The AI data center boom is creating enormous demand, but a growing market does not mean every company supplying it will become a successful investment. Some businesses own facilities and collect long-term lease payments, while others purchase expensive GPUs and sell computing capacity directly. That difference can have a major impact on debt, operating costs, and future profitability. 📉

Eight companies illustrate the divide. Riot Platforms, TeraWulf, and Hut 8 have stronger landlord-style contract stories, while Applied Digital, IREN, and Nebius still face execution or valuation challenges. CoreWeave's heavy debt and Cipher Digital's financing uncertainty raise even bigger questions. For investors considering the AI sell-off, the opportunity may not be in buying every discounted stock, but in identifying which businesses can turn signed contracts and growing demand into sustainable cash flow.

🔍 Explore eight AI infrastructure stocks, the three companies with stronger investment cases, and the financial risks facing the others.

This article also examines AIQ, CHAT, SMH, and ROBT, plus four questions to ask before buying an AI-focused ETF or individual stock.

Imagine setting aside $500 a month for $LMB ( ▲ 6.31% ) 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 Nov 2021 through Oct 2026 (the first buy is on Tuesday, Nov 9, 2021): 60 buys, $30,000 in total.

LMB closed at $7.68 on Nov 9, 2021 and at $53.45 on Nov 6, 2026, a price gain of about 596.0%, or roughly 47.5% a year compounded (the steady yearly rate that would give the same total gain). In that example the $30,000 would have been worth about $79,800 at the Nov 6, 2026 close (about 2.66 times the money put in), and a single $10,000 invested on Nov 9, 2021 would have become about $69,600. Price only: no dividends, fees or taxes, and these are examples, not forecasts.

The highest close of the past 52 weeks was $114.11 on May 5, 2026, so the Nov 6, 2026 close sat about 53.2% below it. The lowest close of the five years was $5.09 on Jun 28, 2022, about 33.7% below the $7.68 start.

Caution: past pace rarely continues. LMB's Nov 3, 2026 release for the quarter ended Sep 30, 2026 showed record revenue of $230.5 million (up 24.9%) and record adjusted EBITDA of $27.4 million, yet the stock closed at $53.45 on Nov 6, 2026, still about 53.2% below its May 5, 2026 closing high of $114.11. Prices can pull back a long way even when the business keeps growing.

The AI boom is entering a more demanding phase. Building data centers and securing computing capacity are no longer enough to convince investors that every company in the industry deserves a premium valuation. As concerns about debt, construction costs, customer concentration, and profitability grow, several AI infrastructure stocks have suffered steep declines.

But a falling share price does not automatically mean a company is cheap. Some businesses own valuable data-center properties and lease them to major technology companies. Others operate the infrastructure themselves, requiring billions in financing to purchase equipment and expand capacity. Those differences can determine whether a company can turn AI demand into sustainable cash flow or become trapped under its own expansion costs.

For ETF investors, this distinction matters. The sell-off is not simply an opportunity to buy discounted AI exposure. It is a reminder to examine what actually supports the companies inside a technology or infrastructure ETF.

The same test applies to today's card stock: Limbach doesn't own data centers or buy GPUs. It services building systems for mission-critical facilities, so the question is how well it turns its $311.2 million in Q3 bookings into cash.

Before looking at individual stocks, it helps to understand the business models behind the sector.

Landlords own the land, buildings, and power infrastructure, then lease computing facilities to customers. Their tenants generally take responsibility for operating the computing equipment. Long-term contracts can provide more predictable revenue, although construction financing and tenant credit quality remain important risks.

Operators own or lease facilities but also purchase GPUs and sell computing capacity directly. This creates more direct exposure to AI demand, but it also requires substantial capital investment. Equipment depreciation, financing costs, and technological changes can put pressure on profitability.

Companies with mixed or developing models may combine leasing, equipment ownership, and other infrastructure services. Their investment potential depends on how successfully they convert contracts and capacity into actual revenue and cash flow.

The distinction is more useful than treating every company as part of the same AI trade.

Riot Platforms attracted attention after announcing a 15-year lease covering its entire Corsicana campus in Texas. The agreement involves approximately 910 megawatts of capacity and has an estimated value of $28 billion over its initial term.

A long-term lease can provide meaningful revenue visibility, particularly when the tenant is a financially strong AI company. However, Riot did not identify the tenant, leaving investors with an important unanswered question: who is ultimately responsible for paying the rent?

There is also a timing issue. Most of the contracted AI revenue is not expected to appear until around mid-2028. In the meantime, Bitcoin mining remains central to the business, accounting for approximately 64% of quarterly revenue in the period discussed.

The estimated cost of completing the campus is another consideration, with buildout expenses projected at $7.6 billion to $7.8 billion.

Riot's opportunity is the transition from Bitcoin mining toward long-term AI infrastructure leasing. The risk is that investors may have to wait years for the new business to contribute meaningfully to financial results.

TeraWulf offers a different kind of landlord investment. The company signed a 20-year lease with Anthropic covering 401 megawatts at its Kentucky data-center campus. The contract is expected to generate approximately $19 billion in revenue over its term.

An important part of the agreement is Google's credit support for project financing. That backing helps address one of the biggest questions facing data-center landlords: whether the tenant has the financial strength to meet its long-term obligations.

The company's latest reported quarter also showed progress. Revenue increased 51% year over year, while AI and high-performance computing lease revenue accounted for approximately 83% of the total. That portion of revenue had been only about $7 million a year earlier.

The additional capacity is expected to come online in early 2028, so execution remains important. Still, TeraWulf has a combination of long-term contracted demand and customer credit support that makes its business model worth watching.

Hut 8 has also secured substantial long-term commitments. Its Beacon Point campus in Texas has two 15-year leases, with total contract value estimated at approximately $19.6 billion. One of the agreements announced in July was valued at about $9.8 billion and involved an investment-grade tenant.

Having an entire gigawatt-scale campus leased reduces uncertainty about finding customers for that capacity. It also gives the company a clearer path toward generating revenue from its infrastructure.

However, Hut 8 carries concentration risk. A large portion of its business is tied to a single campus, meaning operational problems or unexpected changes affecting that location could have an outsized impact.

The contracts strengthen the investment case, but investors should still monitor construction, financing, and the timing of revenue recognition.

Applied Digital has leases covering approximately 1.4 gigawatts across five campuses in North Dakota, Louisiana, and Alabama. The agreements have an estimated total value of $36 billion over their initial terms.

On the surface, this looks compelling. The company has multiple locations and substantial contracted capacity, reducing its dependence on a single campus.

The concern is the financial strength of its tenants. Its flagship Polaris Forge campus is leased to CoreWeave, a company carrying significant debt. Although the lease structure was reportedly reworked to involve an investment-grade special-purpose entity, investors still need to understand who ultimately supports the payment obligations.

Applied Digital's diversified campus footprint is a positive, but a large contract is only as dependable as the financing and customer commitments behind it.

IREN began as a Bitcoin miner and is transitioning toward AI cloud infrastructure. Unlike a traditional landlord, it also purchases GPUs and sells computing capacity, giving it more direct exposure to AI demand and more responsibility for equipment costs.

The company has a five-year, $9.7 billion agreement with Microsoft and a separate five-year, $3.4 billion agreement with Nvidia. Its annualized contracted revenue was approximately $4 billion as of August, with much of its capacity reportedly sold out.

Its latest reported results showed total revenue increasing 66% quarter over quarter, while AI cloud revenue nearly tripled to approximately $200 million. Its backlog reached $20.3 billion, with roughly two-thirds associated with investment-grade customers.

However, a Texas campus experienced delays after Microsoft requested network upgrades to support newer AI models. Completion was targeted for the end of January 2027, with the possibility of further delay into February.

IREN's customer relationships and contracted demand are encouraging. The key question is how efficiently it can bring computing capacity online without allowing delays and equipment costs to undermine its economics.

Nebius has taken a different approach to financing. Rather than relying as heavily on debt, it has used customer prepayments and stock sales to fund expansion.

The company raised approximately $2.8 billion through share sales and carried around $10 billion in debt, considerably below CoreWeave's reported $52 billion.

That does not make the stock inexpensive. Nebius was valued at approximately $39 billion against trailing revenue of roughly $1.4 billion, or around 28 times sales.

Investors are paying a substantial premium for its growth potential. Share issuance also creates dilution, meaning existing shareholders own a smaller percentage of the business as more shares are issued.

Its ability to grow revenue quickly enough to justify that valuation is the central issue. GPU depreciation and the cost of maintaining competitive computing capacity add further uncertainty.

CoreWeave is an AI infrastructure operator, not simply a landlord collecting rent. It purchases computing equipment and sells access to that capacity, creating substantial capital requirements.

As of October, the company carried approximately $52 billion in total debt. Its unsecured bonds had fallen below 80 cents on the dollar, pushing yields above 15%, a sign that credit investors were demanding much higher compensation for risk.

The bull case is still significant. Revenue more than doubled in its latest reported quarter, and its contracted backlog reached approximately $104 billion.

However, Meta and OpenAI account for around $40 billion of commitments. Although both are major AI customers, concentration creates risk if either company changes its spending plans.

The bigger concern is financing. If credit conditions tighten, CoreWeave could face higher borrowing costs just as it needs more capital to expand. Strong demand does not automatically translate into attractive returns when debt obligations are this large.

Cipher Digital has secured long-term contracts, including a 15-year agreement with Amazon Web Services. Having a major customer is a positive, but the company still needs to demonstrate that it can finance the infrastructure required to fulfill its commitments.

The concern is that the funding for its expansion has not been clearly established. Its November 2 earnings announcement did not provide the additional lease or financing developments that investors were hoping to see.

This illustrates an important distinction: a signed contract creates potential future revenue, but it does not eliminate the cost of building the facility.

Until Cipher demonstrates a credible financing path, its long-term lease should not be treated as proof that the business model is secure.

The eight companies illustrate why broad exposure to AI does not guarantee equal investment quality.

Some funds provide exposure to AI infrastructure through semiconductor manufacturers, data-center operators, cloud companies, and technology platforms. Others hold more specialized infrastructure businesses. Their risks can be very different.

For investors evaluating ETFs, several funds offer useful starting points:

These funds offer different approaches, and none should be assumed to hold all eight companies discussed above. Their current holdings, portfolio weights, expense ratios, and concentration should be checked before investing.

The important distinction is between investing in the infrastructure itself and investing in the broader ecosystem that benefits from AI adoption. Semiconductor companies, for example, may benefit from continued demand for computing hardware without carrying the same project-financing risks as heavily leveraged data-center operators.

That does not make semiconductor ETFs risk-free. Chip demand is cyclical, valuations can become stretched, and competition can affect margins. But the underlying business models are different, which is exactly why ETF selection should begin with understanding what the fund owns.

A steep decline can make an AI stock look attractive, but the price chart alone does not tell you whether the business is improving.

Before investing in an AI infrastructure company or an ETF with exposure to the sector, consider four questions:

These questions help separate a temporary market sell-off from a genuine deterioration in business fundamentals.

The long-term demand for AI computing remains a major investment theme. However, the companies supplying that capacity must still prove they can finance expansion, secure reliable customers, manage equipment costs, and generate sustainable cash flow.

Among the eight companies, Riot Platforms, TeraWulf, and Hut 8 have stronger landlord-style contract stories in the analysis. Applied Digital, IREN, and Nebius have opportunities but also meaningful execution or valuation concerns. CoreWeave and Cipher Digital face more serious financing questions.

For investors who prefer ETFs, the lesson is not to avoid AI exposure altogether. It is to avoid treating every AI-related business as interchangeable. A diversified fund can reduce the risk of relying on one company, but it cannot eliminate sector-wide valuation, technology, or market risks.

Tip: Don't buy an AI ETF simply because its holdings have fallen; examine its underlying businesses, debt exposure, customer concentration, and valuation to determine whether the sell-off has created an opportunity or exposed a deeper weakness.

Ready to Revolutionize Your Wealth?

Here's what's waiting for you:

As the world becomes increasingly digital, this question will be on the minds of millions seeking new income streams in 2026.

The answer is—Absolutely!

Thank you for taking the time to read today’s email! Your support is what allows me to send out this newsletter for free every day. 

 What do you think of the new format? Please provide your feedback in the poll below, and if you find the newsletter valuable, feel free to share it with other investors!

Disclaimer: This newsletter is for informational purposes only and should not be considered financial advice. Please consult with a financial advisor before making any investment decisions.