Yes, the AI stock selloff is a warning sign — not because artificial intelligence is fake or useless, but because Wall Street priced a handful of companies for a future that has to arrive exactly on schedule, at exactly the scale promised, with no room for a bad quarter. That’s not a market pricing in a technology. That’s a market pricing in a miracle. And miracles, historically, are not what quarterly earnings calls deliver.
Note: this is a tech-industry opinion piece, not financial advice. Nothing here is a recommendation to buy or sell any security — talk to a licensed advisor about your own portfolio.
The Numbers That Should Worry You
Start with the plain facts, no spin needed. The Nasdaq Composite slid nearly 5% in a single week in July, and the Philadelphia Semiconductor Index — the closest thing Wall Street has to an “AI barometer” — fell 10% in that same stretch, down 20% from its record high and technically in bear-market territory. Alphabet, Microsoft, Amazon, and Meta together are on pace for roughly $725 billion in AI capital expenditures this year, a figure Wall Street itself expects to climb toward $900 billion in 2027. Hyperscaler capex overall jumped 78% in a single year, from $416 billion to $739 billion. Those aren’t fringe numbers from a bear blog — they’re the consensus estimates the same banks issuing 8,000-point year-end S&P targets are working from.
This Is Concentration Risk, Not “AI Is Fake”
The core problem isn’t that large language models don’t work — plenty of people, including everyone reading this on a tech site, use them daily. The problem is how much of the entire stock market’s value now sits on the bet that a handful of companies convert that usage into profit on an aggressive timeline. The top 10 S&P 500 companies account for roughly 41% of the index’s total market cap while generating only about 33% of its earnings — a gap that has to close somehow, either through those companies growing into their valuations or the market repricing them downward. Goldman Sachs’ own Jim Covello put it bluntly: “At some point, you’ve got to make money,” adding that enterprise AI returns have been “underwhelming so far” relative to the infrastructure spend behind them.
The Uncomfortable 1999 Comparison
Ray Dalio, who has spent decades studying market bubbles at Bridgewater, said current indicators show markets “rising close to — not at — the same level in 2000 and… 1929.” Acadian Asset Management’s Owen Lamont published research in June showing that after past tech booms, earnings consistently grew slower than analysts projected — averaging about 7% actual growth against roughly 13% forecasts going back to 1985. Even JPMorgan’s Jamie Dimon, who has every incentive to talk his own book bullish, admitted “it’s gung-ho, folks… there’s a lot of exuberance out there,” name-checking 1972, 1986, 2000, and 2007 as the bubble years this one echoes. When the people running the banks selling AI infrastructure are the ones flagging the exuberance, that’s not noise — that’s a tell.
Devil’s Advocate: Why This Time Might Actually Be Different
To be fair to the bulls, there’s a real argument here. Unlike 1999’s dot-com names — many of which had no revenue, let alone profit — today’s AI spenders are the most profitable companies in corporate history, funding this buildout largely from their own cash flow rather than debt or speculative IPO proceeds. Software engineer Benjamin Horne’s counterpoint deserves airtime too, though it cuts the other way: he’s argued that once you “strip out the massive token subsidies” propping up usage numbers, a gigantic chunk of apparent demand evaporates, since most real-world LLM use — summarizing documents, rewriting emails, basic search — doesn’t actually require frontier-tier models at frontier-tier cost. That’s the tension in one sentence: the companies funding this are genuinely healthy, but a meaningful share of the demand curve they’re betting on may be artificially cheap, subsidized usage rather than durable revenue.
Why This Should Matter to You, Not Just Investors
Even if you don’t own a single share of Nvidia or Microsoft, this selloff affects the products you’ll be using next year. A chunk of the “free” or heavily discounted AI features baked into your phone, your browser, and your favorite apps exist because investors were willing to fund years of losses in exchange for future dominance. If that funding tap tightens — which is exactly what a serious correction does — expect subscription price hikes, feature paywalls, and slower rollout of the AI tools that tech coverage (including ours) has been writing about all year. We’ve already tracked security incidents tied to rushed AI deployment and the chip-supply scramble feeding this spending boom — a slowdown in capital doesn’t undo the technology, but it absolutely changes the pace and shape of what reaches consumers.
What Would Actually Change My Mind
I’d drop this stance the moment enterprise AI spending starts showing up as durable, non-subsidized revenue rather than usage growth propped up by below-cost pricing — real proof that the ROI Covello says is “underwhelming” has turned a corner, sustained over multiple quarters, not one good earnings call. I’d also take it seriously if hyperscaler capex growth started decelerating on its own, the way a healthy market self-corrects rather than getting force-corrected by a crash. Neither of those has happened yet. What has happened is the same handful of voices — bank CEOs, veteran macro investors, the analysts who build the earnings models everyone else trades on — independently arriving at the same uneasy conclusion, from different angles, in the same six-week window. That kind of convergence is rarer than a single bearish headline, and it’s usually worth more attention than the exuberant case gets from the people currently benefiting from it.
Where This Leaves Us
The AI stock wipeout is a legitimate warning sign — not that the technology is a dead end, but that the market’s timeline and the technology’s actual maturity have drifted apart, and the gap is now wide enough that even the banks funding the boom are saying so out loud. History doesn’t say bubbles mean the underlying tech was worthless — the internet obviously wasn’t, after 1999 wiped out trillions in paper value. It says the companies and use cases that survive a correction are rarely the ones the hype cycle crowned as winners at the peak. That’s the pattern worth watching, regardless of what any single stock chart does next week.
Deixe um comentário