What if the biggest opportunity in the AI boom isn't building the smartest model, but owning the infrastructure that makes AI possible? ⚡ Behind every AI application are enormous data centers, specialized computing systems, reliable electricity, and connections capable of moving vast amounts of information. But there's a catch: announcing a project is far easier than getting it financed, permitted, powered, and operational.

Investor Kevin O’Leary estimates that 60% to 65% of announced data center projects may not be completed within 24 months, a forecast that highlights the obstacles facing this capital-intensive industry, not a verified industry-wide failure rate. For companies like CoreWeave (CRWV), Applied Digital (APLD), and IREN, the challenge is turning AI demand into dependable revenue without letting debt, construction costs, and delays undermine returns. Meanwhile, established cloud businesses such as Amazon, Alphabet, and Microsoft offer exposure to computing demand that extends beyond AI alone. The opportunity is substantial, but the winners may be determined less by ambitious announcements than by who can deliver usable capacity at an attractive cost.

🔎 Could a data center with secured power, permits, and customers be more valuable than a much larger project still on paper?

Explore O’Leary’s infrastructure thesis, the risks facing AI data center developers, and five critical milestones to examine before investing in the companies powering the AI boom.

Imagine setting aside $500 a month for Seagate Technology ($STX ( ▲ 3.41% )) for five years, buying a fixed dollar amount on a schedule, a method called dollar-cost averaging. In our example the buy happens on the first trading day on or after the 9th of each month, from Oct 2021 through Sep 2026 (the first buy lands on Monday, Oct 11, 2021, because Oct 9 was a Saturday): 60 buys, $30,000 in total.

STX closed at $80.53 on Oct 20, 2021 and at $639.01 on Oct 16, 2026, a price gain of about 693.5% (about 8 times), or roughly 51% a year compounded (the steady yearly rate that would give the same total gain). A single $10,000 invested at the Oct 20, 2021 close would have become about $79,350. The 60 monthly buys, $30,000 in total, would have been worth about $197,300 at the Oct 16, 2026 close, about 6.6 times the money put in. Price only: no dividends, fees or taxes, and these are examples, not forecasts.

The highest close of the past 52 weeks, and of the five years, was $1,094.04 on Jun 22, 2026 (the intraday high was $1,145.00 on Jun 18, 2026), so the Oct 16, 2026 close sat about 41.6% below it. The lowest close of the five years was $48.49 on Nov 3, 2022, about 39.8% below the $80.53 start. A drop of more than 40% from a peak only four months earlier shows how quickly a multi-year rally can retrace.

Note: a percentage shown beside a $ticker symbol in this issue is that ticker's live daily price move from Stocktwits when the page or email is opened. It is not one of our figures; every figure in this section is a dated close.

Caution: past pace rarely continues. Seagate's revenue and cash flow grew with demand for mass-capacity data center storage, yet the stock fell from its highest close of $1,094.04 on Jun 22, 2026 to $639.01 on Oct 16, 2026, a reminder that share prices can pull back sharply even while the business keeps growing.

The artificial intelligence boom is creating enormous investment opportunities, but the companies attracting the most attention may not necessarily be the ones best positioned to profit over the long term. Behind every AI model, chatbot, and automated business tool sits a growing network of data centers, electricity infrastructure, cooling systems, fiber-optic connections, and digital storage. Building that infrastructure requires billions of dollars, years of planning, and access to resources that are becoming increasingly difficult to secure.

That is where Kevin O’Leary, widely known as Mr. Wonderful from Shark Tank, sees an opportunity. Through his investments in data center projects and his involvement with O’Leary Digital, he argues that investors should look beyond the technology itself and examine the physical infrastructure that makes AI possible. His warning is striking: he estimates that 60% to 65% of announced data center projects may never be completed within the next 24 months.

For your portfolio, that distinction matters. The AI industry can continue expanding even as individual companies struggle, projects get canceled, and heavily indebted businesses face pressure. The opportunity is not simply to own anything associated with AI. It is to identify which businesses have the resources, permits, power, financing, and customers necessary to turn growing demand into sustainable returns.

When AI stocks become volatile, it is tempting to treat the entire sector as one investment. But O’Leary identifies two distinct areas that could benefit from the next several years of digital expansion: power infrastructure and computing capacity.

The first is electricity. AI data centers require substantial and reliable power, and the broader economy also needs additional electricity as businesses digitize more of their operations. Expanding AI use is adding to an existing challenge: the power grid must support growing demand while data center developers compete for suitable locations, permits, and energy supplies.

The second opportunity is computing capacity, including cloud services and digital storage. AI receives much of the attention, but the demand for data storage and conventional cloud computing does not depend entirely on AI adoption. Businesses, telecommunications networks, online commerce, and everyday digital services all require infrastructure to store, process, and move information.

That is why established cloud businesses such as Amazon Web Services (AWS), Alphabet’s Google, and Microsoft remain relevant to this investment theme. Their computing and storage operations serve many customers and workloads beyond AI. This broader demand may provide a different source of growth than investing exclusively in companies whose fortunes depend on building new AI infrastructure.

For an investor with limited time to analyze individual companies, the distinction is useful. Instead of asking only which AI company could become the next market leader, consider which businesses are supplying the essential resources that digital activity requires regardless of which AI model ultimately dominates.

However, a strong long-term industry trend does not guarantee that every company operating within it will deliver strong stock returns. The price paid for an investment, the amount of debt it carries, its ability to execute projects, and the competition it faces still matter.

The scale of the investment required to construct modern data centers helps explain why the sector can be both attractive and dangerous.

According to O’Leary’s interview, a project designed around approximately 1.4 gigawatts of capacity could require around $16 billion to become operational. He described a rough financing structure of $8 billion in equity and $8 billion in debt. These figures illustrate the magnitude of the projects being discussed; they should not be treated as a universal cost estimate for every data center, since costs vary with location, design, equipment, and the infrastructure required.

Developers must do much more than acquire land and construct a building. They need a viable route to construction permits, sufficient power, water access where required, fiber-optic connectivity, environmental approvals, and agreements that make the project financially attractive to future tenants.

Each requirement creates a potential point of failure. A site may look promising but lack the necessary electricity supply. Another may have power available but encounter local opposition, environmental restrictions, or a permitting delay. A developer may even spend substantial money on engineering and preliminary planning before learning that a critical approval will not be granted.

O’Leary estimates that preliminary planning costs can reach $20 million to $30 million per gigawatt of planned capacity, including permitting and engineering work. He also argues that developers should aim to secure a path to construction permits within approximately 18 months, while investors may be reluctant to finance projects that cannot demonstrate a credible permitting path within 24 months.

The financial consequences of delay are significant. Capital is committed before a facility generates revenue, and the longer a project takes to become operational, the longer investors must wait for returns. Delays can also affect the economics of the project and make financing more difficult.

The important question is not just how much a company plans to build. It is whether that company can realistically get its project operating and earning money.

O’Leary’s prediction that 60% to 65% of announced data center projects may not be completed is a central part of his investment thesis. It is his estimate, not an independently established industry-wide failure rate, and it should be evaluated accordingly.

His reasoning is that announcements are not the same as completed infrastructure. A project can attract attention and appear in company plans long before its developers secure the approvals, power, financing, and customer commitments needed to proceed.

Some projects may have fundamental technical or logistical problems. Others may struggle to obtain water rights, connect to fiber networks, secure adequate electricity, or access the natural gas and generation equipment required to produce their own power. In certain jurisdictions, assumptions about connecting to existing electrical infrastructure may no longer be sufficient, forcing developers to rethink how they will supply their facilities.

Political opposition adds another layer of uncertainty. Local governments and communities can influence whether a project proceeds, while proposed restrictions or moratoriums can delay construction or prevent it altogether. O’Leary believes political risk is one of the most important reasons many announced projects will fail to reach completion.

For investors, this creates a potential advantage for companies that have already overcome major obstacles. A developer with land, permits, power agreements, and credible construction plans may be in a stronger position than a competitor that has announced a much larger project but has yet to establish a viable path forward.

Scarcity could also support the value of sites that have cleared these hurdles. If many proposed projects never materialize, the supply of genuinely buildable data center capacity may be considerably smaller than the headline announcements suggest.

Still, there is an important distinction between a valuable project and an attractive stock. Even a well-positioned developer can be overvalued, overleveraged, or exposed to unfavorable contract terms. Scarcity alone does not guarantee a good investment.

The recent volatility discussed in the interview highlights how quickly sentiment can change for companies associated with data center construction and AI infrastructure.

CoreWeave $CRWV ( ▼ 0.44% ), a specialized cloud computing provider, was described in the interview as having fallen approximately 25% over a few days during the market pullback. Applied Digital $APLD ( ▲ 6.65% ) and IREN were also cited as companies that had dropped roughly 50% from their summer highs at the time of the discussion. These are figures reported in the interview, not current price changes or independently verified market data.

Those declines illustrate the concerns investors have about the cost of building infrastructure, the amount of debt required, and the time it takes for projects to generate sufficient cash flow.

A company can benefit from increasing demand for AI computing while still facing considerable financial risk. If its growth depends on large upfront investments, its financial results may be sensitive to construction schedules, equipment costs, financing conditions, and the timing of customer contracts.

That is especially important when a company commits capital before a facility is fully operational. Debt payments and other obligations can create pressure even if the long-term business opportunity remains intact.

The key is to avoid treating every price decline as either a bargain or a warning to sell. A lower share price may create an opportunity if a company's underlying economics remain sound, but it can also reflect genuine concerns about execution, financing, or future profitability.

When examining CRWV, APLD, and IREN, the more useful questions concern their individual financial positions and operating progress: How much capital must each company commit? How dependable are its customer agreements? Can it obtain the power and approvals it needs? How much financing will be required before the business can generate sustainable cash flow?

Those questions matter more than the assumption that all data center stocks should rise simply because AI demand is growing.

Electricity is not merely an operating expense in the data center business. Its availability, cost, and reliability can determine whether a project is feasible in the first place.

O’Leary described looking for locations where power costs are below roughly six cents per kilowatt-hour. He highlighted Saudi Arabia's exceptionally low electricity costs, as well as Finland, Norway, and Alberta, Canada, for their access to different energy resources. Saudi Arabia was cited at approximately one cent per kilowatt-hour, while Finland, Norway, and Alberta were described as having electricity costs below six cents in favorable circumstances.

These figures are examples from the interview, not guaranteed rates available to every developer. Actual energy costs depend on contracts, location, infrastructure, and other commercial terms. Access to cheap electricity also does not eliminate political, financing, or construction risks.

The underlying point remains important: a data center with dependable, affordable power may have a meaningful advantage over a competing project that cannot secure the same resources.

Norway was highlighted for its available hydroelectric power and its approach to choosing locations and working with communities. Finland was discussed in connection with nuclear power and more detailed scrutiny of how much electricity new projects consume. Alberta offers access to natural gas resources, while developers in the United States must contend with differing rules and political conditions across states.

O’Leary also discussed investments across several jurisdictions, including Norway and Finland through Bitzero, a publicly traded company he described as having land, power, and permits in those markets. His investments in projects in Utah and Alberta were described as private rather than publicly traded.

This geographic approach reflects a practical lesson: when the success of an investment depends heavily on local infrastructure and government decisions, spreading exposure across different jurisdictions can reduce dependence on any single market. It cannot eliminate risk, but it may help limit the damage if one location encounters delays or restrictions.

O'Leary's preference for the infrastructure behind AI follows a familiar investment idea: supplying the essential tools can be less speculative than betting on which individual technology will dominate.

A developer does not necessarily need to determine which AI model will become the market leader. It needs to secure the resources that customers require to operate computing facilities. That means land, permits, power, connectivity, and a structure that can accommodate the tenant's equipment.

This is also why he sees an advantage in companies that have already progressed beyond the announcement stage.

A project with existing operations has demonstrated at least some ability to build and run infrastructure. A company that has secured important permits and power contracts may have removed obstacles that still threaten competitors. Meanwhile, a project that remains private may offer development potential but can be much harder for investors to value and exit.

Bitzero was presented as an example of the public-market side of this approach. O’Leary pointed to its existing operations, public-market access to capital, locations, and power contracts as advantages compared with a private project that is only beginning development. He also indicated that he prefers having both publicly traded investments and private projects in his broader portfolio.

For you, the distinction comes down to visibility and liquidity. Publicly traded stocks can generally be bought or sold through the market, although their prices can fluctuate sharply. Private projects may offer different opportunities, but they can be difficult to sell, may require longer holding periods, and can expose investors to risks that are harder to assess.

A project's progress can change its prospects considerably, but that does not make every early-stage development a worthwhile investment. Before committing money, it is essential to understand what has actually been secured, what remains uncertain, and how much additional capital is needed.

O’Leary views data center development partly through the lens of commercial real estate. The comparison is useful because these facilities require land, substantial upfront investment, long-term tenant arrangements, and physical infrastructure that must remain useful over time.

In the interview, he compared data centers with specialized climate-controlled storage facilities used by pharmaceutical companies. Both businesses can require precise operating conditions and significant upfront development work. He described seeking returns of roughly 17% or more on certain developments and referenced historical capitalization rates in the 8% to 12% range for acquisitions that could later trade at lower rates of 4% to 5%.

These are his investment comparisons, not assured returns for data center projects. Actual performance depends on the price paid, operating costs, financing, lease terms, occupancy, and market conditions. A lower capitalization rate also does not automatically translate into a particular investor return.

He also said he generally prefers using no more than 30% leverage in the projects he discusses. That reflects his stated approach rather than a universal rule for the industry. Debt can increase returns when projects perform well, but it can magnify losses and create pressure when construction is delayed or revenue falls short.

The longer-term appeal is that successful facilities can generate cash flow from tenants while supporting employment and tax revenue in their surrounding communities. Land may also appreciate, although that outcome is not guaranteed, and specialized data center sites are not interchangeable with ordinary real estate.

Community relations therefore become part of the investment case. Residents may object to noise, water consumption, land use, or pressure on local infrastructure. Developers must show how a project will address these concerns and what benefits it can provide. O’Leary argued that locating facilities near enough to communities to provide employment, while maintaining a reasonable distance from homes, can help balance access to workers with local concerns.

The lesson is that real estate value does not come from owning land alone. In this sector, the land must be suitable, permitted, connected, powered, and acceptable to the surrounding community.

The data center opportunity becomes easier to assess when the investment thesis is broken into specific milestones rather than broad promises about AI growth.

1. Permits and environmental approvals

A company should be able to explain where its project stands in the permitting process, which approvals remain outstanding, and whether the timetable is realistic. Land ownership without a credible route to approval may offer far less value than it initially appears to.

2. Reliable power and connectivity

The project needs a clear plan for electricity, not just an assumption that power will eventually become available. Fiber-optic connectivity is also important because customers need fast and reliable connections to their computing infrastructure.

3. Construction funding and debt exposure

Large projects require significant capital before they produce revenue. Investors should examine whether a company has enough funding to progress through construction, what obligations it has already taken on, and how additional financing could affect existing shareholders.

4. Customer demand and contract visibility

Demand for computing capacity is encouraging, but a general increase in industry demand is not the same as a contract for a specific facility. The timing, terms, and reliability of customer commitments help determine whether the project can generate the cash flow needed to justify its cost.

5. A realistic route to profitability

A project should have a credible path from planning and construction to operation and cash generation. If permitting delays push revenue further into the future, the economics may deteriorate even if demand remains strong.

These milestones are especially relevant when evaluating companies such as CoreWeave, Applied Digital, and IREN. Their exposure to the same broad industry does not mean they have identical financial risks or project prospects. Each company requires its own assessment.

The concern about overbuilding is worth taking seriously. If too much capacity becomes available at once, data center operators could face greater competition, pressure on pricing, and lower returns on expensive infrastructure. Companies that borrow heavily to build facilities could be particularly vulnerable if revenue fails to meet expectations.

O’Leary disagrees with the view that announced projects will translate into widespread excess capacity. His argument rests partly on his estimate that many projects will never be built and partly on his assessment that demand for computing and power remains well above available supply. He suggested that demand for computing capacity could be around three times the capacity currently available, although this was his estimate rather than an independently verified industry statistic.

He also pointed to demand from large cloud providers, government agencies, and other businesses. In his view, these customers plan well ahead because they must coordinate facility availability with expensive computing equipment and their own capital spending.

The timing matters. A project that becomes operational when customers need capacity may be better positioned than one that finishes after competing facilities have already entered the market. Conversely, even a technically sound project can disappoint if construction delays or financing costs undermine its expected returns.

Investors should therefore avoid treating either extreme as certain. The industry may have strong long-term demand while individual companies still overbuild, pay too much for assets, or fail to deliver projects on schedule. Scarcity in one location does not guarantee profitability in another.

One of the more important points in the interview is that AI's economic benefits do not have to come exclusively from companies developing AI systems or building data centers.

O’Leary argued that AI tools are already helping businesses improve productivity and cash flow. He suggested that companies across different industries could benefit from adopting these tools, even if AI is not their primary product.

That creates a broader way to think about AI exposure. Microsoft, Alphabet, and Amazon are relevant through their cloud and digital infrastructure businesses, while companies in other sectors may benefit by using AI to improve their own operations. The potential public-market opportunity could also expand if Anthropic eventually goes public, although that possibility was discussed hypothetically in the interview and should not be treated as a confirmed listing.

O’Leary also mentioned quantum computing as a possible longer-term investment area, particularly in relation to security concerns surrounding increasingly capable computing systems. This is a separate, more speculative theme, and the interview did not identify a specific quantum computing stock to buy.

For a busy investor, the practical implication is to distinguish between companies that sell the infrastructure, companies that develop the technology, and companies that use the tools to improve their businesses. Each group has different revenue drivers and risks. Diversifying across business models may provide exposure to the broader adoption of AI without making the entire investment case depend on one narrow part of the industry.

The AI infrastructure boom offers a compelling long-term story, but compelling stories can still produce disappointing investments when the price is too high or the underlying business cannot deliver.

O’Leary's warning about data center cancellations is a useful reminder that announcements are not proof of future revenue. The companies that ultimately benefit may be those that secure power, complete permitting, finance construction responsibly, and establish dependable customer relationships. The firms that cannot overcome those obstacles may struggle even if demand for AI continues to grow.

That does not mean every established company is safe or every early-stage developer is a poor investment. It means the evidence supporting each investment should match the risk being taken. A publicly traded company with operational facilities may offer more visibility and liquidity, while a private project may offer potential upside with less flexibility to exit. Neither structure eliminates the possibility of loss.

For a portfolio built around long-term goals, it is worth deciding in advance how much exposure to this theme makes sense. O’Leary described data center development as a substantial part of his own investment activity, but his allocation reflects his circumstances, resources, and tolerance for risk. It should not be treated as a suitable allocation for every investor.

A measured approach can begin with businesses whose financial positions and operations are easier to understand, then evaluate more specialized infrastructure companies according to their balance sheets, project milestones, and valuations. Any position should be sized so that a sharp decline, a canceled project, or a prolonged delay does not derail the broader plan.

The central question is not whether AI will remain important. It is whether a particular company can convert that importance into durable earnings and attractive returns for shareholders.

The next phase of the AI boom may reward the companies that solve the least glamorous problems: securing electricity, obtaining permits, connecting facilities, controlling construction costs, and delivering capacity when customers need it. Those constraints may create opportunities, but they also separate a promising industry from a sound investment.

For your portfolio, the strongest AI thesis is not simply that demand will grow. It is that the company you own has a credible, financially sustainable way to meet that demand.

Want a simple next step for the $500 habit?

Our Fast Track Blueprint walks through a trusted online broker setup and a clear portfolio starting path.

Thanks for reading. This format is built to be fast to open, clear to understand, and useful enough to act on, without pretending past returns continue forever.

Caution: Past pace rarely continues. All figures here are approximate and are shown as examples, using price only (no dividends, fees, or taxes). Past performance is not a forecast; this is education, not advice.

Disclaimer: This newsletter is for informational purposes only and is not financial advice. Consult a qualified advisor before investing.