Bad News, Good News, Who Knows
You know the story of the farmer whose horse runs away—the neighbors call it bad news, the wise farmer just shrugs and says “who knows,” and sure enough, the horse comes back leading three more. Good news, bad news, who knows: he never rushes to a verdict, because he’s learned that today’s calamity often turns out to be tomorrow’s blessing, and vice versa.
I think about that farmer a lot these days, because the headline out of the AI world this week is that things are slowing down. And I’m not sure yet whether that’s bad news or good news. Who knows.
The Slowdown Nobody Wanted to Admit
Dario Amodei can be right and wrong at the same time. He may be absolutely right that frontier AI models are advancing faster than our ability to understand, secure, and govern them. But he’s also talking his book, and it’s worth being a little suspicious of how convenient that overlap is. Anthropic is preparing for an eventual IPO, and Amodei has found a pitch that does double duty: telling investors that Anthropic has built the most valuable product humanity has ever conceived of, while simultaneously convincing them the company alone can be trusted to wield something that powerful responsibly. The warnings about existential risk aren’t just caution—they’re also marketing. The scarier the technology sounds, the more essential Anthropic’s stewardship of it becomes, and the more that stewardship is worth to the people writing checks.
Elon Musk has his own motives. He’s spent years warning about AI risk while simultaneously raising capital and pushing xAI forward as aggressively as possible. A coordinated slowdown would help preserve the valuations of the leading companies, prevent everyone from destroying themselves in an expensive arms race, and give xAI, Anthropic, and OpenAI more time to grow into the enormous valuations they already need.
Slowing Down Without Actually Slowing Down
That may be exactly the needle the industry needs to thread. Slowing the release of frontier models does not require slowing data-center construction. So far, there’s no evidence that spending on chips, power, and data centers has meaningfully slowed. It doesn’t need to. These companies can keep building the infrastructure while taking more time to test their models, develop safeguards, and—most importantly—find sustainable ways to monetize the software.
This is also about justifying a very high valuation at IPO, and then achieving it. If Anthropic or OpenAI go public at a valuation approaching $1 trillion or higher, that valuation opens the door to enormous amounts of capital. They can then reinvest that capital back into the system—buying chips, leasing data centers, funding research, building the next generation of models. The safety narrative buys them time; the IPO supplies the capital; the capital keeps the ecosystem growing.
The Real Test Comes Later
But the real test won’t be the IPO. It will come two years afterward, when investors ask whether these companies can generate enough revenue and profit from the software to justify the valuations and the infrastructure spending. At some point, “this technology will change humanity” has to become a repeatable business model. The industry has to prove that AI can create economic value beyond raising money and buying more compute.
What Investors Should Take From This
We need more cybersecurity, because increasingly capable agents create more vulnerabilities. We need more data centers, because even a more deliberate frontier still requires enormous computing capacity. We need to stay ahead of China, because America can’t unilaterally stop while its strategic competitors keep moving forward.
I believe capitalism will work this out. Capital will flow toward the companies producing genuine value and eventually away from those that can’t monetize their technology. There will be excess, volatility, and failures along the way, but competition will force these businesses to become more efficient, safer, and ultimately profitable. America should build the infrastructure, secure the technology, and win the race.
Bad News, Good News, Who Knows
So is the AI slowdown bad news or good news? I keep landing back on the farmer, shrugging at his neighbors. Maybe the slowdown is a sign that the industry finally has to answer the question it’s spent years avoiding—can this actually make money, safely, at scale? Or maybe it’s just the version of the story where the horse runs away, and we’re all standing at the fence assuming the worst before we’ve seen what comes back with it.
Bad news, good news, who knows. We wait and see.