Anthropic is going public soon, and OpenAI disclosed a trillion dollars of capital-expenditure obligations last October.
The public reading of Dario Amodei's letter calling for a slowdown is that the risk has become too large to manage. Ram Ahluwalia's reading is that private venture markets pay for momentum and storytelling, and public markets do not.
"So this is all coordinated. When they come out and they concur, that means the message and the memo was socialized. Conversations were had, messages were sent, everyone was aligned."
Ahluwalia runs Lumida Wealth, which invests client money in alternatives and digital assets; he says he was one of three artificial-intelligence bulls at a dinner of investors last year and that he gets more constructive on semiconductors with every passing day.
The full episode is covered here so you can skip it. 17 minutes of audio, 12 minutes of reading.
Here are the 10 arguments that matter.
👤 Speaker: Ram Ahluwalia, CEO and Founder of Lumida Wealth Management, a registered investment adviser focused on alternative assets and digital assets
📰 Published: 16 September 2026 on the Lumida Wealth YouTube channel and podcast feed
🔴 YouTube | ⏱️ 17 min | ✅ Time saved: 5 min
Key Takeaways
He reads the coordinated slowdown letters as capital-markets positioning ahead of an Anthropic listing, not as a change in the risk assessment
Public markets impose discipline that private rounds priced on storytelling do not
A falling share price after an initial public offering turns vesting stock into a retention problem, which is what he thinks the labs are managing
He says the chips get bought regardless, because Meta will absorb whatever capacity the others give up
Chain of reasoning and looping are engineering workarounds, not discoveries on the order of the transformer paper
He compares them to duct tape, and says reinforcement learning was already on the whiteboard
People leaving the frontier labs are telling him development is going more slowly than the public thinks
A 90% priced-in rate increase means the surprise is asymmetric: no hike produces a rally, a hike shifts everything to the guidance
He says the Fed should never have cut last year, and should not be tightening two months before an election
1. Rates Are Topping Out
The episode opened on the Federal Reserve decision two days ahead, and his position on yields was that the move is close to over.
"I think you're topping out around here. This is what capitulation looks like."
"90% chance. That means it's largely priced in." That is the market-implied probability of an increase at the coming meeting.
The asymmetry is what he wants listeners to hold. "So, if you don't get a rate hike, markets are going to rally quite a bit."
"Now, if you do get a rate hike, what'll matter is the forward guidance. Is it one and done? Is it an adjustment hike? Or are we expecting a trio of these hikes?"
"I don't think rate hike is a foregone conclusion." His reading of the committee is that Christopher Waller was fairly dovish a few weeks ago and Neel Kashkari is always hawkish — a man, he said, who has been in bonds a long time and has not really invested in stocks.
On the chairman he thinks the hawkishness is rhetoric. Warsh has been talking about productivity gains from artificial intelligence, which Ahluwalia read as speaking softly and carrying a big stick: "So he's just jawboning here."
2. The Flattening Case
The counter-intuitive part of his rates view is that a hike would be good for the long end.
A credible tightening posture is what brings long yields down, because it is evidence that the central bank will fight inflation.
"You could see a world where short-term rates go up and long-term yields go down. It's like a flattening."
His complaint is about the cuts that came before. "They never should have done these rate cuts." The economy already had momentum, with capital spending and government spending running.
"The economy is like a massive tanker in the ocean. You're not supposed to be changing rates like this quickly. You have to be more deliberate, longer term oriented."
So any move now is corrective rather than new policy — undoing the recent cuts.
The timing still bothers him. "And again, you have midterms coming. It doesn't make any sense, I don't think, for them to hike rates this close to the elections."
3. The Coordinated Memo
Turning to the letter from Anthropic's chief executive arguing for a slowdown on grounds of catastrophic risk, Ahluwalia's first observation was about the chorus rather than the content.
Sam Altman and Elon Musk concurred, which he does not read as coincidence. "So this is all coordinated. When they come out and they concur, that means the message and the memo was socialized. Conversations were had, messages were sent, everyone was aligned."
The exception was Meta. Mark Zuckerberg's response, as he relayed it, was "We need more AI faster. We need to go faster."
He placed that response in sequence behind the president, who had called artificial intelligence a major opportunity and reinforced it at a summit by phoning Jensen Huang, who put him on speaker.
His aside on the room was about the market capitalization in it, which he put at seven to eight trillion dollars with Meta's president, Huang and other companies present. He said he wished he had been there and joked about sending an artificial-intelligence agent on a drone to take notes.
4. Why Anthropic Wants It
His explanation for the timing is the financing calendar, and it is the core of the episode.
"Anthropic is going to go public soon."
OpenAI needs capital too. "Last year in October 30th, Sam disclosed they have a trillion dollars in capex obligations." His summary: "You got to pay those bills."
"Anthropic also has a lot of capex obligations." He reads its obligations as smaller than OpenAI's, which he offered as the reason Claude Code runs more slowly than Codex, and said the company now has to manage to free cash flow and earnings.
The two markets pay for different things. Private venture markets are trend-driven, run on momentum and storytelling, and can support far more froth; public markets impose discipline, and everyone can see how recent listings have traded after their offerings.
The consequence is a talent problem, not a valuation problem. "He's lost a lot of talent to Anthropic already." Amodei himself left OpenAI to build Anthropic.
"If your stock price drops after you go public, you have these RSUs, you have these vesting schedules, they're underwater, and you start to lose even more talent."
5. The Chips Get Bought
Whatever the labs say about pace, Ahluwalia's investment conclusion does not move, because the demand for compute does not depend on any one buyer.
"Any chip that flies off of TSM's conveyor belt is going to get bought." He said the same of anything coming off Micron's or SK Hynix's lines.
Meta is the marginal buyer that makes it true, taking up whatever spare capacity the others leave.
"Will the chips get bought? The answer is yes." That, he said, is what matters for the forward path of earnings.
"What does slowdown mean? It might impact something in 3 years from now." Three years, in his framing, is forever away, and a great deal of new information arrives before then.
The demand he expects to arrive includes new entrants, open-weight models and sovereign buyers such as the United States government and Saudi Arabia.
American small business is barely started. He cited a penetration rate for these models among small businesses of about 20%: "So, we got a long ways to go."
"And what about those humanoids? We haven't seen those yet." That is another bid for compute.
"So, I'm not worried about this semi compute story at all."
6. A Slowdown Nobody Says
The part he flagged as least discussed is what people leaving the frontier labs are telling him privately.
Friends at the frontier labs, and people who have left recently, have started to suggest that the pace of development is going more slowly than people think.
He worked through what that means for the definition of general intelligence, noting there are differing definitions and that one is simply being generally intelligent. Jordi Visser's version, which he relayed, is a number: "Jordy called AI today is 150 to 160 IQ."
His own test is practical. "A practical test is can AI do real work? The answer is yes, it can. Can it do labor substitution? No, it can't. Can it do that soon with agents? Yeah, probably next year."
"Like certain functions are just gone. You don't need a UX designer anymore, for example, right? But that's a very narrow slice of the labor market."
7. Hacks, Not Breakthroughs
His technical argument is that the last two years of progress came from engineering rather than from discovery.
Chain of reasoning — sequencing prompts that feed on each other — he calls a clever engineering workaround. "And it's a token consumption hog."
The second is looping, and he credited Meta's new head of artificial intelligence with it, along with Zuckerberg for swapping out Yann LeCun, the Turing Award winner, for a hands-on builder who delivered Muse, which he called incredible and first in response time.
The loop puts two models in a cycle: one writes code, the other acts as the evaluation function and assesses the output, and the pair iterate until the result improves. "And he said these setups are creating the output of 100 engineers."
The benchmark he measures these against is one paper. "So the great breakthrough in AI was the paper attention is all you need." He called it the seminal breakthrough of the field: "It will go down in the history books as one of the greatest papers ever written", alongside the peer-to-peer electronic cash paper and Einstein's theory of relativity.
Against that, chain of reasoning and looping are not novel insights. Reinforcement learning is in the same category: "But the thing is RL is a well-known machine learning concept. It's on the whiteboard".
His image for the difference is a television character. "It's like duct tape. It's practical. It unlocks value. It's not an incredible breakthrough." MacGyver, he said, was his favorite show growing up.
8. Scaling Still Works
The one thing he does not dispute is that more inputs produce better models.
"You can throw more data, more training and more compute and the models get better." Trillion-parameter models are better, there is more data to train on, and more compute is coming.
So models and artificial intelligence will keep improving.
But the rate of improvement is the issue. "But I think part of the story here is innovation is slower than what they've expected" — which he offers as a second reason to manage expectations around a slowdown.
9. China Won't Slow Down
The external constraint on any American slowdown is that the other side has already declined.
China said publicly it is not going to slow down. Ahluwalia flagged that he had only seen the headline and wanted to look more closely at exactly who in China had said it.
"If China doesn't slow down, how does the US slow down?"
"Meta's not going to slow down. And if you have compute on demand, it gets monetized."
His analogy is launch capacity. "So, if you have Nvidia chips, you can monetize it. You spend on Nvidia, then you turn around and rent it, and that's how you make money." He tied the same logic to SpaceX being on track for a hundred billion dollars of revenue.
He gave a live example of spare capacity finding a buyer, saying OpenAI's excess compute let it generate revenue serving Claude's coding workload when Claude was overwhelmed.
10. Self-Regulation First
The last driver he names is political, and it has a date on it.
The midterm elections are coming and, on his reading of betting markets, the Democrats may take the House.
"Dario is trying to take the venom out of the sting by embracing self-regulation first", and so are Altman and Musk. He called it a very smart thing to do.
"You want to lead with self-regulation otherwise it's imposed upon you."
His criticism is of the delivery, not the motive. "I think this could have been communicated much much much better. There's still way too much drama around this stuff." He said he thought David Sacks's take on it was generally fairly accurate, from clips rather than the full episode.
His own assessment of the risk is that there is a long way to go before it is real. "They're going to unlock an incredible amount of opportunity, but we're not facing any kind of imminent catastrophic risk."
The cost of the current messaging, he argues, is adoption. "I had some family members ask me about this over the weekend", and his view is that the tone undermines national adoption of a technology he says will raise living standards for everybody.
Bonus Insights
The episode opens with an automated headline read covering drone attacks that forced Saudi Arabia to shut a key pipeline, a 10-year Treasury yield briefly above 5% for the first time since 2007, Wall Street Journal reporting on oil executives warning about a potential closure of the Strait of Hormuz, and worries about the pace of artificial-intelligence progress after a mathematics breakthrough. Ahluwalia credited Grok for the headlines.
His sentiment check is a dinner table. At an investor dinner last year he counted himself, Jordi Visser and Anthony Pompliano as the only bulls in the room. "You got a two-sided market. Everyone's not all in."
"Like every passing day, I'm getting more and more constructive on these semis." "It's like there's a lot of good value and opportunity there."
Visser, whom he interviewed on another podcast recently, spent his career at Weiss Multi-Strategy Advisers in New York.
Ahluwalia's bottom line is that the slowdown letters are a financing and political event rather than a technical one: the labs need public-market discipline and legislative cover, the underlying progress is engineering rather than discovery, and none of it changes the fact that every chip made will be bought.
Products, Companies & Tools Mentioned
Anthropic and OpenAI (The two labs he says are managing toward public-market discipline and a trillion dollars of capital-expenditure obligations respectively)
Meta (The buyer he says will absorb spare chip capacity, and the company behind Muse)
Nvidia, TSMC, Micron and SK Hynix (The supply chain he says will sell everything it makes whatever the labs announce)
SpaceX (His analogy for monetizing capacity on demand, and his hundred-billion-dollar revenue reference)
Watch the full episode:
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: