How to Invest in AI Companies Without Picking the Wrong Ones

The AI investment boom has created more wealth—and more wealth destruction—than almost any technology sector in recent memory. Every quarter, another company rebrands itself as an “AI company” and watches its stock price soar, only to come crashing back to earth when investors realize there’s no actual revenue underneath the hype. The difference between making money in this space and becoming another cautionary tale comes down to knowing what to look for beyond the press releases.

I’ve spent years analyzing technology investments, and I can tell you that the biggest mistakes I see aren’t from people who picked the wrong AI company—they’re from people who never asked the right questions in the first place. The good news is that evaluating AI companies isn’t actually that complicated once you know where to look. The bad news is that most investors skip those steps because they’re seduced by the story rather than scrutinizing the fundamentals.

Let me walk through what actually matters.

Distinguish Between AI Infrastructure and Application Companies

The most important distinction in AI investing is whether a company builds the tools that enable AI or actually uses AI to deliver value. This changes how you evaluate risk, growth potential, and competitive positioning.

Nvidia is an infrastructure company. They design and sell the GPUs that training large language models requires. Their revenue has exploded because every company building AI needs their chips. When you invest in Nvidia, you’re betting on the entire AI ecosystem continuing to grow—you don’t need any single AI company to succeed, as long as the industry as a whole expands.

Application companies are different. Microsoft has integrated AI capabilities across Office 365 and Azure. They can leverage AI to increase their existing revenue streams without needing to convince customers to buy something entirely new. Salesforce has added AI features to their CRM platform—they’re monetizing AI by making their existing product more valuable to customers who already pay them.

The insight here is that infrastructure companies tend to have more predictable revenue in a growing market, while application companies offer higher upside if they successfully transform their business but face execution risk that infrastructure players avoid. Neither is inherently better, but mixing them up in your analysis leads to flawed conclusions.

Before you buy, ask yourself: does this company need AI to succeed, or is AI just a marketing layer on top of an existing business?

Scrutinize the Actual Revenue, Not Just the AI Narrative

This is where most retail investors get burned. A company announces an “AI partnership” or launches an “AI feature,” and suddenly it’s treated like a growth stock regardless of whether any money actually follows.

Look at the revenue numbers. Specifically, look at how much revenue actually comes from AI-related products versus how much is traditional business dressed up in new language. Some companies generate meaningful AI revenue today—Microsoft’s Azure AI services, for instance, contribute billions to their cloud business. Others are simply tacking AI onto products that customers already buy for other reasons.

The financial statements will tell you this if you know where to look. Read the earnings call transcripts and pay attention to how executives describe AI revenue. Are they giving specific numbers, or are they using phrases like “AI is integral to our strategy”? The vague language usually signals that AI revenue is either tiny or doesn’t exist in any meaningful way.

I remember when BuzzFeed saw a temporary stock price surge after announcing AI content generation plans in early 2023. The revenue never materialized in any detectable way, and the stock gave back those gains within months. The lesson is clear: beautiful narratives about AI don’t pay dividends. Actual revenue does.

When evaluating any AI investment, demand to see the money. If the company can’t show you concrete AI revenue today, treat any valuation premium as speculative rather than fundamental.

Evaluate the Competitive Moat

In traditional industries, a company might sustain competitive advantage through brand recognition, distribution networks, or regulatory capture. In AI, the moat question is more immediate: can this company survive when better-funded competitors enter their space?

Data advantages can create real moats. Companies with proprietary datasets that are difficult or expensive to replicate—like medical imaging databases in healthcare AI or customer behavior data in enterprise software—have defensible positions. But data advantages erode over time as competitors find alternative data sources or as public datasets improve.

Network effects matter enormously in AI. Consider how Microsoft benefits from the feedback loop between Azure AI services and their enterprise customer base. Every company that uses their AI tools generates data that improves Microsoft’s models, which attracts more customers. This virtuous cycle is incredibly difficult to break once established.

Technical talent is often cited as a moat, but I think it’s overrated as a sustainable advantage. Top AI researchers can be hired away, and the best ones frequently leave to start their own companies. What you can’t hire is the accumulated institutional knowledge, proprietary techniques, and years of training data that go into a mature AI system.

Ask yourself: what would happen if Google, Microsoft, or Meta decided to compete directly in this company’s niche? If the answer is “they could replicate this relatively easily,” the moat is too shallow to justify investment.

Analyze Management Incentives

I’ve seen brilliant technical teams build amazing AI products and watched their stock get demolished because management was selling shares into every rally. The AI talent wars have made this problem worse—when companies are paying executives and engineers partly in stock, there’s sometimes more incentive to maintain a high stock price than to build a durable business.

Look at the insider buying and selling patterns. Are executives accumulating shares or dumping them? Pay particular attention to sales that happen after lock-up periods expire or around major product announcements. If the people running the company don’t think the stock is worth holding, why should you?

Check the compensation structures. Some companies have shifted toward cash-heavy compensation in the current environment, which signals confidence in maintaining talent without needing stock incentives. Others are still heavily equity-dependent, which can create problematic incentives around short-term stock performance.

The most telling sign is often what executives say versus what they do. If the CEO is constantly talking about how the stock is undervalued while simultaneously selling shares, that disconnect tells you something important about what they actually believe.

Understand the Regulatory Environment

AI regulation has moved from abstract policy discussion to concrete legal reality faster than most investors anticipated. The EU’s AI Act began implementation in phases starting in 2024, creating strict requirements for certain AI applications and outright banning others. China has implemented its own AI regulations. The US has issued executive orders and will likely see legislative activity in the next few years.

For investors, this regulatory environment creates both risks and opportunities. Companies that rely on collecting and processing large amounts of personal data face significant compliance costs and potential liability. Facial recognition companies, for instance, have seen their business models disrupted by privacy regulations that were unimaginable five years ago.

On the other hand, companies that build AI systems designed for compliance from the ground up—or that operate in sectors with strong regulatory moats—can benefit from rules that their competitors struggle to meet. Enterprise AI companies selling to regulated industries often welcome regulation because it creates barriers to entry that startups can’t easily clear.

The smart move is to think about regulatory tailwinds and headwinds before you invest, not after. Ask whether the company’s core technology could be significantly impacted by upcoming rules. If you’re uncertain about the regulatory outlook, that’s often a reason to pass rather than guess.

Resist the Valuation Temptation

AI companies trade at extraordinary valuations by historical standards, and that reality isn’t necessarily a reason to avoid them—but it is a reason to be more discriminating. The difference between a company that’s expensive and one that’s overpriced comes down to growth expectations and the probability of achieving them.

A company trading at 100x earnings is only overpriced if it fails to grow into that valuation. If that same company is growing revenue at 50% annually and has a clear path to profitability, the multiple might actually be reasonable. The key is matching the valuation to realistic growth scenarios rather than assuming that any premium is automatically unjustified.

What concerns me more than absolute valuation is when companies have priced in perfection. Look at what Wall Street expects versus what the company has actually delivered. If expectations have been ratcheting up every quarter and the stock only goes up when they beat the already-elevated numbers, you’re looking at a setup where even good news might not be enough.

One useful exercise: calculate what the stock price would be if growth slowed to half the current rate. If that number still seems reasonable relative to the current price, the valuation might be justified. If it would represent a 50% decline, you’re paying for a perfect outcome.

Look at Customer Concentration and Retention

This is one of the most boring aspects of AI investing, which is exactly why it matters so much. The flashiest AI company with the best technology can still be a terrible investment if they can’t keep customers or if a handful of clients account for most of their revenue.

Customer concentration risk is particularly acute in the enterprise AI space. Some companies have contracts with a handful of large customers that represent the majority of revenue. When one of those customers shifts priorities—or builds their own internal AI capability—revenue can collapse overnight. I’ve seen this happen repeatedly, most recently with companies that were heavily dependent on a small number of large tech customers for their AI services.

Retention metrics tell a different story. Software companies that succeed typically show net revenue retention above 100%, meaning existing customers are spending more over time. This is even more important in AI, where the best companies create products that become more valuable as customers use them more. If a company can’t retain and expand existing customers, whatever new customer acquisition they’re achieving is just filling a leaky bucket.

Ask for these numbers directly if they’re not in the financials. A company that’s confident in its customer relationships will volunteer this information. One that deflects should make you suspicious.

Accept That You Probably Can’t Pick the Winners

Here’s the uncomfortable truth about AI investing: predicting which specific company will dominate a decade from now is essentially impossible, even for professional investors with resources far beyond what retail investors can muster. The technology is evolving too quickly, and the competitive dynamics shift with every major model release.

What you can do is avoid the obvious losers. Companies without clear competitive advantages, without differentiated technology, without solid unit economics, and without management teams aligned with shareholders—these are the ones that will destroy capital regardless of how hot the AI sector appears.

The other thing you can do is resist the FOMO that drives people to buy at the top. The AI rally of 2023 saw many companies triple or quadruple in value based on announcements that hadn’t produced any revenue. Investors who bought at those levels are still waiting to break even. Waiting for a pullback isn’t always the right strategy, but it’s almost always better than buying at mania prices.

Consider whether you’d be comfortable holding the stock if the AI hype disappeared tomorrow. If the only reason you want to own it is because everyone else is talking about AI, that’s a signal to reconsider.

The Bottom Line

Investing in AI companies without picking the wrong ones comes down to treating them like businesses rather than magic. The technology is transformative, but the investment principles haven’t changed: you need to understand what you’re buying, what it’s worth, and whether the people running the company share your interests.

The AI sector will continue producing both incredible opportunities and spectacular blowups. Your goal isn’t to avoid all blowups—that’s impossible in a space growing this quickly. Your goal is to avoid the ones that are obviously avoidable, the ones where the warning signs were visible to anyone willing to look. That’s not about being smarter than the market. It’s about being more disciplined.

The market will keep generating AI stories that sound compelling. Your job is to look past the story to the underlying business. That’s where actual returns are made.