WisdomTree has introduced a new financial product that truly deserves more attention. Not due to its structural novelty—it's still an ETF, a standard vehicle—but because it signals a critical divergence in how Wall Street is interpreting investment in artificial intelligence. While the mainstream market clings to buying shares of Nvidia and indirectly, OpenAI, this new ETF zeroes in on infrastructure: data centers, liquid cooling providers, specialized chip manufacturers for inference, and the energy companies that power these megawatt giants. Interestingly, many in the market have overlooked this dimension.
Photo: Igor Omilaev on Unsplash
The thesis is simple: if AI is set to consume 10% of global electricity by 2028—as some Morgan Stanley analysts project—then the real value lies not just in who trains the models, but in who enables those models to operate. WisdomTree is betting that the next decade belongs not so much to research labs but to those who build the digital and physical highways for those labs on an industrial scale.
The Hidden Strategy Behind the Timing: Why Now and Not Two Years Ago
WisdomTree is no rookie in thematic products. It has been launching ETFs related to cryptocurrencies, semiconductors, and disruptive technology for years. However, it waited until 2026 to launch a vehicle dedicated exclusively to AI infrastructure. Why? Because the market has matured, moving away from simple speculation about who will win the large model race.
In 2023 and 2024, the prevailing narrative revolved around LLMs. All institutional funds wanted a piece of Microsoft, Google, and Anthropic. But that story is changing. Training costs have skyrocketed: training GPT-4 cost over $100 million, and GPT-5 will likely exceed $500 million. Larger models need more GPUs, more bandwidth, more cooling. And that's where WisdomTree saw its opportunity.
The ETF includes companies like Vertiv (cooling systems), Equinix (data center colocation), ASML (advanced semiconductor manufacturing), and Schneider Electric (energy management). These names are less flashy than OpenAI or Anthropic, but they are the ones capturing the margin when a 175-billion-parameter model needs continuous inference for 18 months.
The Mistake Retail Investors Make: Confusing Hype with Infrastructure
Many investors see "AI" and rush to buy Nvidia stock. However, Nvidia has already reached a market capitalization of over $3 trillion. Its P/E multiple is at 65x. Honestly, how much more can it grow without its growth justifying the valuation? WisdomTree is betting that the next value multiplier lies in the lower layers of the stack: the undersea cables connecting data centers across continents, the renewable energy systems powering clusters of 50,000 GPUs, the construction companies specializing in buildings with extreme thermal dissipation.
This ETF also includes fiber optics companies and low-latency network providers. When Meta trains Llama 4 in a distributed cluster between Virginia and Oregon, network latency matters as much as raw computing power. If you lose 200 microseconds per hop, your training slows by 8%. That 8% loss translates to weeks of engineering time and millions in operating costs.
Which Companies Are In and Why They Matter More Than Nvidia Now
Photo: Steve A Johnson on Unsplash
WisdomTree's ETF doesn't reveal its exact composition until the first quarterly rebalance, but leaked regulatory documents suggest significant weighting in three categories:
1. Cooling and Energy Equipment Manufacturers
Vertiv and Schneider Electric are leaders in this space. Vertiv reported a 42% year-over-year growth in liquid cooling systems sales in Q4 2025. The reason is simple: AMD's H100 and MI300 chips generate so much heat that conventional air conditioning is no longer viable. Companies are switching to direct-to-chip cooling, which requires entirely new infrastructure. Meanwhile, Schneider Electric is selling complete microgrids to hyperscalers looking to power 200MW data centers without relying on public power grids.
2. Colocation and Connectivity Providers
Equinix and Digital Realty are the invisible landlords of AI. Equinix operates over 240 data centers in 70 countries. When a European startup needs low-latency inference for users in Singapore, it doesn't build a data center from scratch: it rents racks at Equinix. Digital Realty reported that 38% of its new contracts in 2025 were driven by AI clients, up from 12% in 2023. Demand is booming, and these players capture the rent without taking on technological risk.
3. Specialized Semiconductor Manufacturers
Here's where the strategy sharpens. WisdomTree doesn't just buy Nvidia. It also includes companies like Marvell Technology and Broadcom, which manufacture network switching chips and custom inference accelerators. Broadcom designed Google's TPU v5. Marvell makes the switches connecting 10,000 GPUs in a single training pod. These companies don't have Nvidia's glamour, but their margins are equally high, and their demand is guaranteed as long as models keep growing.
The Case of ASML: The Bottleneck No One Mentions
ASML manufactures the EUV lithography machines that make 3nm chips possible. Each machine costs $200 million, and only ASML produces them. TSMC, Intel, and Samsung are 100% dependent on ASML for the chips powering AI. If ASML faces a production issue or a geopolitical blockade (China is already off its client list), the entire AI supply chain would collapse. WisdomTree is betting that ASML will have unlimited pricing power over the next decade.
The Risks WisdomTree Isn't Shouting About But Exist
No financial product is perfect, and this ETF has two structural risks that the prospectus barely mentions.
Risk 1: Overexposure to Cyclical Capex
Infrastructure companies require a lot of capital. If Microsoft, Google, and Meta decide to reduce their data center spending for two consecutive quarters—something that happened in 2022 during the post-pandemic contraction—stocks of Vertiv, Equinix, and Schneider could drop 30% in a few months. The ETF isn't geographically diversified enough to mitigate this risk. 68% of its exposure is to companies with more than 50% of revenue coming from North American clients.
Risk 2: The Colocation Business Model Could Break
Equinix and Digital Realty charge for rack space and electricity. But if hyperscalers decide to build their own data centers—like Meta is already doing in Texas with a 500MW complex—the colocation business would lose its best clients. Equinix already reported that Meta reduced its footprint by 12% in 2025. If that trend generalizes, 40% of the ETF's portfolio is at risk.
The Paradox of Energy Efficiency
There's a subtler risk that almost no one is modeling: AI is improving its energy efficiency faster than expected. GPT-4 needs 10x less energy per token than GPT-3. If this trend continues, demand for energy and cooling infrastructure might hit a lower ceiling than projected. Companies producing 200MW systems might find their clients only need 80MW. That would destroy the growth projections justifying current valuations.
How It Compares to Other AI ETFs: Why This One Stands Out
Numerous AI ETFs already exist. The Global X Robotics & Artificial Intelligence ETF (BOTZ) has been around for years. The ARK Autonomous Technology & Robotics ETF too. So what makes WisdomTree's different?
Pure Focus on Physical and Digital Infrastructure
BOTZ includes software companies, industrial robotics, and generic semiconductors. WisdomTree excludes software entirely. There is no exposure to Salesforce, Adobe, or Palantir. It's 100% companies that build or maintain the physical and network infrastructure layer that allows AI to function.
Absence of Overvalued Mega-Caps
ARK has 15% in Tesla, betting on Full Self-Driving as AI. WisdomTree doesn't include Tesla. It also doesn't include Apple. It focuses on companies with P/E ratios between 20x and 35x, avoiding the valuation traps that sank many tech funds in 2022.
Agnostic Quarterly Rebalancing
Most AI ETFs are passive and replicate pre-built indices. WisdomTree uses an active selection committee that can rebalance exposure each quarter based on hyperscaler capex data. If Google announces a 20% data center spending cut, WisdomTree can adjust its Equinix exposure in the next rebalance without waiting for the underlying index to react.
Why AI Founders Should Pay Attention to This Move
This ETF isn't just a product for institutional investors. It's a market signal. When WisdomTree stakes an initial $500 million in AI infrastructure, it's saying: "We believe the next decade isn't about who has the best model, but who controls the resources that make training and deploying those models possible."
If you're the founder of an AI startup, this trend has direct implications:
1. Infrastructure Costs Are Going Up, Not Down
The 2022 narrative was that the cloud would get cheaper with growing competition. However, if WisdomTree is right and infrastructure demand outpaces supply, AWS, GCP, and Azure prices will rise. Startups optimizing their architecture to minimize GPU and bandwidth usage will have a significant competitive advantage.
2. Regionalization Matters More Than Ever
If energy and cooling costs spike in Virginia and Oregon, hyperscalers might shift workloads to regions with cheaper electricity and cooler climates. Startups designing their systems for geographic inference distribution—Iceland, Norway, Canada—can cut operational costs by 40%.
3. Colocation Contracts Are the New Competitive Edge
If you can negotiate a direct contract with Equinix or Digital Realty, you have guaranteed access to infrastructure even if the market tightens. Startups waiting until the last minute to scale infrastructure will face 18-month waitlists for high-density racks.
The Real Play: WisdomTree Is Betting Against the Software-Defined Narrative
The subtext of this ETF is a philosophical bet: that AI can't be fully virtualized. For 20 years, the tech industry believed software was eating hardware. But AI is flipping that narrative. You can't train a trillion-parameter model with just elegant abstractions in Kubernetes. You need real watts, real cooling, real fiber optic cables.
WisdomTree is betting that the next generation of unicorns won't be pure software companies but those that control their physical infrastructure vertically. Tesla doesn't just design autonomous driving software; it manufactures its own inference chips. OpenAI doesn't just train models; it's negotiating directly with energy utilities to secure 500MW commitments. Vertical integration is back, and companies that exist solely in the cloud will be displaced by competitors owning their full stack.
Is This the End of the Cloud-Native Era?
Not exactly. But we're seeing the limits of abstraction. When your app requires 100ms guaranteed latency and your model is 500GB, you can't rely on AWS's shared infrastructure. You need your own metal, your own network, your own energy strategy. Startups that grasp this sooner will have a five-year advantage over those still believing "the cloud is infinite."
To Close: Wall Street Finally Understands AI Is Physical, Not Just Code
WisdomTree isn't launching an ETF because it thinks AI is a fad. It's launching it because it finally understands that infrastructure is the true moat in this industry. Models can be replicated, data can be scraped, but the 200MW of dedicated energy and liquid cooling systems that allow a cluster of 50,000 GPUs to operate can't be copied over a weekend.
If you're a founder, investor, or simply someone interested in understanding where institutional money in AI is headed, this move by WisdomTree is more revealing than any keynote from OpenAI or Google. Smart money isn't betting on who trains the best model. It's betting on who controls the oxygen those models need to breathe.
Does your startup have a plan for when inference costs double in the next 18 months?