover costs.
Comments