June 20, 2026 · 12 min read
AIPO targets the overlooked side of AI investing: not the chips or the models, but the power grid, transmission infrastructure, and data center hardware that has to be built before a single inference can run. With $550M+ AUM and the ETF.com Best New Thematic ETF award, it has caught investor attention — here is what is inside it.
AIPO is the Defiance AI & Power Infrastructure ETF, launched on July 24, 2025 — one of the first ETFs to explicitly link the AI buildout to its underlying power infrastructure requirements. The fund is managed by Defiance ETFs, a boutique thematic ETF provider known for forward-looking sector bets.
AIPO's core thesis is straightforward: AI models require electricity, and the US power grid needs a massive, multi-decade upgrade to support the AI data center buildout. The ETF invests in both sides of this equation — the companies building AI compute infrastructure (data centers, cooling, power distribution) and the companies generating and transmitting the electricity those facilities consume.
To qualify for inclusion, holdings must derive at least 50% of their revenue from AI and power infrastructure activities. This screen keeps the fund focused — it's not a broad utilities ETF, it's specifically targeting the AI power demand story. The fund holds 78 companies as of mid-2026, spanning utilities, industrials, REITs, and energy.
Since its July 2025 launch, AIPO has attracted $550M+ in assets — remarkable for a fund less than a year old — and won the ETF.com Best New Thematic ETF award for 2026. Its YTD return of approximately +28% through June 2026 reflects the market's recognition that AI power demand is not hypothetical; it is here now and accelerating.
The scale of AI power demand is unlike anything the US grid has planned for. A single ChatGPT query uses approximately 10× more electricity than a Google search. Scale that across billions of daily AI interactions, and the power math becomes staggering.
A single 100-MW AI data center — and hyperscalers are building dozens of these — draws roughly the power of 80,000 US homes simultaneously. Microsoft alone announced a $100B multi-year capex plan for AI infrastructure. Google, Amazon, and Meta are running comparable programs. The power demand created by these plans is not a future projection; it is showing up in utility earnings calls, transformer order books, and FERC interconnection queues right now.
The US grid was not designed for this. Most of the existing transmission infrastructure was built in the 1970s–1990s for a very different demand profile. The consequence: a data center wanting to connect to the grid faces a 3–5 year FERC interconnection queue backlog. Transformers have 2–3 year lead times. Grid construction labor is in shortage. AIPO's holdings are the companies solving these bottlenecks.
| Ticker | Company | Weight | Role in the AI power thesis |
|---|---|---|---|
| VST | Vistra Corp | 8.5% | Nuclear + gas power generation; largest competitive power producer in the US; runs Comanche Peak nuclear plant |
| CEG | Constellation Energy | 8.2% | Largest US nuclear operator; signed landmark Microsoft/Three Mile Island deal; sells 24/7 clean power to hyperscalers |
| PWR | Quanta Services | 7.8% | Largest US electrical contractor; builds high-voltage transmission lines and grid interconnections for data centers |
| GEV | GE Vernova | 7.3% | Gas turbines, wind turbines, and grid equipment; massive order backlog from data center electrification |
| NRG | NRG Energy | 6.9% | Competitive power generation and retail energy; growing data center power purchase agreements |
| ETR | Entergy | 5.8% | Regulated utility serving the US Southeast; significant data center load growth in its territory |
| EQIX | Equinix | 5.2% | Largest data center REIT; interconnection infrastructure that carries AI model traffic between clouds |
| WLDN | Willdan Group | 4.6% | Energy efficiency and grid modernization engineering; helps utilities handle new AI load growth |
| CLFD | Clearfield | 3.9% | Fiber optic infrastructure connecting data centers; backbone of AI inference delivery networks |
| AME | AMETEK | 3.4% | Electronic instruments and power management systems; precision components used in data center power infrastructure |
Several ETFs now compete in the AI power and grid infrastructure space. Here is how they stack up:
| Ticker | Fund Name | Exp Ratio | AUM | Focus | Top Holding |
|---|---|---|---|---|---|
| AIPO | Defiance AI & Power Infrastructure | 0.75% | $550M+ | AI data center power + grid infrastructure | VST / CEG |
| GRID | First Trust Nasdaq Clean Edge Smart Grid | 0.58% | $1.1B | Smart grid & electric infrastructure broadly | Itron / AMETEK |
| AMPS | Accelerate Power ETF | 0.49% | $180M | Power generation for AI and data centers | VST / NRG |
| PSIF | Principal Power & Energy Infrastructure | 0.65% | $90M | Utilities + pipelines + power infrastructure | NEE / SO |
AIPO is the most explicitly AI-focused of the group — its 50% revenue screen from AI/power infrastructure is stricter than GRID's broader smart grid mandate. AMPS is the closest competitor on thesis; GRID is more established and liquid. AIPO's higher ER (0.75%) vs GRID (0.58%) reflects its narrower, more curated mandate.
The most surprising development in the AI power story is the nuclear renaissance. AI hyperscalers — Microsoft, Google, Amazon, and Meta — have all signed or are pursuing nuclear power purchase agreements for a simple reason: nuclear provides 24/7 baseload power with zero carbon emissions. Wind and solar cannot guarantee 24/7 availability. Data centers need 99.9999% uptime.
The Microsoft / Three Mile Island deal. In September 2024, Microsoft signed a 20-year power purchase agreement with Constellation Energy to restart Unit 1 of Three Mile Island — the infamous plant that partially melted down in 1979. Renamed the Crane Clean Energy Center, it came back online in late 2024. This single deal established that hyperscalers are willing to go to extraordinary lengths to secure clean, 24/7 power.
Constellation Energy (CEG) is AIPO's second-largest holding for this reason. As the largest US nuclear operator with 25% of US nuclear capacity, CEG is uniquely positioned to sell 24/7 clean power to data centers. Its stock rose 3× in 2024–2025 largely on the back of hyperscaler power demand.
Small Modular Reactors (SMRs). Beyond existing plants, companies like Oklo (backed by Sam Altman) and NuScale are developing SMRs — compact nuclear reactors that can be sited directly adjacent to data centers. Oklo went public via SPAC and has signed letters of intent with multiple data center operators. These are long-dated speculative bets (first SMRs unlikely before 2030–2032), but they represent the next chapter of the AI power story.
Vistra (VST), AIPO's largest holding, combines nuclear with natural gas — giving it multiple "always-on" power sources that data centers prize above all else.
The US electric grid needs more than $2 trillion in upgrades over the next two decades — and AI data centers are accelerating that timeline. The grid was built for a world of centralized coal and gas plants serving residential load. Today it must accommodate distributed renewables, EVs, and now massive data center clusters that can draw 500 MW from a single site.
Transformer shortage. Large power transformers — the critical hardware that steps voltage up and down across the grid — are facing a 2–3 year backlog. There are fewer than a dozen manufacturers globally capable of producing the largest units. GE Vernova, Eaton, and ABB are running at full capacity. Data center operators are reserving transformer capacity years in advance.
Quanta Services (PWR) is AIPO's largest position outside of power generators. Quanta is the largest US electrical contractor — the company that physically builds transmission lines, substations, and grid interconnections. Every new data center that needs to connect to the grid is a Quanta Services contract. The company has $30B+ in backlog as of 2026, driven primarily by grid expansion for data centers and renewable interconnection.
FERC interconnection queue backlog. The Federal Energy Regulatory Commission manages the queue for new power generators and large loads connecting to the grid. As of 2026, the queue has over 2,000 GW of proposed projects waiting for approval and grid studies — much of it from data centers. This creates both a bottleneck (slow connection timelines) and an opportunity (for grid infrastructure companies like Quanta and Willdan that speed up the process).
While nuclear provides 24/7 clean baseload, natural gas plays a different role in the AI power ecosystem: on-demand flexibility. Gas peaker plants can go from cold to full output in 15–30 minutes, making them ideal backup power when renewable generation drops or demand spikes unexpectedly at a data center campus.
Hyperscalers are increasingly building data campuses with on-site gas turbine generators as backup — not just UPS systems, but full turbines capable of running the entire facility for days. This is driving demand for GE Vernova's aeroderivative gas turbines and NRG's distributed generation products.
Beyond domestic gas, the AI boom is accelerating US LNG export infrastructure. European and Asian countries seeking energy security are bidding for US LNG — and the AI electricity demand means US natural gas producers face rising domestic demand competition. NRG Energy, one of AIPO's larger holdings, is positioned on both the power generation side (competitive wholesale electricity) and the retail energy side (selling power directly to commercial and industrial customers including data centers).
Solar and wind energy are cheaper per kWh than ever before — but they have a fundamental limitation for data centers: intermittency. The sun doesn't shine at night. The wind doesn't blow on calm days. A data center that goes offline for even minutes loses millions in SLA penalties.
Hyperscalers are buying renewable energy at scale — Microsoft has committed to 100% renewable energy matching. But "matching" is an annual accounting construct. In practice, when the sun isn't shining and the wind isn't blowing, the data center is drawing power from the same fossil fuel grid as everyone else.
This is why nuclear and gas remain central to the AI power thesis. They provide the 24/7 baseload that solar and wind cannot reliably deliver. AIPO's holdings reflect this reality: the largest positions are in nuclear (CEG, VST) and gas (NRG, GEV) — not solar developers. Grid infrastructure (PWR, ETR) ties everything together, carrying power from wherever it is generated to wherever it is needed.
The conclusion for AIPO investors: this is not a clean energy ETF. It is an AI infrastructure ETF that happens to include significant nuclear and gas exposure because those are the only technologies that can meet data center power requirements today.
Power Purchase Agreements (PPAs) are long-term contracts between electricity generators and large consumers — in the AI era, between nuclear/gas plants and hyperscalers. These deals are transforming the economics of AIPO's top holdings.
Microsoft / Constellation (Three Mile Island). Microsoft signed a 20-year PPA with Constellation Energy to purchase 100% of the output from the restarted Three Mile Island Unit 1 (now the Crane Clean Energy Center). The deal is priced at approximately $115–120/MWh — roughly 2–3× the spot wholesale price in PJM, the mid-Atlantic electricity market. Microsoft is paying a premium for the guarantee of 24/7 carbon-free power. For Constellation, this deal locked in decades of predictable revenue on a plant that was about to be retired.
Google / Kairos Power. Google signed a deal with Kairos Power to purchase electricity from planned SMR (small modular reactor) plants — the first commercial SMR power purchase agreement in history. Google has committed to purchasing electricity from 500 MW of SMR capacity by 2035. This validates the SMR market and benefits the nuclear supply chain companies AIPO holds.
Amazon / Talen Energy. Amazon signed a data center co-location deal with Talen Energy to build a 960 MW data center campus adjacent to the Susquehanna nuclear plant in Pennsylvania — the largest data center campus ever proposed. FERC initially rejected it on grid reliability grounds, but the structure established a new model: nuclear plant + data center co-located for maximum efficiency.
Why PPAs matter for AIPO investors. The companies in AIPO's portfolio — particularly Constellation (CEG), Vistra (VST), and NRG — are increasingly transitioning from commodity power sellers (subject to volatile spot prices) to contracted power sellers with long-term agreements. This transformation reduces earnings volatility and supports premium valuations. Each new hyperscaler PPA announcement is a direct catalyst for these stocks.
AIPO trades on Nasdaq under the ticker AIPO. It is available at all major US brokers including Fidelity, Schwab, TD Ameritrade, Interactive Brokers, and Robinhood. There is no minimum purchase requirement beyond one share.
Position sizing. Given AIPO's thematic concentration and relatively small AUM ($550M), it is best sized as a satellite position — typically 3–8% of a total portfolio. Investors who want broader AI exposure without the power-specific concentration may combine AIPO with a broader AI ETF (like BOTZ or ROBO) or simply hold individual positions in its top holdings directly.
Building AIPO yourself. The top 10 holdings represent roughly 55% of AIPO's weight. An investor with $50,000+ could replicate significant exposure by buying VST, CEG, PWR, GEV, NRG, ETR, and EQIX directly — avoiding the 0.75% ER. The tradeoff: you must rebalance manually, do not get the long tail of 68 smaller holdings, and spend more time managing the position. For most investors, paying 0.75% for managed exposure is the right choice.
Tax location. AIPO's income (from utility dividends) is likely mostly qualified dividend income — suitable for taxable accounts. The ETF's growth-oriented utility holdings generate relatively modest dividend distributions. AIPO can be held in taxable, traditional IRA, or Roth accounts — it does not have MLP or REIT pass-through complications at the ETF level.
Comparing AIPO to individual stock picks. If you already own PWR, CEG, or GEV individually with high conviction, adding AIPO creates significant overlap. Check your portfolio's existing utility and industrial exposure before sizing. A useful rule: if your existing holdings overlap more than 25% with AIPO's top 10, the marginal diversification benefit is limited.
The most common question investors ask about AIPO is: "Why not just buy NVIDIA?" It is a fair question. NVIDIA is the dominant supplier of AI training GPUs — if AI infrastructure is growing, NVIDIA should benefit most directly. Here is an honest comparison of the two approaches:
The sophisticated answer: own both. NVDA for maximum AI upside with higher volatility. AIPO for picks-and-shovels infrastructure exposure with less valuation risk and more earnings predictability. AIPO's correlation to NVDA during sell-offs is meaningfully lower — in a scenario where AI chip demand disappoints, NVDA could fall 40–50% while AIPO's utility holdings cushion the portfolio.
AIPO occupies a specific niche in a portfolio — it is not a core holding, but a thematic satellite position for investors who have a view on AI infrastructure spending.
AIPO's approximately +28% YTD return through June 2026 — against an S&P 500 up roughly +11% — reflects several overlapping catalysts that hit AIPO's holdings simultaneously.
The +28% return has not come evenly distributed — it reflects a series of discrete catalysts rather than smooth appreciation. This pattern is typical of thematic ETFs: long stretches of consolidation punctuated by sharp moves on specific news. Investors who bought after the March 2026 run-up paid meaningfully higher prices than January buyers. Dollar-cost averaging into a position like AIPO (rather than a lump sum) helps smooth entry across different catalyst periods.
To size the opportunity correctly, it helps to understand the scale of what is being built. The current generation of AI data center investment is not a short-term surge — it is a fundamental re-plumbing of the US and global power and computing infrastructure.
These numbers represent a demand shock that plays out over 10–20 years, not 1–2 years. AIPO's top holdings — particularly grid constructors like Quanta and power equipment makers like GE Vernova — have order books extending 3–5 years. The revenue visibility is unusually high for equity investments. This is the core of the AIPO bull case: it is not speculative, it is contracted.
AIPO is one of the most interesting thematic ETFs launched in 2025–2026 because its thesis is both specific and verifiable: AI requires power, the grid needs upgrading, and specific companies are being paid to do that work right now. Unlike speculative AI ETFs that own pre-revenue companies, AIPO holds utilities, contractors, and power producers with real earnings, real order books, and real cash flows.
The main concern is valuation — VST, CEG, and PWR have already run hard on this thesis. New investors are not buying cheap. The 0.75% ER is also a meaningful headwind versus building a custom basket of the same stocks directly.
For investors who want broad, managed exposure to the AI power infrastructure theme without picking individual utility and industrial stocks, AIPO is the cleanest vehicle available. Size it as a thematic satellite — 3–8% of a portfolio — alongside core equity exposure, not in place of it.