In today’s issue, Robert Rapier examines which AI companies are earning their valuations — and which are burning capital without proof. But there’s another way to position for AI that sidesteps the “show me” test entirely: own the essential-service companies getting paid to build the boom, regardless of which application wins. Robert has spent decades in exactly these stocks →

For much of the past two years, investors were willing to give companies the benefit of the doubt on artificial intelligence. Announcing a new AI product, data center, partnership, or spending plan was often enough to generate enthusiasm. The assumption was that the profits would eventually follow.

That period is ending.

Investors are not turning against AI. Recent earnings showed that demand for cloud computing and AI services remains exceptionally strong. But the market is becoming much more selective about which companies deserve credit for that demand and which ones are merely spending enormous sums in hopes of capturing it.

The question is shifting from “How much are you investing in AI?” to “What are investors getting in return?”

The Market Is Drawing Distinctions

The contrast was clear in the latest round of technology earnings.

Microsoft (NSDQ: MSFT) reported quarterly revenue of roughly $90 billion, while Azure cloud revenue grew 43%. Microsoft Cloud generated $59.3 billion, and Microsoft 365 Copilot surpassed 30 million paid users. Those figures gave investors tangible evidence that AI investment was contributing to revenue growth rather than simply increasing expenses.

Amazon (NSDQ: AMZN) told a similar story. Amazon Web Services revenue increased 37% to $42.2 billion, its fastest growth in more than four years. AWS’s contract backlog reached $496 billion, providing substantial visibility into future demand. Investors rewarded those results even though Amazon raised its annual capital-spending plan to approximately $220 billion and reported negative free cash flow over the latest period.

In both cases, the spending was enormous. But so was the evidence that customers were willing to pay for the resulting capacity.

Other companies received a less forgiving response.

Alphabet (NSDQ: GOOGL) reported strong revenue and cloud growth, but investors focused on the cost of delivering that growth. The company raised its expected 2026 capital spending to between $195 billion and $205 billion, while heavy investment pushed quarterly free cash flow below zero. The shares initially declined despite results that exceeded revenue expectations.

Meta Platforms (NSDQ: META) also reported solid revenue growth, but earnings fell short of expectations as the company continued to increase spending on AI infrastructure. Meta raised the lower end of its annual capital-spending range, and the stock sold off following the report.

The market’s reaction was not a vote for or against artificial intelligence. It was a demand for evidence.

Test No. 1: Where Is the Revenue?

The first test for any AI investment is straightforward: What product or service is generating revenue?

That question is easier to answer for some companies than for others. Cloud providers can measure AI-related computing consumption. Software companies can disclose paid subscriptions or higher revenue per customer. Advertising platforms can demonstrate whether AI tools are improving targeting, engagement, or conversion rates.

Investors should be cautious when a company talks extensively about AI but provides little information about customers, pricing, adoption, or revenue.

Management does not necessarily need to disclose a separate AI revenue line. Many AI features are being integrated into existing products. But investors should be able to identify some measurable economic benefit, such as faster cloud growth, rising subscription revenue, higher margins, improved customer retention, or greater employee productivity.

Otherwise, “AI strategy” may simply be a label attached to ordinary technology spending.

Test No. 2: Can the Company Afford the Buildout?

The second test is financial capacity.

AI infrastructure requires chips, servers, networking equipment, data centers, electricity, cooling systems, and skilled employees. The upfront capital requirements are enormous, but the costs do not end when construction is completed.

Depreciation rises as new equipment enters service. Power and maintenance expenses continue. Chips may need to be replaced long before a data-center building reaches the end of its useful life. Microsoft has said that a large share of its recent capital spending went toward relatively short-lived assets such as graphics processors and central processing units.

That makes free cash flow especially important.

A company can report strong earnings while capital spending consumes much of the cash generated by the business. That does not automatically make the investment unwise. Amazon spent heavily for years building a logistics network that eventually became a major competitive advantage. But investors need to consider how long the spending will continue, when the assets may begin producing adequate returns, and whether the balance sheet can support the process.

The strongest AI companies can fund investment from existing operations. Weaker companies may need to issue debt or stock, accept years of negative cash flow, or cut spending elsewhere.

Test No. 3: What Protects the Profit?

The third test is competitive advantage.

Demand for AI may grow rapidly without every participant earning attractive returns. The automobile transformed the economy, but most early automakers eventually disappeared or were absorbed by stronger competitors. The internet reshaped commerce and communication, but countless internet companies vanished along the way.

The same distinction will apply to AI.

A durable advantage could come from proprietary chips, scarce power access, data-center capacity, valuable data, customer relationships, software distribution, or the ability to spread development costs across a large installed base. Microsoft, Amazon, Alphabet, and Meta already have billions of users or deeply established enterprise relationships. That gives them ways to distribute AI products that a startup may not have.

Infrastructure providers may also benefit without needing to predict which individual AI application becomes dominant. Semiconductor manufacturers, electrical-equipment suppliers, utilities, data-center operators, and networking companies can serve as toll collectors on the broader buildout.

But even here, investors must remain selective. A business is not automatically attractive merely because it sells something used in a data center. Valuation, customer concentration, debt, contract terms, and competitive pressure still determine whether revenue growth translates into shareholder returns.

The Big Picture

The AI boom is not ending, but it is maturing.

The first stage rewarded bold promises and ambitious spending plans. The next stage will reward companies that can show a clear connection between investment, customer demand, revenue, and cash flow.

That is a healthy development. Transformative technologies produce both extraordinary winners and expensive disappointments. The market is beginning to separate the two.

Investors evaluating an AI-related stock should ask three questions:

Where is the revenue? Can the company finance the investment? What protects the eventual profit?

Companies with convincing answers may continue to thrive even as the market becomes more demanding. Those relying mainly on excitement may discover that the words “artificial intelligence” no longer guarantee an enthusiastic response.

The three questions I’ve laid out here — where is the revenue, can the company finance the buildout, what protects the profit — are the right tests. But answering them requires staying close to earnings every quarter and making ongoing calls on companies still proving themselves.

The portfolios I manage in Utility Forecaster take a different approach. Rather than evaluate which AI application earns its keep, I focus on the essential-service companies — utilities, pipelines, grid operators — that get paid to build the infrastructure every AI winner runs on. You don’t have to predict which company passes the “show me” test. You just need to own the companies they all have to pay.

Last year my Income Portfolio returned 10.7% with a 4.8% yield, moving less than half as much as the market. My Growth Portfolio returned 16.5%. See the companies I believe are best positioned for this buildout →