The 3 AI Fears Driving Markets in 2026: DeepSeek, CapEx Returns, and the Legacy Software Threat

TL;DR The AI anxiety gripping markets in early 2026 is not one fear but three: DeepSeek challenging U.S. AI dominance at a fraction of the cost, hundreds of billions in CapEx with uncertain ROI timelines, and growing panic that AI will destroy legacy software businesses. Each is a legitimate concern—but the market has a history of overshooting on the downside just as it overshoots on the upside.

Separating the Three Fears

Most investors treat "AI anxiety" as one vague feeling. That is a mistake.

There are three distinct fears driving the market sell-off in AI-related stocks, and they deserve separate analysis because they have very different implications.

Fear 1: DeepSeek and the Crack in American AI Dominance

A Chinese AI lab produced a model that appeared to match the best American AI models at a fraction of the cost. This rattled the entire investment thesis behind the AI buildout.

The assumption had been clear: Nvidia, Microsoft, Google, and OpenAI had an almost insurmountable lead. Massive infrastructure, hundreds of billions in capital, years of head start.

DeepSeek suggested otherwise:

• You may not need hundreds of thousands of Nvidia GPUs to build a world-class model
• The moat around American AI companies may not be as wide as assumed
• More efficient models could slow hardware demand sooner than expected

This is a legitimate concern. Competition is real. China is serious about AI. If Nvidia's stock price assumes decades of continued dominance, any crack in that assumption has significant valuation implications.

Think of it this way. Driving 5 miles with 5 minutes of buffer? Fine. Driving cross-country with 5 minutes of buffer? Unrealistic. The further out you project dominance, the more things can go wrong. This is why the 30-year Treasury pays far more than the 90-day bill.

Fear 2: The CapEx Question—When Does the Money Come Back?

Microsoft, Amazon, and Alphabet have collectively committed hundreds of billions to AI infrastructure over the next few years.

Initially, the market cheered. AI is the future. Spend the money. Build it.

Now investors are asking harder questions:

• When does this investment pay off?
• What does the payoff actually look like?
• What if it takes longer than stock prices assume?

Capital expenditures come directly from cash flow. When Microsoft commits $80 billion to AI infrastructure, that is $80 billion not returning to shareholders. It needs to generate strong returns to justify the spending.

Current AI revenue is real but still relatively modest compared to the magnitude of investment. The question is not whether AI will generate value—it almost certainly will. The question is whether the payoff timeline matches what current stock prices assume.

This is not AI skepticism. It is healthy scrutiny of the math.

Fear 3: AI Eats Legacy Software Alive

Adobe, Salesforce, Intuit, ServiceNow—companies long viewed as having some of the most stable recurring revenue streams in technology—are suddenly being treated as if they are in the early stages of extinction.

The Centrini report amplified these fears, extending them from software into financials, real estate, and insurance.

But what does the data actually show?

AI has been building for over three years. During that time, most of these software companies have been actively integrating AI into their products. Adobe, for instance, has actually seen demand increase since AI took center stage.

Will AI significantly disrupt these businesses? Probably. But likely not in the way most people assume. Could it permanently eradicate them? Possible but not probable.

Here is the framework I find useful: at a sufficiently low price, you can buy a basket of 4-6 major companies that the market says AI will destroy. For all of them to be bad investments, every negative scenario would need to play out almost exactly as feared. That is about as unlikely as every positive scenario playing out exactly as hoped.

Heads I win, tails I don't lose much. When prices reflect maximum pessimism, the risk-reward math shifts dramatically in the investor's favor.

FAQ

Q: Should I sell Nvidia because of DeepSeek? A: DeepSeek represents real competition, but it does not invalidate Nvidia's business. The question is how much future growth the stock price already assumes. If decades of dominance are priced in, even small cracks in the thesis can cause large price declines. Judge by comparing price to intrinsic value, not by headline fear.

Q: What happens if AI CapEx does not generate returns? A: Microsoft, Amazon, and Google would face valuation compression. But these companies have enormous cash-generating businesses beyond AI. Delayed returns are more likely than total failure—and stock price corrections could create buying opportunities.

Q: Are legacy SaaS companies like Adobe really at risk from AI? A: Short-term, the panic is likely overblown. These companies are integrating AI into their products and have high switching costs. Long-term risk exists, but at sufficiently low prices, the risk-reward tilts favorable. The key is not to overpay for the perceived safety of these businesses during euphoria or to sell them in a panic during fear.