Recent thematic ETF data is sending an increasingly clear message about the AI infrastructure trade. Investors are not abandoning the long-term power-demand thesis; if anything, the underlying fundamental case is becoming stronger. What is changing is the way capital is being allocated within that theme. The market is increasingly distinguishing companies that are already converting electricity scarcity and AI capital spending into revenue, backlog and free cash flow from companies whose valuations still depend primarily on projects that may not generate commercial cash flow for several years.
That distinction is especially important in the current macro environment. Oil prices remain elevated as the U.S.-Iran conflict continues, Treasury yields are under pressure, and investors are waiting for Friday’s CPI report to clarify the Fed outlook. At the same time, AI spending remains exceptionally strong: TSMC reported a 53% increase in August revenue, reinforcing the view that hyperscale computing investment remains intact even as higher real rates challenge long-duration equity valuations.
The result is a market increasingly willing to fund AI hardware, electrical infrastructure and existing power assets, but less willing to pay indiscriminately for every company associated with AI, nuclear power or clean energy.
The Flow Data: Investors Are Funding the Physical AI Buildout
Semiconductors remain the clearest confirmation of that behavior. Semiconductor ETFs attracted approximately $2.97 billion during the latest week compared with $2.28 billion over the entire rolling one-month period. SMH alone attracted roughly $1.64 billion and SOXX another $1.34 billion. The category also gained about 5.1% on an AUM-weighted basis during the week after declining over the preceding month.
That is an important flow pattern. The market is not merely buying an AI narrative; it is allocating capital toward the part of the AI ecosystem where demand is already appearing in revenue. The same cannot be said for software, where the category lost approximately $434 million during the week after posting a small positive flow over the rolling month. IGV accounted for most of that deterioration with roughly $449 million of weekly redemptions.
Power infrastructure is beginning to show a similar pattern. Infrastructure ETFs attracted approximately $263 million over the latest week versus $236 million over the entire one-month window, even though the group remains down more than 4% on an AUM-weighted basis over the month. PAVE alone absorbed roughly $236 million during the week. That combination—weak trailing returns and accelerating inflows—looks much more like accumulation than performance chasing.
Electrification funds tell a similar, if less dramatic, story. GRID has received approximately $231 million over the past month, while the broader electrification cohort remains in positive flow territory despite recent price weakness. By contrast, broad Clean Energy ETFs lost roughly $198 million during the latest week, including approximately $160 million from ICLN.
The implication is important: investors are increasingly treating rising electricity demand as an infrastructure problem rather than simply a clean-energy theme.
AI Is Creating a Real Electricity Demand Shock
The fundamental reason is straightforward. AI data centers are changing a U.S. electricity system that had spent decades operating in a relatively slow-growth demand environment. EIA now expects U.S. electricity sales to reach a record 4,135 billion kilowatt-hours in 2026, almost 2% above 2025, and increase another roughly 2% to 4,211 BkWh in 2027. The agency specifically identifies data-center development and increased manufacturing activity as important drivers of commercial and industrial electricity consumption.
The longer-term estimates are even more significant. Department of Energy research estimates that data centers accounted for approximately 4.4% of U.S. electricity consumption in 2023 and could represent 6.7% to 12% by 2028. In energy terms, that would mean data-center consumption rising from approximately 176 terawatt-hours in 2023 to somewhere between 325 and 580 TWh only five years later.
This demand is different from ordinary economic growth because it is arriving in very large increments at very specific locations. A major AI campus can require hundreds of megawatts of reliable power, often on a development schedule much shorter than the time required to permit and construct new generation and transmission infrastructure. FERC highlighted precisely this issue in June when it ordered all six regional grid operators under its jurisdiction to justify or reform the rules governing connections for data centers and other large loads. The Commission explicitly described the challenge as delivering greater “speed-to-power.”
That concept may ultimately be one of the most important investment insights in the power-demand trade. In a capacity-constrained system, an existing megawatt that can be delivered in 2027 can be economically more valuable than a theoretically cheaper megawatt that might become available in 2031.
The Real Bottleneck Extends Far Beyond Power Plants
The power-demand opportunity therefore extends well beyond utilities or generating companies. Every incremental gigawatt has to travel through an enormous physical supply chain. Generation requires turbines and related equipment. New generating assets need transmission lines. Transmission requires substations and transformers. Data centers need switchgear, backup systems, electrical distribution and increasingly sophisticated thermal management because higher rack densities produce extraordinary amounts of heat.
That is why companies such as GE Vernova (GEV), Eaton (ETN), Quanta Services (PWR) and Vertiv (VRT) have become particularly important examples of what we would call the cash-flow side of the power-demand trade.
GE Vernova reported $24.2 billion of second-quarter orders, an 88% organic increase, while revenue rose to $11.1 billion. Its gas-power equipment backlog and slot reservations reached 116 GW, and the company generated $5.1 billion of free cash flow during the quarter—more than it produced during all of 2025. Those figures make the investment case tangible: customers are already reserving equipment and paying GEV to supply a system experiencing genuine capacity scarcity.
Eaton is benefiting one layer further downstream. Twelve-month rolling orders in Electrical Americas were up 41%, reflecting strong demand for the equipment necessary to distribute and control electricity in data centers, industrial facilities and other high-load applications. The company’s products are valuable regardless of whether the marginal electricity is ultimately produced by gas, nuclear, wind or solar.
Quanta Services provides another example. Second-quarter revenue reached $9.56 billion, compared with $6.77 billion a year earlier, while free cash flow approached $900 million and total backlog reached $53.4 billion. Quanta is being paid to build the transmission, distribution and utility infrastructure required to connect new loads to the system. It therefore participates in the demand cycle without having to predict which generation technology ultimately captures the largest share.
Vertiv is positioned even closer to the server rack. The company reported second-quarter sales of $3.27 billion, up 24%, while adjusted free cash flow reached $925 million. Its power-management and thermal-management products become more important as AI data centers increase computing density and therefore electrical and cooling requirements. Vertiv now expects approximately 31% organic sales growth for 2026.
These businesses can certainly become overvalued, and none is immune to a capex slowdown. But fundamentally they differ from concept-stage companies because AI power demand is already visible in their reported revenues, orders and cash flows.
GRID Illustrates the “Picks and Shovels” Version of the Trade
GRID is a useful ETF example of this distinction. Its largest holdings as of September 8 included Schneider Electric at 9.32%, Eaton at 8.88%, Johnson Controls at 8.42%, ABB at 7.88% and Quanta Services at 7.57%. Together, those businesses provide electrical equipment, automation, cooling, transmission infrastructure and other components required to expand and modernize the grid.
The attraction of that portfolio is that it does not require an investor to decide today whether the winning incremental power technology will be natural gas, conventional nuclear, SMRs, solar or some combination of all four. Transformers, switchgear, transmission systems and electrical-management equipment are required across almost every plausible scenario.
That helps explain why GRID has continued to attract capital even after declining approximately 3.4% over the preceding month. Investors appear willing to buy the infrastructure bottleneck on weakness rather than wait for momentum to turn.
Existing Generators Have an Advantage That Cannot Be Manufactured Quickly
Existing power generators represent another version of the cash-flow trade. Constellation Energy (CEG) already owns and operates nuclear assets capable of producing round-the-clock electricity, giving the company something a development-stage reactor company does not possess: power that can be contracted on a commercially relevant timetable.
Constellation reported second-quarter adjusted operating earnings of $2.55 per share, raised full-year guidance to $11.50–$12.50 per share and signed an additional 920 MW of long-term power purchase agreements. The economic logic is increasingly compelling. Hyperscalers do not simply need cheap theoretical power; they need dependable power delivered soon enough to support billions of dollars of data-center investment.
This creates scarcity value for operating plants. Existing generation can potentially monetize AI-related demand before a new gas plant, transmission project or next-generation reactor receives permits, financing, construction approval and final commissioning.
Concept Stocks Offer More Optionality—and Much More Execution Risk
At the other end of the spectrum are companies such as Oklo (OKLO), NuScale Power (SMR) and other advanced-reactor developers. Their potential addressable markets may be enormous if AI-related electricity growth forces utilities and hyperscalers to adopt new nuclear technologies at scale. But buying these companies today is economically very different from buying GEV, ETN, PWR, VRT or CEG.
The bulk of the value in a development-stage nuclear company lies in future events. Investors must underwrite regulatory approvals, final plant designs, fuel availability, financing, construction costs, customer agreements, successful commissioning and eventual commercial production. Each milestone can create enormous equity value if successful, but the underlying cash flows remain further away and less certain.
That makes these companies much more sensitive to both execution risk and interest rates. A dollar of cash expected many years from now is worth less when the discount rate rises. By comparison, a supplier filling a multi-billion-dollar order book today can reinvest or return cash while the power buildout is occurring.
This does not mean concept stocks cannot outperform. In fact, their optionality means they can produce extremely strong returns if commercialization probabilities improve. It simply means investors should recognize that they are underwriting a different financial proposition even though the stocks share the same “AI power demand” label.
Some Nuclear ETFs Combine Concept Stocks and Existing Cash Flows
The distinction gets particularly interesting at the ETF level because many nuclear funds deliberately blend these two investment cases.
NLR, the VanEck Uranium and Nuclear ETF, is a good example. As of September 8, its largest position was Constellation Energy at 9.09%, and it also held Public Service Enterprise Group at 6.03% and CGN Power at 4.22%. Those holdings provide exposure to existing generation and utility economics. At the same time, NLR held NuScale at 4.99%, Oklo at 4.06%, X-Energy at 2.34% and Nano Nuclear at 1.65%, giving investors meaningful exposure to future-reactor optionality as well.
SMRF, the ALPS Nautilus SMR, Nuclear & Technology ETF, provides another—and somewhat different—example of the same barbell. The fund is explicitly designed to cover the broader nuclear and SMR value chain rather than functioning simply as a portfolio of development-stage reactor companies. ALPS describes the strategy as combining nuclear and SMR exposure with a curated AI and technology sleeve, effectively linking growth in computing demand with the infrastructure required to supply firm power.
That Is Why Power Demand and Clean Energy Are Diverging
One of the more revealing signals in this week’s data is that the structural electricity-demand trade is strengthening even as traditional clean-energy funds struggle to retain assets.
ICLN gained 3.7% during the week but suffered approximately $160 million of redemptions. The Clean Energy category overall lost nearly $198 million during the week and roughly $99 million over the rolling month.
That divergence suggests investors are no longer treating electricity-demand growth as synonymous with renewable-energy adoption. A hyperscale data center ultimately needs reliable electricity; it does not necessarily care whether the marginal electron is produced by an existing nuclear plant, a new combined-cycle gas turbine, a solar project backed by storage or another source.
That shifts the investment emphasis toward availability, reliability and connection speed. GEV can benefit because utilities need new turbines. ETN can benefit because every system needs electrical distribution. PWR can benefit because somebody has to build the transmission and substations. CEG can benefit because it already owns generation. GRID can capture much of that infrastructure ecosystem in a single portfolio. NLR and SMRF can layer future nuclear optionality on top of existing power-generation and supply-chain exposure.
Higher Rates Are Raising the Bar for Thematic Investing
The current Treasury environment makes this distinction even more relevant. The September 10 morning headlines show long yields under continued upward pressure as investors weigh oil-driven inflation, fiscal concerns and skepticism that Treasury buybacks alone can materially relieve pressure at the long end.
That backdrop favors business models capable of producing near-term cash. It is consistent with the broader ETF flow data as well. Factor and quantitative strategies attracted approximately $250 million over the latest week and $1.23 billion over one month, while low-volatility ETFs received approximately $159 million during the week despite weak price performance.
The common denominator is not simply “defense.” It is cash-flow visibility.
Investors appear willing to own secular growth, but increasingly want evidence that the secular theme is moving through the income statement. Semiconductor demand has that evidence. Electrical-equipment companies have it. Transmission contractors have it. Existing generators have it. Many advanced nuclear developers may eventually have it, but today their valuations require investors to assume that a series of future milestones will be successfully completed.
Bottom Line
The most useful way to think about the AI power-demand trade is no longer as one monolithic theme. It is an ecosystem extending from semiconductor fabs and data-center servers through cooling equipment, switchgear, transformers, transmission lines, power generation, uranium and ultimately next-generation reactor technologies.
The strongest part of the current fundamental setup remains where demand has already become orders, backlog, electricity contracts and free cash flow. GEV, ETN, PWR, VRT and CEG are examples of businesses monetizing the shortage today. GRID provides diversified exposure to many of those physical bottlenecks.
NLR and SMRF occupy an interesting middle ground. Both combine portions of the existing nuclear and power system with exposure to companies whose value depends much more heavily on the future commercialization of advanced reactors. SMRF is particularly useful as an example because, despite its explicit SMR branding, its mandate includes traditional nuclear generation and transmission exposure alongside the emerging technology side of the trade.
The broader lesson from this week’s flows is that investors have not lost faith in thematic growth. They have simply become more demanding about how quickly the theme translates into cash. This is the equity side of the duration trade.
Sources
- ETFThemes.com September 10 thematic return and flow data for semiconductor, software, infrastructure, electrification, clean-energy and nuclear ETF flows and returns.
- September 10 Morning Headlines for oil, rates, AI demand and CPI/Fed context.
- U.S. Energy Information Administration — September 2026 electricity outlook
- U.S. Department of Energy — data-center electricity demand study
- FERC — large-load integration and speed-to-power initiative
- GE Vernova Q2 2026 results
- Quanta Services Q2 2026 results
- Vertiv Q2 2026 results
- Constellation Energy Q2 2026 results
- First Trust GRID holdings
- VanEck NLR holdings
- ALPS SMRF fund page and its published fact sheet describing the fund’s nuclear generation/transmission, reactor-development, uranium and AI sleeves.
Disclaimer: The information provided by ETFThemes.com is for informational purposes only and should not be construed as an offer or solicitation to buy or sell any security. All investments and investment strategies involve risk, and past performance is not indicative of future results. Opinions and analysis reflect judgment as of the publication date and are subject to change without notice. Thematic ETFs may contain companies with materially different business models, financial characteristics and stages of commercialization. References to cash-flow generators, concept stocks or similar classifications are analytical descriptions rather than formal investment categories. Fund holdings and weights can change, and investors should review current fund disclosures and conduct independent due diligence before making investment decisions.
