The OpenAI-Anthropic IPO Wave Could Pull the Entire AI Infrastructure Trade Into Focus — Some Possible Protagonists: $VRT / $ETN / $CEG

The artificial intelligence trade is entering a more physical phase.

For the last several years, the market has mostly treated AI as a software, semiconductor and cloud story. The center of gravity was obvious: large language models, GPUs, hyperscalers, enterprise adoption, AI copilots, model performance, inference costs and the rising arms race between the largest technology platforms.

That story is not over. Far from it. But the next leg of the AI trade may be less about the model itself and more about the infrastructure needed to keep the model alive.

Artificial intelligence does not run in the abstract. It runs inside data centers. Those data centers need land, permits, fiber, transformers, substations, switchgear, backup power, thermal management, water or advanced cooling, long-term power contracts, grid interconnection, and, increasingly, dedicated sources of firm electricity.

That is why the expected IPO paths of OpenAI and Anthropic matter far beyond the private AI-company universe.

Anthropic announced on June 1, 2026, that it had confidentially submitted a draft S-1 registration statement to the U.S. Securities and Exchange Commission for a proposed initial public offering. OpenAI later confirmed that it had also submitted a confidential S-1, while clearly stating that it had not decided on timing and that remaining private for some period may still be useful to the company.

Once private AI leaders begin moving toward public-market visibility, investors usually do not wait passively for the final listing date. They start looking for liquid proxies. They look for companies that already trade, already report earnings, already have public financials and already sit somewhere inside the value chain that the new IPO narrative could make more visible.

That is exactly what happened in the space trade.

SpaceX did not need to be directly available to every public-market investor for the space economy to become a momentum theme. The mere idea of SpaceX as the dominant private anchor created a reference point. It made the broader ecosystem easier to understand. It pulled attention toward listed proxies: launch providers, satellite companies, space infrastructure names, defense-space contractors, lunar economy plays and other companies positioned around the same theme.

OpenAI and Anthropic could do something similar for AI. But with one major difference: the AI infrastructure chain is much broader and much more energy intensive.

SpaceX pulled attention toward space infrastructure. OpenAI and Anthropic may pull attention toward the infrastructure of intelligence itself: compute, cloud, power, data centers, nuclear, gas, grid construction, electrical equipment, cooling, fiber, storage, onsite generation and backup power.

The IPO catalyst is bigger than the IPO itself

The public-market importance of OpenAI and Anthropic is not limited to their potential market capitalizations. Yes, the valuations could be enormous. Yes, the IPOs could become landmark Wall Street events. Yes, institutional investors would finally get official financial disclosures, risk factors, customer concentration details, revenue growth metrics, cost structure, compute spending, partnership exposure and capital needs.

But the deeper market signal would be this: AI model companies are no longer only venture-backed technology stories. They are becoming infrastructure-scale industrial actors.

That shift changes the way investors think. When a private company is still opaque, the market can tell almost any story about it. When it moves toward an IPO, the story becomes more financial, more measurable and more comparable.

Investors will not only ask how fast revenue is growing. They will ask how expensive that revenue is to serve. They will ask how much gross margin is consumed by compute. They will ask how dependent the model company is on hyperscaler cloud partners. They will ask how much capex is required to keep scaling. They will ask whether inference demand is accelerating faster than the infrastructure required to support it.

They will also ask harder physical questions. How much power is required? How much data center capacity is already locked up? How fast can grid interconnection happen? What happens if local communities push back? What happens if regulators force hyperscalers and large-load customers to absorb more of the grid-upgrade cost?

Those questions do not point only to OpenAI and Anthropic. They point to the companies around them.

The AI ecosystem needs chips. But chips need servers. Servers need racks. Racks need power distribution. Power distribution needs switchgear. Switchgear needs transformers. Data centers need cooling. Cooling needs electricity. Electricity needs generation. Generation needs grid connection. Grid connection needs transmission, substations, permitting, engineering, construction and regulatory approval.

That is the real AI stack. For public-market investors, the listed opportunity set may therefore extend well beyond the usual mega-cap AI names.

The numbers are now too large to treat as a footnote

The power issue is no longer a minor side note in the AI story.

The International Energy Agency estimates that global electricity consumption from data centers could roughly double by 2030, reaching around 945 TWh in its base case. That would represent just under 3% of global electricity consumption. The IEA also expects data center electricity use to grow much faster than electricity demand from other sectors, with accelerated servers driven mainly by AI adoption growing especially quickly.

The IEA also stresses a crucial point for investors: the absolute global share can look manageable, but data centers concentrate in specific locations. That makes local integration into the grid potentially more challenging than the headline global percentage suggests.

Goldman Sachs Research has estimated that global power demand from data centers could rise 165% by 2030 compared with 2023 levels. In the United States, Goldman has also pointed to a rapid expansion in construction spending and strong leased-data-center occupancy across most U.S. markets.

McKinsey frames the buildout as one of the largest infrastructure cycles in modern history. In a March 2026 analysis, McKinsey estimated that global spending on data centers could reach $7 trillion by 2030, while warning that the buildout will depend on capital availability, energy resources and long-lead industrial equipment capacity.

Berkeley Lab, in a report produced for the U.S. Department of Energy, found that U.S. data center load growth had tripled over the past decade and was projected to double or triple by 2028. The report estimated that data centers consumed about 4.4% of total U.S. electricity in 2023 and could consume between 6.7% and 12% by 2028, depending on broader electricity growth.

Reuters reported on June 18, 2026, that the Federal Energy Regulatory Commission ordered U.S. grid operators under its jurisdiction to reconsider how they connect very large energy users, including data centers. The proposed orders give grid operators and transmission owners 60 days to justify current rules or propose changes, with a focus on speeding connections while protecting grid stability and power bills.

Texas is moving separately. Reuters also reported on June 18, 2026, that the Public Utility Commission of Texas approved ERCOT’s first “Batch Zero” process for large electricity users, grouping qualified projects of at least 75 MW into a single study. ERCOT said it was tracking more than 438,000 MW of large-load requests, nearly 89% of them from data centers.

From AI software to AI industrialization

The first AI trade was easy to understand. Buy the obvious compute leaders. Buy the hyperscalers. Buy the software platforms with AI narratives. Buy anything that could plausibly benefit from enterprise AI adoption.

The next phase is harder, but potentially more durable.

AI industrialization requires physical capacity. A model can be copied. A data center campus cannot be copied overnight. A GPU cluster can be ordered, but it still needs power. A new data center can be announced, but grid interconnection may take years. A company can sign a cloud deal, but the cloud provider still needs to secure generation, transmission, transformers, cooling systems and backup capacity.

This is why the AI infrastructure trade may become broader than the AI software trade. The model layer may receive the attention, but the constraint may sit underneath it.

The bottleneck creates a second layer of listed proxies. Some are already obvious winners. Some are crowded. Some are cyclical. Some are speculative. Some may disappoint if valuations outrun fundamentals. But together they form a public-market map of the AI infrastructure economy.

The right way to analyze this theme is not to say that every name in the chain is automatically attractive. That would be lazy. The right way is to separate the chain into layers, then monitor who has real orders, real backlog, real pricing power, real balance-sheet capacity and real exposure to the bottlenecks that matter.

The AI infrastructure watchlist: public-market proxies across the chain

The following table is not a buy list, not a model portfolio and not a recommendation. It is a public-market map of U.S.-listed companies that investors may monitor as the OpenAI and Anthropic IPO narratives develop and as the market continues to reprice the infrastructure required to scale artificial intelligence.

Some of these companies are direct beneficiaries. Others are indirect beneficiaries. Some are large, diversified and lower-beta. Others are small, volatile and highly sensitive to retail momentum. The common thread is that each one touches a part of the AI infrastructure chain.

Tier 1: the direct infrastructure beneficiaries

The cleanest part of the trade is not necessarily the most exciting part, but it may be the most important. This is where names such as Vertiv, Eaton, GE Vernova, Quanta Services, EMCOR, Powell Industries and Sterling Infrastructure enter the discussion.

These companies do not need to own the best AI model. They do not need to compete with OpenAI or Anthropic. They do not need to win the chatbot interface. Their relevance comes from the fact that the AI boom cannot scale without the physical systems they provide.

A modern AI data center is not just a building filled with GPUs. It is an electrical and thermal-management machine. High-density AI racks create extreme power and heat challenges. As workloads move from traditional cloud computing toward training, fine-tuning, inference and agentic AI, the intensity of the physical infrastructure increases.

That creates demand for liquid cooling, power distribution units, UPS systems, switchgear, grid equipment, transformers, substations and specialized electrical engineering. Vertiv has become one of the most recognizable pure plays because it sits directly in power and thermal infrastructure for data centers. Eaton is broader, but its electrical systems, power management and switchgear exposure make it a core AI infrastructure name.

GE Vernova is broader still, but its grid and power-equipment exposure gives it a central position in the electrification theme. Quanta and EMCOR represent the construction and installation layer: the companies that help turn power demand into actual physical projects.

This matters because AI demand is moving faster than utility timelines. A model can go viral in days. A data center can be announced in months. But a transmission line, substation or power interconnection can take years. That timing mismatch creates the bottleneck. And bottlenecks are where public-market narratives often concentrate.

Tier 2: power supply, grid scarcity and firm electricity

The next layer is power itself. This is where the AI trade becomes uncomfortable for anyone who wants a clean, simple narrative.

The world wants AI to run on clean energy. Hyperscalers want to sign renewable power purchase agreements. Governments want innovation without emissions. Local communities want jobs without higher power bills, water stress or noise. Utilities want to serve new demand without destabilizing the grid. Regulators want speed without cost shifting. Investors want growth without stranded assets.

Those objectives do not always fit neatly together.

AI data centers need large amounts of electricity, and they need it reliably. Intermittent generation can help, but high-density AI workloads require firm power, backup capacity and grid stability. That is why nuclear, natural gas, storage and transmission all enter the discussion.

Constellation Energy has become a central public-market name because nuclear power offers large-scale, carbon-free, firm electricity. Vistra has attracted attention because of its combination of power generation, gas exposure, battery storage and nuclear assets. NRG, Southern Company, American Electric Power, Dominion, Exelon, NextEra and CenterPoint all sit somewhere inside the utility and regional load-growth debate.

Dominion is especially relevant because Virginia has become one of the most important data center markets in the world. If AI growth strains power availability in data center corridors, utilities in those regions become unavoidable players in the story.

Nuclear fuel and nuclear technology names such as Cameco, BWX Technologies, Uranium Energy, Energy Fuels and NuScale Power are further away from the immediate data center buildout, but they may become part of the broader narrative if the market concludes that AI requires a nuclear renaissance. Some of those names are far more speculative than others. That distinction should not be ignored.

Natural gas also matters. Whatever investors think about long-term decarbonization, the near-term reality is that gas-fired power can be dispatchable, scalable and relatively fast compared with many other forms of generation and transmission expansion. Williams, Kinder Morgan, EQT, Cheniere and Baker Hughes are not pure AI names. But if data center load growth drives more gas generation, pipeline capacity, turbine demand or LNG-related infrastructure, they become part of the second-order map.

Bloom Energy is another important case. Its onsite power systems have made it a favorite name in the data center power-constraints narrative. That does not make it low risk. It is still more speculative than the large equipment and utility names. But narratively, it sits very close to one of the most urgent questions in AI infrastructure: what happens when the grid cannot deliver enough power quickly enough?

Tier 3: data centers, cloud, networking and compute

The data center owners and digital infrastructure names are the most obvious part of the chain, but not always the simplest.

Equinix and Digital Realty are direct data center REITs. They own and operate critical digital infrastructure. Iron Mountain has also become increasingly relevant because of its growing data center business. These companies offer a more real-estate-linked way to think about the AI infrastructure boom.

The challenge is that data center REITs can face a mixed setup. Demand may be strong, but so can power costs, construction costs, financing costs and customer concentration. A data center is valuable only if it can secure power, interconnection, cooling and customers at attractive economics.

Networking is another critical layer. Arista Networks is one of the cleanest AI networking names because large AI clusters require extremely high-performance networking inside and between data centers. Broadcom also belongs here because of its custom silicon and networking exposure. Cisco, Ciena, Coherent and Lumentum provide broader or more specialized exposure to the traffic and optical-connectivity side of the AI buildout.

The compute layer remains essential. Nvidia, AMD, Broadcom, Micron, TSMC, Super Micro, Dell and HPE are all part of the physical AI stack. They are not energy plays in the narrow sense, but they are the reason the energy demand exists. Higher compute density increases the pressure on power and cooling systems. More AI servers mean more racks, more heat, more electricity and more grid strain.

Finally, the hyperscalers remain the demand engines. Microsoft, Alphabet, Amazon, Oracle and Meta are not only AI software companies. They are among the largest infrastructure spenders in the world. Their capital expenditure plans shape demand for chips, servers, data centers, power contracts, fiber, energy storage and grid interconnection.

In that sense, the OpenAI and Anthropic IPO narrative will not exist in isolation. It will be interpreted through the lens of the hyperscalers that fund, host, compete with or depend on similar AI infrastructure.

Why the OpenAI-Anthropic IPO process could pull attention into these names

The market likes clean stories. OpenAI and Anthropic could give public investors a clean story: the frontier-AI model layer is moving toward public-market scrutiny.

But once investors see the filings, they will likely focus on the messy details underneath the story. Revenue growth will matter. Enterprise adoption will matter. Competitive position will matter. But compute cost, cloud dependence, capital intensity and infrastructure constraints may matter just as much.

If the filings show enormous growth but also enormous infrastructure needs, investors may start looking for companies that benefit from that spending rather than companies that must absorb it. That is the difference between being the customer and being the supplier.

If an AI model company must spend aggressively to scale, the suppliers of power equipment, cooling systems, data center construction, grid upgrades, networking gear and firm electricity may become the more practical public-market proxies.

This is why the SpaceX comparison is useful. SpaceX did not make every space stock a good business. It did not eliminate execution risk. It did not make valuations irrelevant. But it created a thematic anchor. It helped the market think about the space economy as a real investable ecosystem. That attention spilled into listed proxies.

OpenAI and Anthropic could do the same for AI infrastructure.

The listed proxies will not all be equal. Some will have better margins. Some will have better backlogs. Some will have stronger balance sheets. Some will already be expensive. Some will be speculative and dangerous. Some will rise only because traders want exposure to the theme. But the direction of attention could be powerful.

The bullish scenario

The bullish scenario is straightforward.

AI demand continues to grow rapidly across enterprise software, consumer tools, coding, research, advertising, video generation, robotics, autonomous agents and industrial automation. OpenAI and Anthropic move closer to public listings, increasing institutional focus on the economics of frontier AI. The market begins to treat AI infrastructure as a decade-long investment cycle rather than a short-term capex burst.

Under this scenario, data center demand remains strong, hyperscaler spending stays elevated, power constraints become more visible, and companies with direct exposure to electrical infrastructure, cooling, grid construction and firm power see continued investor interest.

The biggest beneficiaries could be the companies closest to the bottleneck. That may include data center power and cooling suppliers, electrical equipment manufacturers, grid contractors, transmission builders, nuclear power owners, select utilities in key data center regions, and onsite power providers.

In this scenario, the AI trade broadens. It is no longer only Nvidia and the hyperscalers. It becomes Vertiv, Eaton, GE Vernova, Quanta, Constellation, Vistra, Dominion, Bloom Energy, Arista, Dell, Super Micro, Digital Realty and a much wider set of infrastructure names.

The bearish scenario

The bearish scenario is also real.

AI infrastructure could be overbuilt. Efficiency gains could reduce the expected electricity intensity of future models. Investors could discover that some AI revenue is less profitable than expected once compute costs are fully visible. Governments could impose stricter rules on data center power use, water use, emissions or grid-cost allocation. Local communities could block projects. Utilities could face political pressure if data center demand pushes retail electricity costs higher.

There is also a valuation risk. Many of the best-known AI infrastructure names have already moved significantly. A strong theme does not automatically mean an attractive entry point. If expectations become too aggressive, even good companies can disappoint.

Small and mid-cap thematic proxies are especially dangerous. Names tied to fuel cells, hydrogen, nuclear development, uranium, speculative power systems or small-cap infrastructure can move quickly when retail sentiment accelerates, but they can also reverse violently when financing, execution or profitability questions return.

That is why the watchlist should be treated as a map, not a recommendation. The AI infrastructure theme is real, but not every company connected to it will create shareholder value.

Key risks to monitor

Timing risk

OpenAI and Anthropic have filed confidential S-1 documents, but no final IPO dates have been confirmed. The process could take time. Companies can delay, change plans or wait for better market conditions. A thematic trade can begin before an IPO, but it can also fade if the calendar becomes too uncertain.

Regulatory pressure

AI companies are facing increasing scrutiny over safety, data, content, competition and social impact. Data centers are facing growing scrutiny over power consumption, land use, emissions, water demand and local community costs. The stronger the AI buildout becomes, the more political the infrastructure question becomes.

Grid availability

If interconnection queues remain slow, the market may have to distinguish between companies with real near-term demand and companies exposed only to long-term hopes. Power availability could become a gating factor for new AI capacity.

Cost inflation

Transformers, electrical equipment, labor, cooling systems, land, permits and construction timelines can all become constraints. Some suppliers may benefit from demand, but customers may resist pricing. Margins will matter.

AI economics

If AI monetization disappoints, if inference costs remain too high, or if enterprises slow adoption, the infrastructure buildout could be repriced. The strongest AI infrastructure thesis depends on sustained demand for compute.

Crowding

Once a theme becomes obvious, some stocks become crowded. The best narrative names may already price in years of growth. Investors should separate business quality from stock momentum.

What would confirm the thesis?

Several developments would support the idea that the AI infrastructure trade is becoming a major public-market theme.

First, more concrete OpenAI and Anthropic IPO disclosures. The actual S-1 documents, once public, could show how much infrastructure intensity sits behind the AI model layer. Investors will look for revenue growth, compute costs, cloud relationships, capital needs and risk disclosures.

Second, continued hyperscaler capex strength. If Microsoft, Alphabet, Amazon, Meta and Oracle continue to guide for heavy AI infrastructure spending, the entire supply chain remains supported.

Third, more power deals. Long-term electricity contracts, nuclear restart agreements, onsite power projects, gas-generation partnerships, renewables-plus-storage contracts and dedicated power arrangements would all reinforce the theme.

Fourth, stronger backlogs from electrical and construction suppliers. Companies such as Vertiv, Eaton, GE Vernova, Quanta, EMCOR, Powell and other infrastructure providers can validate the theme through orders, backlog, margin trends and commentary.

Fifth, faster regulatory action. If FERC, ERCOT and other regulators continue adapting interconnection and large-load rules, the market will have more evidence that data center demand has become a systemic power issue.

Sixth, local opposition. This sounds negative, but it can also confirm scarcity. If communities push back against data centers because of power, water, noise or cost concerns, it reinforces the idea that infrastructure capacity is becoming valuable.

Merlintrader bottom line

The OpenAI and Anthropic IPO process could become one of the most important public-market catalysts in the AI cycle. But the most interesting trade may not be limited to OpenAI or Anthropic themselves.

The deeper question is what the public market decides to reprice around them.

If the market treats these IPOs as proof that frontier AI is becoming a permanent infrastructure-scale industry, attention could move quickly into the listed companies that build, power, cool, connect and operate the physical AI stack.

That includes direct data center infrastructure names such as Vertiv and Eaton. It includes grid and electrification companies such as GE Vernova, Quanta and EMCOR. It includes firm-power and nuclear names such as Constellation and Vistra. It includes utilities in key data center regions. It includes power scarcity, gas, storage and onsite generation names. It includes data center REITs, networking suppliers, server companies and hyperscalers.

The market has already seen how a private leader can create public momentum across an entire listed ecosystem. SpaceX helped investors think about the space economy through public proxies. OpenAI and Anthropic may do the same for AI infrastructure.

The difference is scale.

AI is not just another software cycle. It is becoming a physical demand shock for electricity, compute, land, cooling, networks and power systems. The next phase of the AI trade may belong not only to the companies building the models, but also to the companies keeping the lights on.

Related Merlintrader reading