AI Stocks Bounced Back. What Actually Changed?

Apparently, a week was enough time for Wall Street to worry about the future of AI, reconsider, and decide it would like considerably more of it.

Monday’s rally was substantial. On September 21, the Nasdaq Composite climbed 2.26% to a record close, while the PHLX Semiconductor Index gained 4.3%, according to Reuters’ closing report. The enthusiasm followed the September 14 selloff we covered last week, when AI safety concerns and rising Treasury yields had investors questioning the same businesses.

I take the rebound seriously because consumer AI is giving investors something concrete to assess. Lower bond yields helped, too. What bothers me is how easily “the stock went up” becomes the entire argument for why the stock deserved to go up.

Muse gives the spending a purpose people can understand

Meta’s shares rose 11.4% on Monday, while AMD gained roughly 10% and reached a $1 trillion market value for the first time, Reuters reported.

Much of the excitement centers on Muse. Announced on September 8, the assistant is designed to handle tasks such as sending emails and booking travel. Access is free, with paid subscriptions for heavier use.

I can see the commercial appeal. Booking a trip involves enough tabs, comparison shopping and minor administrative misery that a reliable service could earn its monthly fee. That’s a more persuasive reason to subscribe than getting a slightly better answer to a question you asked out of boredom.

That only works if it’s reliable. If an assistant saves you twenty minutes and then requires twenty-five minutes of checking, you’ve acquired a small management position.

The early reception gives investors a reason to revisit their assumptions. My view is that paid consumer AI looks plausible, particularly for recurring tasks people dislike. I’d still want to see how many customers keep paying once the novelty wears off, and what it costs to serve them.

There’s a lot inside a revenue forecast

In Tuesday’s reporting on Muse, Reuters described Jefferies’ projection of $10.8 billion in annual revenue by 2027, based on assumptions including a billion users and 3% converting to paid plans.

Three percent of a billion means 30 million paying customers. If they supplied all $10.8 billion, each would spend an average of $360 a year, or $30 a month. That’s an implied average, not an announced plan price.

Would you keep a $30 subscription after the first few months? I’d want it handling a recurring chore well enough that canceling felt inconvenient. That’s the kind of habit I’d look for in customer data. Computing and support costs would then tell us how much of the subscription revenue could become profit.

Chipmakers have their own version of this homework. Their results depend on customer orders, delivery schedules, pricing and margins. A popular consumer app could support infrastructure demand. I’d look for the supplier’s own order and margin disclosures before changing its earnings assumptions.

The bond market made investors’ arithmetic friendlier

Oil prices fell on Monday, and the 10-year Treasury yield retreated below 5%, according to Reuters. That helped the backdrop for growth stocks.

When yields fall, investors may accept a lower required return on stocks, which raises the present value of expected future cash flows. Companies with much of their anticipated profit years away can be especially sensitive to that calculation.

This is why a rally can make financial sense before a company raises its earnings forecast. Investors might see less risk or better growth further out.

Consider a hypothetical stock trading at $100 with expected annual earnings of $5 a share. You’re paying 20 times those earnings. At $120, with the same $5 estimate, you’re paying 24 times.

In that example, the next year’s expected earnings haven’t changed. You’re paying extra for growth beyond that year, a lower assessment of risk, or some combination of the two.

Give the good news the same scrutiny

My objection is to selective skepticism. If we dismiss every decline as an emotional overreaction and celebrate every rally as proof of our insight, we’ve built an investment process with remarkably convenient results.

Try this with one AI-related holding: open its latest results and write down the revenue outlook, profit outlook and any disclosed customer commitments. Add the date and a link. Then compare those figures with what you were relying on before the rally.

If the company has supplied better evidence, explain what improved. If you’re paying a higher multiple for the same near-term outlook, write down what longer-term growth you expect to justify it. Include what you’d need to see in margins or cash flow as that growth arrives.

For the AI holding you own, what would the company have to deliver over the next year to justify the expectations you have today?