Wall Street Can Price an AI Failure. It Can’t Price Extinction.

Wall Street sees AI revenue and capital spending. Researchers see catastrophic extinction risk. Both stories belong to the same companies.

KEY TAKEAWAYS

• AI has become a central driver of equity returns and economic activity. The enormous capital expenditures behind it also feed directly into GDP.

• Two AI narratives are unfolding at once: one about revenue and growth, and another about catastrophic danger. Investors and researchers are examining the same technology from radically different perspectives.

• The practical investment risk is more likely reputational or operational than existential. Tesla and CrowdStrike demonstrate how damaged trust or a serious failure can quickly affect market value.

• Anthropic’s dispute with the Pentagon tests whether taking a principled stand can strengthen enterprise trust. That decision may sacrifice substantial near-term revenue while potentially improving the company’s long-term positioning.

• Markets can model spending, regulation, litigation, and slower growth. They cannot responsibly insert human extinction into a discounted cash-flow model.

MY HOT TAKES

• Wall Street is not ignoring AI’s existential debate so much as lacking a useful mechanism for pricing it. A risk without reliable probabilities or historical data does not fit neatly into conventional valuation.

• The most actionable AI danger for investors is an ordinary-looking corporate disaster. A defective product, damaging controversy, or regulatory confrontation is more likely to hit a portfolio than a science-fiction scenario.

• Anthropic is turning safety from a defensive policy into a potential brand advantage. Refusing certain government use cases appears to be a calculated bet that trust will ultimately command a premium.

• The AI investment thesis now depends on spending at a scale that leaves little room for disappointing revenue growth. Investors are betting that future cash flows arrive before today’s capital commitments become financially uncomfortable.

• Both the optimists admiring AI’s skyline and the skeptics inspecting its foundations may be right. The real mistake is assuming that either vantage point tells the entire story.

• You can quote me: “Nobody underwriting this IPO is modeling their own mortality into a discounted cash flow.”

Two sides–same story. I was walking my pup Eloise the other day–and I got thinking. 🤔 It was one of our favorite walking routes along the Hudson River. On those walks, I take in the views and marvel at the beautiful architecture set on a backdrop of an ancient natural wonder that has served as an economic artery for centuries. 🌆Ella–on the other hand is busy observing every fire hydrant, signpost, building foundation, and tree trunk (yes, we have trees in the city 😉). If you’ve ever visited NYC, you know that it has a rather distinct bouquet (that is putting it politely), and one can’t avoid taking in those either. For Eloise, of course, the smells and sights she is taking in are completely different from what I am.

This past week brought us some interesting insights on a much pondered topic in the markets. That topic: Artificial Intelligence. Have you heard about it? Lol, sure you have. No matter who you are, it has already touched your life in so many ways–some, you may not even realize. For investors, it has dominated the financial markets and economic narrative for the past couple of years. Rightly so, as it has been the driving force behind superlative returns in equities since entering the stage. It’s not just markets, it’s the economy as well. All that spending that we hem and haw about feeds right into GDP. While we all marvel at the ins, outs, ups, and downs of the AI trade, there is a completely different narrative that runs alongside what is happening on Wall Street. It would appear that some folks are not looking at the shiny skyline, but rather, the fire hydrants, the sign posts, etc. that sit below our eyeline, far enough for our noses to recognize certain nuances that may be of importance.

And this week, the fire hydrants had plenty to say. If you've been anywhere near social media, or even just glanced at a headline over someone's shoulder on the subway, you've probably caught wind of the latest round of AI doomsday talk–the kind that isn't coming from cranks on the fringe, but from people who've actually built the thing. Insiders, researchers, the people who sat in the room–going public, sometimes loudly, with the fear that what they helped create could genuinely get away from us. It's a strange kind of headline to sit next to quarterly earnings calls and CAPEX guidance, and yet there they are, running in parallel, barely acknowledging each other. One story about revenue. One story about the end of the world. Same company, same week, two completely different readers. Walk with me…

I got an email earlier in the week that put a finer point on it than I'd managed on my own. One of my favorite journalists from Axios wanted to know how investors are supposed to weigh all this talk–the kind Eloise would sniff out at ground level–about AI potentially being catastrophic, right as the biggest name in the space is getting ready to go public. Do investors mind buying something that could, in theory, kill them? It's a fair question, and it deserved a fair answer, so let me give you the long version here.

“We live in interesting times, indeed”–I said as much in my email back to her. It's telling that Anthropic's own founder has put a NUMBER on the odds that his creation could disrupt, or even endanger, humanity. That's partly marketing–safety as the top concern, as it should be–and it isn't just him; a chorus of researchers, current and former, have said versions of the same thing out loud this year. This week alone gave us a fresh case study on both ends of that chorus. Anthropic published its latest threat intelligence report, documenting misuse it disrupted across seven harm categories between December and August–cyber operations, surveillance, scams, and, in the case that ought to stop you cold, a blocked request for help drafting a grant application aimed at making a virus more dangerous. One day earlier, a researcher who'd trained models at both Anthropic and OpenAI resigned publicly, saying–in a thread that went viral before most of us had our morning coffee–that the people building this technology privately believe it could kill us all by decade's end (that’s right, KILL), even as they sound measured about it in public. That's Eloise at the fire hydrant. Fascinating, unsettling, and utterly unpriced by anything resembling…er, a financial model.

Here's the part I actually build into my valuation, though, and it's not the extinction scenario. The risk I'd underwrite is reputational, not existential–AI was always going to become an emotional flashpoint, transformational enough that sentiment alone has already swung valuations across the infrastructure names and the hyperscalers on very little hard evidence. Tesla is the useful template for how that plays out when a brand takes the hit directly: three straight years of declining brand value now, Musk's political detours a real and measured drag, and yet the stock itself has proven tradeable through all of it. CrowdStrike is the other template, and the sharper one for AI specifically–not a breach, not a doomsday event, just a flawed software update that knocked a meaningful chunk off the market cap in a single session and took a long, grinding stretch to claw back. That's the more realistic dress rehearsal for how an AI stumble actually shows up in a portfolio. Not the Terminator. An operational failure with a dollar figure attached.

You can see the same instinct playing out in Anthropic's running fight with the Pentagon, and I'd call it the cleanest real-world test case we've got. The Department of Defense wanted unrestricted access to Claude, and Anthropic said no to autonomous weapons and no to domestic mass surveillance, and got itself labeled a “supply-chain risk” for the trouble–a designation that, as of late August, a federal judge has now ruled unlawful and retaliatory. That fight has REAL, measurable financial stakes attached to it, potentially billions in lost revenue, and Anthropic picked principle over revenue not once but twice, in public, ahead of the biggest liquidity event in its history. I don't think that's a company stumbling into reputational risk on the way to an IPO. I think it's a deliberate bet–that being seen as the lab willing to walk away from Pentagon money is worth more in enterprise trust than what it costs in the short run. There's a mountain of academic literature on whether that kind of corporate principle-taking actually pays off financially–I wrote my doctoral dissertation on a corner of it, so I'll spare you the full bibliography, but I'll give you the punchline: the research is inconclusive. 😉 The IPO is the live market test of that thesis.

And that test is happening at the exact moment Wall Street is running its own, entirely separate model on the same company–and the same sector. Alphabet posted its first negative free cash flow since its 2004 IPO this summer, burning through nearly $6 billion in a single quarter as capital spending topped $44 billion. Amazon's trailing free cash flow swung negative too, even with AWS revenue up better than a third year-over-year. Add it up across the big hyperscalers and you're looking at combined 2026 capital spending pushing past $700 billion, all of it a bet that the revenue curve bends upward before the balance sheets get too stretched to keep funding it. That's the rent roll. That's what Wall Street actually knows how to price–a dollar spent today against a dollar of cash flow expected sometime out past 2029, discounted back to the present with all the false precision our models are so good at pretending we have.

Here is the reality, and it's the same one I gave that journalist, just dressed up in the building metaphor. Markets are extraordinarily good at pricing risk that has a season, a probability distribution, and a claims history behind it–which is exactly why we can price a hurricane and can't price an extinction event. When Anthropic eventually files its public S-1, you will not find a line item for "the technology could end humanity." You'll find "public and regulatory backlash against AI"–a real, disclosed, quantifiable risk factor that is a proxy for the reputational dynamic, not a probability-weighted doomsday clause. Nobody underwriting this IPO is modeling their own mortality into a discounted cash flow–except maybe a life insurance actuary, and even he'd tell you the data doesn't exist to do it responsibly. What investors ARE modeling is whether backlash, litigation, or a Pentagon-style standoff slows revenue growth. That's tradeable. That's a very different animal than what Eloise is sniffing out at the two or three rows of bricks at the base of the building.

Which brings me back to the Hudson River. I looked up at that skyline again on the walk home, rent rolls and cap rates dancing in my head the way they do–and Eloise was still down at ground level, working through whatever story that fire hydrant was telling her. Same buildings. Different questions. Both of us were right about what we were looking at–we just weren't looking at the same thing, and neither, it turns out, are the two audiences currently deciding what Anthropic and its cousins are actually worth. With respect to the death of humanity at the hands of AI, I’d rather not put a real probability on that tail risk. I am willing to say it is not zero, but also not enough so that I would just go YOLO crazy and start reciting that famous economists’ saying “we are all dead in the long run.” Sleep easy for now…let’s revisit when those long-awaited S-1s come out.

YESTERDAY’S MARKETS

U.S. stocks fell for a fourth straight session yesterday, with the Dow Jones Industrial Average down 0.6% to 52,064, the S&P 500 off 0.58%, and the Nasdaq Composite down 0.65%. Oil prices pushed past $100 a barrel, with WTI crude settling at $102.48 (up 6.7%) and Brent crude at $107.63 (up 5.9%), both the highest closes since mid-May. Treasury yields rose broadly, with the 10-year note touching its highest level since 2023. August producer prices rose 0.4% month-over-month, matching consensus estimates.

NEXT UP

• CPI/Consumer Price Index (August) is expected to come in unchanged at 3.4% with Core expected to ease slightly to 2.4% from 2.5%. 👈This will drive markets today and in the days leading to next week’s FOMC meeting.

• University of Michigan Sentiment (September) may have slipped to 51.0 from 51.7.

• Next week we will get Retail Sales, housing numbers, Industrial Production, and Leading Economic Index. The event of the week will be the FOMC meeting and the press conference that follows. You better show up right here to stay ahead of the storm.