OpenAI's own internal projections, as reported by The Information and cited on the show, point to a $14 billion loss in 2026 and cumulative losses of roughly $44 billion by 2028. A single next-generation data center, Travis Hoium said, runs about $35 billion on its own.
The public version of the past week is that frontier-lab leaders have become frightened of their own technology. The reading this panel keeps returning to is that the same week is also the last one before two of those labs have to explain their spending to public-market investors.
"But I also think one of the loudest voices calling for caution are the same people holding the largest stakes. You have to question what the motivation is."
Tyler Crowe hosts the show; Rachel Warren and Travis Hoium are both stock market analysts at The Motley Fool, and between them they cover healthcare and the consumer economy.
The full episode is covered here so you can skip it. 27 minutes of audio, 18 minutes of reading.
Here are the 10 takeaways that matter.
👤 Guests: Rachel Warren and Travis Hoium, Stock Market Analysts at The Motley Fool
🎙️ Host: Tyler Crowe, Stock Market Analyst at The Motley Fool
📰 Published: 14 September 2026 on The Motley Fool
🟣 Apple Podcasts | 🔗 Episode page | ⏱️ 27 min | ✅ Time saved: 9 min
Key Takeaways
The people calling loudest for an AI slowdown are the ones with the largest positions in it
Bridgewater's Greg Jensen was an early investor in both OpenAI and Anthropic
The slowdown chatter lands in the same week as two IPO pipelines and a spending problem
OpenAI's projections point to a $14 billion loss in 2026, on the panel's reading of The Information's reporting
Compliance costs do not slow the biggest labs down, they price out everyone smaller
Nothing in the episode resolves the question, and the panel says that is the honest answer
Everyone involved can be sincerely worried and also acting in their own interest
Consumer discretionary is the worst-performing sector of the year while energy and tech are up sharply
Some beaten-up consumer names now trade on 10 to 15 times earnings
AI's real work in healthcare is unglamorous: compressing the decade and the billions it takes to get a drug approved
Contract research organizations are not dying from AI, but the ones billing for headcount are exposed
A dividend is a message about which investors a company wants, not just a use of cash
1. Slow Down, or Spend Less?
Crowe opened with the week's news and a hypothesis about why it all arrived at once.
His summary of the week: "There has been a fair share of we need to slow AI development chatter out there. But this past weekend, that conversation appeared to hit a fever pitch." He listed employees leaving Anthropic and OpenAI over safety concerns, and Sam Altman and Dario Amodei both calling for frontier-model development to slow down, "lest they go out of control" in what Crowe said were Altman's words.
He also flagged the hedge-fund contribution, which he treated with open skepticism: "Even the CIO of the hedge fund Bridgewater Associates was on podcasts over the past week, like talking about human extinction and the probability was higher than 10%, which is, you know, kind of silly if you think about it or startling depending on how you want to look at it."
The timing is what he thinks matters. "This isn't anything new, but it does appear to come at a very specific time where both OpenAI and Anthropic are on the precipice of IPOs. And spending on these businesses is getting tougher to swallow, especially at the frontier level where the bulk of their spending is going."
His own framing, offered as the cynical one: "The cynical view, at least to me, is that all of this slowdown chatter comes at a time when they want to slow down spending more than anything else without disrupting business growth."
The question he put to the panel was a spectrum, from "AI will be the death of us" at one end to a company managing a spending-versus-growth problem at the other.
Hoium's answer started in the middle and leaned cynical: "Honestly, I think the truth is probably somewhere in the middle, but I do tend to take a bit of a more cynical view to what we've been hearing." He was careful to add that this does not mean the safety concerns are fake — only that there is "a lot of calculations going on behind the scenes."
2. The Math Behind the Alarm
The case for the cynical reading was made with numbers, and they are the hardest thing in the episode.
The change in who is watching: "You're kind of moving out of the easy money venture phase. These are companies that are anticipating to have huge entrances into the public markets where they're going to face a very different level of scrutiny than they have in the private space."
On OpenAI, citing internal projections reported by The Information: "They're expecting a $14 billion loss in 2026. They could have cumulative losses of about $44 billion by 28."
The unit cost that makes those losses hard to grow out of: "You know, think about how a single next-gen data center runs about $35 billion on its own. This is not something that Wall Street is necessarily going to be forgiving on."
On Anthropic, whose chief executive has spent roughly a year telling the industry to slow down: "Meanwhile, we also saw a report from the information that Anthropic has locked in $517 billion in compute commitments. That's 14.8 gigawatts of capacity. It's enough to rival just for scale a dozen nuclear reactors." That commitment, the panel noted, is funded out of revenue and a confidential IPO pipeline, and much of the money goes to Alphabet, Amazon, Microsoft and SpaceX.
Hoium's closing point on scale is that this is not a side debate: trillions of dollars and several multi-trillion-dollar companies are on the other side of it, "So this isn't something that we should take lightly."
3. Who Owns the Warnings
The second half of the cynical case is about the people making the argument rather than the argument itself.
On Bridgewater's Greg Jensen: "He's not a neutral bystander. He was one of the earlier investors in both OpenAI and Anthropic."
What Jensen said, as relayed on the show: "He's the one that was telling Bloomberg on a recent podcast, there's a 30 to 60 percent chance of a catastrophic AI disaster in the next few years."
Set against his firm's filings: "Bridgewater's own SEC filings show that has been building up positions in Nvidia, Broadcom, Amazon."
The mechanism by which safety rules become a moat: "So we're seeing this push for regulation in compliance costs, but it's more likely to just price out the open source community and the small players who can't afford the legal overhead. The biggest labs can absorb it."
Hoium said he believes the risks are real and still thinks the incentive question has to be asked. "But I also think one of the loudest voices calling for caution are the same people holding the largest stakes. You have to question what the motivation is."
4. Everyone Can Be Right
The panel's answer to its own question is that the competing explanations are not mutually exclusive, and that the coordination is itself the odd part.
What struck the show about the weekend was the unanimity: the leaders of the frontier labs almost all agreed with what Dario Amodei had written, which was described as stark and as feeling "a little bit coordinated."
The refusal to pick a side is deliberate: "But I also came away thinking that all of these things can be true." They can be genuinely worried about safety; critics can be right that this is ladder-pulling and regulatory capture to build a moat; and everyone can also be wrong about all of it.
The credibility problem is longevity. "These companies are crying wolf a little bit because they've been doing this for years." The example given is Amodei himself, one of the industry's biggest critics, who went on to found what the show called arguably the leading lab — "a lot of cognitive dissonance going on here."
The historical read is that harm is likely and unpredictable: "And the history of technology says that something bad will probably happen. We just don't know what that is." The analogy offered was the internet, which nobody expected to produce isolation and mental-health problems among young people.
The oddest feature of the debate, in the panel's description, is that it sounds like the labs asking to be protected from themselves — that if they keep developing this, they will do something bad.
5. The Laws Already Exist
One strand of the answer is that the liability regime being asked for is largely already in place.
The argument: "And one of the things that's most resonant to me is that there are laws in place for a lot of these things. If you build a product that goes out and hurts people or steals things, it is your fault."
The Hugging Face incident is the live test. The panel's understanding is that laws were broken and felonies committed in it, and that the industry has been glossing over the consequences. The hypothetical extension: if AI systems break into financial institutions and steal money, does anyone go to prison?
The investing conclusion is that prevention is not how the United States works. "And ultimately, we're investors here. Something bad is probably going to happen. And the question is going to be then what are we going to do? Because especially in the U.S., we don't typically act first." Action comes after the event, as it did after the global financial crisis.
Crowe's response was to imagine the disclosure language. "I feel like reading the S-1, for Anthropic and OpenAI of seeing like in the risk sections like, our AI agents might commit felonies and we don't know if it might happen." He said he does not normally read the risk section line by line and might have to now.
6. The Regulatory Drawbridge
Crowe put a specific historical pattern to the panel: that this is the Silicon Valley script, not a new event.
His framing: "And it really feels like the Silicon Valley playbook that we've seen before, where it's this land and expand role of like you saw with Google, where it dominated search or Meta where it started to dominate social media." Regulation then arrives — his example was GDPR in Europe — intended to make things fairer, and what happens instead is the opposite: "But what ends up happening is these giants, are the only ones that can handle the regulatory compliance to make it happen."
Warren declined to make it purely cynical. "I think it's a nice happenstance that it also is to their competitive advantage if the latter is pulled up." But she thinks the fear is genuine, for a reason she acknowledged sounds strange on an investing podcast.
Her explanation of why: "Again, we're an investing podcast, but there is much more of a religious view to this in a lot of ways because they don't really know how this thing works. They can't explain to you why AI is doing the things that it's doing." The sequence she describes is sincerity first and commercial benefit second — "And oh, by the way, if it happens to help our business, that's a nice byproduct."
Hoium restated his position and then conceded the other half of it. "I mean, look, the genie is out of the bottle, so to speak." He repeated that the unity among lab leaders traces back to growth expectations as they enter the public markets.
The concession is that the technology is genuinely ahead of its makers: "But I do also think that we see that these leaders are talking about a technology that they have helped to create and develop that is rapidly outpacing their ability to control and even fully understand it."
What follows from that, on his account, is that the brake is the only available instrument. "Well, at least at the short term, the answer is you have to pull the brakes a bit. You have to scale back development to understand how to regulate it." He called AI much harder to set legal guardrails around than anything Silicon Valley has produced before.
7. What We Don't Know Yet
Warren interrupted the segment with a possibility nobody on the show could rule out.
"Tyler, I just want to bring up one quick thing. I wonder if something happened at one of these labs that we don't yet know about."
Her precedent for the delay is the same incident the panel keeps returning to: "We didn't know about the Hugging Face incident until long after it happened."
The pattern she is pointing at is the sudden agreement. "Is there a reason that all these people who seem to be fighting each other to lead the world, the AI world, why are they suddenly all on the same page?"
Crowe extended the joke to a hypothetical news cycle two months out, in which it emerges that there was an incident at OpenAI, and then moved the show on, promising a palate cleanser after the break.
8. The Consumer Underneath
Asked what he is watching into the end of the year outside the frontier-lab story, Hoium named the consumer, and he named it as a worry rather than an opportunity.
The structural question: "I have to wonder how much of the economy is being held up by that AI trade." The market is at or near all-time highs, and the spending driving it is concentrated in one place.
The damage is already visible in the tape. "And as I look at consumer, particularly shoe and apparel stocks have gotten absolutely hammered this year. Restaurant stocks have not done well because, you know, same store sales comps are rough. Margins are getting squeezed because they have no pricing power." Gasoline prices, he added, keep rising with no sign of stopping.
The risk he is describing is a sequencing risk. "But if that AI trade does slow down and then we find out we have a weak consumer underneath, that's probably not great for investors." His diagnosis is that the market is thinking about two things, and the AI trade is loud enough that the consumer has fallen by the wayside.
The other side of the same fact is price. "But that said, I'm finding a lot of really interesting opportunities because some of those stocks that have been beaten up are trading for 10 to 15 times earnings." For a company that can grow long term, he said, this is an interesting time to start building positions.
Housing is the connected leg. Home equity loans are, in his description, almost a thing of the past because rates are prohibitively expensive, and a depressed housing market feeds through into discretionary spending.
The sector numbers he read off a chart during the recording: "I'm looking at a chart right now and consumer discretionary is the worst performing sector right now on the market. It's down 5.3% year to date. Well, things like energy and tech are up 28 to 30%."
9. AI's Quiet Job in Pharma
Warren's answer to the same question was healthcare, and she offered it explicitly as the positive counterweight to the first segment.
What AI is doing there is not flashy: "It is really helping to refine and optimize the long help processes of drug development, discovery, optimizing clinical trials." Her caveat: "It's not replacing doctors and scientists."
The size of the problem it is attacking: "I mean, you have to think about how the average drug, it can take over a decade, a couple billion dollars on average to bring a drug to market. 90% fail in later stage trials anyway."
Her framing of the role: "And so AI is really being used as an automation and data engine to attack a lot of the bottlenecks that have historically made drug development so slow."
Her first example is Moderna, whose cancer vaccine work with Merck has drawn investor attention, and its system Maestro, which orchestrates personalized vaccines: "It reads a patient's tumor sequence. It predicts the best immune response."
Her second is a smaller name: "Krystal Biotech is another really interesting company, they make redosable gene therapies for rare skin disease, and they are using AI and machine learning to run real-time quality control in their production lines."
Crowe put the contract-research question to her — MedPace Holdings, IQVIA and their peers, which he called "the worker bees of the healthcare industry" and the firms that actually run clinical trials for startup biotechs. The bear case is that AI removes the drug-discovery work they sell; the bull case is that faster discovery means more candidates and therefore more trials to run.
Warren's answer is that it splits the industry rather than ending it. The exposed model is the billable one: "I mean, like, historically speaking, CROs would make money by billing for the number of people typing data, monitoring sites, redlining contracts." Those rote administrative tasks are what user-friendly AI tools are absorbing, and some biotech startups have delayed signing CRO contracts.
The offsetting force is volume. "But because AI is accelerating the early stage drug discovery process, which is one of the most time intensive phases of developing a drug, we are seeing a lot of opportunities for these more tech-fluent CRO organizations." She expects large, well-connected firms such as MedPace to be fine, and said "There are certainly some of these kind of old school mom and pop organizations that may not stand up to the test of it."
10. Dividends and Who Buys
The mailbag question came from a listener, James in Atlanta, who said he understands dividends from energy companies but struggles with growth companies paying them.
The question as Crowe read it: "So the question is, we'd love to hear everyone's thought on how you view dividends, both in the return calculation and the company's messaging." The listener cited ExxonMobil and Chevron as the models he understands and Disney and Alphabet as the ones he does not — "I struggle to see the value of a growth stock paying one."
Hoium's first answer is that a dividend is a signal about the shareholder register. "One of the things you have to look at is a company is trying to communicate with investors, what sort of investor they want to attract."
He split the listener's two examples. "So when you bring up two interesting examples, you know, Alphabet would be a case of, hey, we got so much cash. We got to return this somehow. Dividend, I guess." Disney, he said, is different: it wants dividend investors in its pool, and historically that is where it plays.
His proposal is for something American companies mostly refuse to do. "With that said, I would love to lobby for irregular dividends." The problem he is naming is the expectation of permanent increases: "If you own a private business, you have a great year. You just pay yourself more this year and maybe don't pay yourself as much next year if it's not such a good year. We should do that in public markets."
His warning about what a dividend usually indicates: "But that said, one of the things that I think about with dividends is it's actually a lot of times a sign that a company doesn't have as many investment opportunities as they once did. You don't see young companies paying dividends."
The failure mode he traces is circular: a weak underlying business forces a dividend cut, the cut drives out the investors who bought it for the dividend, and the share price falls further.
His bottom line: "But at the end of the day, whether you're looking at dividends or not, make sure the business is on solid footing because that's ultimately the most important thing."
Warren's case for dividends starts by rejecting the premise of the question. "And I mean, just injecting cash back into the business doesn't necessarily create more growth."
What she reads a dividend as: "I mean, it's a discipline signal, right? I mean, it's telling you that a company isn't going to hoard cash when the balance sheet can afford it." It is also cash on a schedule, which can be reinvested.
She added the demand argument — a large share of funds and conservative pension money is mandated to hold only dividend-paying stocks, so paying one widens the buyer base and puts support under the price.
Her caveats were tax efficiency against buybacks, and the fact that a dividend is not a guarantee: "Even Disney pulled its dividend back during the pandemic briefly." She holds both growth and dividend businesses herself.
Bonus Insights
Hoium has pitched the irregular-dividend idea to a chief financial officer in person and been turned down flat. "I interviewed a CFO in the oil and gas industry once, and I proposed that idea of like an irregular dividend." The reply, as he told it: "And he's like, I would get murdered in the capital markets if I ever tried to do that."
His worked example of a company that does pay this way is a casino operator he named as Melco Crown, while noting the name has changed more than once, which he said pays out a fixed share of net income so the dividend moves around. The trap he wants to avoid is the opposite case — a 7% yield being paid out of more than the company's net income and free cash flow, which he called nuts.
Hoium's caveat on his own proposal is that it cannot be introduced mid-life: "But you have to kind of establish a track record with the market and be like, This is what we do. And if you haven't done it before, you're probably going to get in trouble with the market when you do it."
Crowe noted that Hoium was effectively arguing for American companies to behave more like European ones, which he said might be a first on the show.
The panel's bottom line is that the AI safety argument and the AI spending argument arrived in the same week for reasons nobody can disentangle from the outside, that the regulation being asked for would fall hardest on everyone except the labs asking for it, and that the more immediately investable questions are elsewhere — in a consumer sector where beaten-up names are already trading at 10 to 15 times earnings, and in the unglamorous work AI is doing inside drug development.
Products, Companies & Tools Mentioned
OpenAI and Anthropic (Both approaching IPOs, both led by executives now calling for a slowdown, and both carrying spending the panel says public markets will scrutinize differently)
Bridgewater Associates (Greg Jensen's warning about a catastrophic AI disaster, set by the panel against the firm's own disclosed positions)
Nvidia, Broadcom and Amazon (The names the panel says Bridgewater's SEC filings show it building up while warning about AI risk)
Alphabet, Microsoft and SpaceX (Where much of Anthropic's compute commitment is said to be going; Alphabet also features in the dividend discussion as a company with more cash than uses for it)
Hugging Face (The incident the panel keeps returning to as the test of whether existing law already covers AI harm, and as evidence that these events surface late)
Google and Meta (Crowe's precedents for the land-and-expand pattern in which regulation arrives after a company is already dominant)
Moderna and Merck (Warren's lead example: the cancer vaccine partnership, and Moderna's Maestro system for orchestrating personalized vaccines)
Krystal Biotech (Redosable gene therapies for rare skin disease, using machine learning for real-time quality control on the production line)
MedPace Holdings and IQVIA (The contract research organizations at the center of the question about whether AI kills or feeds clinical-trial services)
Disney (The listener's example of a growth company paying a dividend, and Warren's example of a dividend that was pulled during the pandemic)
ExxonMobil and Chevron (The energy dividends the listener says he does understand)
If this was worth your time, send it to someone closer to the industry than you are.
Get the latest market chatter as it happens: