Wall Street analysts had Nvidia growing revenue 36%; the company guided to 70% and is approaching $100 billion a quarter. Jason Bodner was in the room in San Francisco for that call, and he has built his current buy list entirely out of what the company said next about its own bottlenecks.
The week's headlines said the opposite — Dario Amodei and Sam Altman calling for a slowdown, a former researcher putting a greater than 10% chance on AI killing everyone within a decade. Bodner's reading is that the people asking to be regulated have looming initial public offerings and large press operations.
"They meter messaging, they collar messaging, they spin it."
Bodner runs the research firm Money Flows, which tracks unusual institutional buying in individual stocks, and spends his working week on company-by-company research.
The full interview is covered here so you can skip it. 34 minutes of audio, 15 minutes of reading.
Here are the 10 calls that matter.
👤 Guest: Jason Bodner, who runs the research firm Money Flows, which tracks unusual institutional inflows into individual stocks
🎙️ Host: Bridget Bennett, who presents MarketBeat's investing videos and its Monday live shows
📰 Published: 15 September 2026 on YouTube (MarketBeat)
🔴 YouTube | 🟢 Spotify | 🟣 Apple Podcasts | ⏱️ 34 min | ✅ Time saved: 19 min
Key Takeaways
The AI safety warnings are a lobbying move, not a confession, and they raise the barrier to entry for everyone smaller
Licensing and capital minimums would let the incumbents buy the companies that cannot clear them
Nvidia guided to 70% revenue growth against a 36% analyst estimate, and named its own two bottlenecks
Memory first, for at least two years; silicon photonics after that
The three largest memory makers all publicly agree the shortage runs for years
SK Hynix's chief executive sees it out to 2030; Samsung says through 2027 into 2028
This is a sentiment problem rather than an earnings problem: 52% earnings growth in the second quarter, 30% without Google and Meta
Micron's forecast earnings go from $73.23 a share this year to $155 next year and $169 in 2028
It trades on a forward price-to-earnings multiple of 5.9 against that
Nvidia is putting $2B into Coherent and has signed a multi-billion-dollar purchase commitment with it
Tower Semiconductor is the wafer layer underneath photonics, with earnings forecast to double and then nearly double again
Even a full stop to AI development would not cut demand, because adoption is still running ahead of capacity
1. The Warnings Are a PR Play
Bennett opened on the week's dominant story: Anthropic, OpenAI and others calling for the brakes and for regulation, with the market selling off on it. She asked Bodner for his take.
He started by conceding the headlines are frightening. Dario Amodei called for pumping the brakes to make sure development is safe; Sam Altman said he supports that wholeheartedly; a former researcher at one of the major companies said there is "a greater than 10% chance that AI kills all of humanity in less than 10 years."
Then he asked what the speakers get out of it. "But I asked myself, why would major CEOs with looming IPOs say this type of stuff?"
"They meter messaging, they collar messaging, they spin it." His point is that these are companies with large public-relations departments that know exactly what they are doing, and that Altman has been pushing for regulation since 2023.
The mechanism he describes is regulatory capture, and he spells out each step. Frightening the public forces the government to take the subject seriously. Taking it seriously produces licensing and minimum capital requirements. Those requirements entrench the existing companies for years and make AI part of everyday life — allowing it into health care, into government, into everything.
The competitive consequence is the part he cares about: the barrier to entry for smaller competitors rises, so the incumbents can cherry-pick and buy up other AI companies.
His historical analogy is aviation. Airlines themselves asked for a federal commission to regulate everything because they could not do it alone. "That didn't ground airplanes. It actually just made it more accessible for billions of travelers."
Bennett extended it: the first passengers had to trust the aircraft would fly, and making consumers feel safe often means government regulation.
2. September's Worst Weeks
Asked how the market is responding, Bodner put the selling in a seasonal frame and walked through his firm's own research on the screen.
His calendar of the year so far: April and May were huge rallies out of dire March lows, June lost momentum around the SpaceX listing, July was whippy for AI names — especially around the Situational Awareness blow-up — and August was surprisingly strong.
The research he showed measures weekly rather than monthly returns for the S&P 500 back to 1990, through midterm election years. The first week of September is usually decent and the second week reasonably strong, at 20 basis points. Weeks three and four are where it turns, and week five is historically the weakest of all.
The segment was recorded in week three, so his read is that the market is heading into the historically ugly stretch, with more of it in the first week of October.
After that the seasonal picture flips. Through October and into November it tends to be smooth sailing as clarity arrives on who is winning the midterms, and November and December tend to be extremely good months in a midterm year.
The longer series he cited runs to 1926: quarters one through three of a midterm year are historically weak for the S&P 500, down 1%, and the fourth quarter is where the gains are.
His conclusion from that is a bullish setup despite the fear.
Bennett added the near-term catalyst on top of the seasonality: a Federal Reserve meeting the following Wednesday, landing in an already weak week.
3. Sentiment, Not Earnings
Bennett asked what a historically weak stretch plus two large negative catalysts means for a longer-term investor. Bodner's answer separated the mood from the fundamentals.
He listed what Wall Street is reacting to: a war in Iran, oil at $100, fears over the Fed, inflation, low approval ratings for President Trump heading into the midterms, and general uncertainty. Wall Street hates that, so it reduces risk — and once clarity arrives, he says, the setup is there.
His bullish case extends past the fourth quarter. History says not only is the fourth quarter of a midterm year strong, but the following 18 months tend to be too.
The diagnosis he keeps returning to: "This is a sentiment problem, not a valuation problem, and not an earnings problem."
The evidence is the second-quarter reporting season just finished. Roughly 86% of the S&P 500 beat earnings estimates and roughly 76% beat sales estimates, and the index reported 52% earnings growth for the quarter, which he called shocking.
Strip out what he described as the one-time growth at Google and, he believes, Meta, and it is still 30% earnings growth.
"So companies are firing on all cylinders. They're making more money. The AI buildout is real."
The imbalance underneath it is where he sees the opportunity: "Too much demand, too little supply for hardware supporting this AI buildout."
4. Nvidia's Blueprint
Bodner said he was in San Francisco two and a half weeks earlier listening to the Nvidia earnings call, and that everything on his list comes from what was said on it.
The guidance itself: "He shocked the world. He doubled projected estimated revenue growth." Nvidia foresaw 70% where Wall Street analysts saw 36%, and the company is approaching $100 billion of revenue a quarter.
The company's own constraint was memory. Nvidia said memory is a bottleneck for at least the next two years, that it has shored up its own supply, and — Bodner paraphrasing — that the numbers would be better still without the bottleneck.
Three of the largest memory makers said the same thing independently. SK Hynix's chief executive called it the worst year for supply in the industry's history and sees a bottleneck out to 2030; Samsung expects shortages to persist through 2027 into 2028; Micron said much the same. Together, Bodner said, they account for a vast majority of memory.
The second bottleneck Nvidia named is silicon photonics, which Bodner then explained for the audience. A chip is connected to memory by copper wire, which is very fast at the chip level. As chip counts rise, copper runs into resistance, heat and speed problems, especially over distance. The fix is to move fiber optics and lasers down to the chip itself.
His argument for why the bottleneck persists even in a slowdown is adoption, not development. The pulsing "thinking" indicator on a chatbot prompt is capacity failing to keep up with everyone using it at once — so even if model development stopped on the chief executives' warnings, the bandwidth problem remains.
"Memory and photonics are the two biggest bottlenecks to the AI infrastructure."
5. Micron's Earnings Ladder
The first of his three names is the one he expects viewers to know. He was explicit that he is a proponent of understanding what you own, and defined the product before the position.
High bandwidth memory is the memory stacked at the GPU level to increase the speed and depth of data transfer into the processor. SK Hynix is the number one manufacturer; Micron is number two and, he said, the fastest-growing.
His pricing argument follows directly from the supply commentary. With demand spiking and supply spoken for years ahead, prices head higher and so do operating margins — and Nvidia is working with memory suppliers to add capacity because it needs more.
His image for why the memory matters: a GPU is the world's fastest chef, and ingredients spread all over town slow it down.
The earnings ladder he showed is the core of the call: $73.23 a share forecast for full-year 2026, $155 a share the following year — a double — and $169 a share for full-year 2028. He described the price/earnings-to-growth ratio as very cheap and the forward price-to-earnings multiple as 5.9.
His caveat is that this assumes the estimates hold, and he expects them to keep being revised higher given how far second-quarter results beat.
On the objection that the stock has already run, he was unmoved: "Guys, that's just the beginning." He put the run at 800% to 900%, and said a shortage this well telegraphed is already showing up in the earnings revisions.
He also walked through his own firm's scoring. Micron carries a Money Flows overall score of 79, split into technical and fundamental components.
"You may think 79 is like a C plus. It doesn't work that way. It works more like a RPM meter on a car. You don't want it over revving." The firm's sweet spot is 70 to 85, which he called about the best a stock can get.
The flow chart behind it shows unusual institutional inflows for the better part of a year, which he says usually precedes a rise in price. A forced liquidation shows at the March lows, after which the stock ran from $321 a share to a peak of $1,200.
The fundamentals he read off: one-year sales growth of 49%, three-year sales growth of 20%, estimated earnings growth of 112%, three-year estimated earnings growth of 409% and a profit margin of 22.8%, on very little debt.
6. Charge Whatever They Want
Bennett put a line from a different analyst on the show weeks earlier to him — that with this demand and this shortage, Micron can essentially charge whatever it wants.
Bodner agreed without qualification. "Well, I think you just summed it up. They can charge whatever they want." Supply is spoken for years out, and the company is spending to build more capacity.
The reason the shortage cannot be fixed quickly is physical: "You don't put up chip manufacturing plants overnight."
"So, this bottleneck isn't going anywhere."
His demand argument is that semiconductor content keeps rising everywhere. He said he cannot think of anything in his life other than his dog that does not have a chip in it or will not — the refrigerator, the washing machine — even if those do not need high bandwidth memory.
The figure he gave for cars is over 2,000 chips on average, at last count.
He pointed at the Stargate initiative, announced at $500 billion and mocked at the time, as evidence of how far the build has run since.
7. Nvidia's $2B in Coherent
Before the second name, Bodner explained how his firm categorizes the market: for the past ten decades investing has been organized by sector, and Money Flows instead looks at what he calls frontiers — AI, photonics, robotics, space.
Coherent is his photonics name, and he placed it at the center of that world. It supplies lasers, optical components, transceivers that transmit and receive the data signal, advanced materials packaging and other photonics technology.
What the shift actually does: data moves as pulses of light rather than electrical signals over copper, which raises speed and reduces heat, and Coherent gives broad exposure across the optical ecosystem rather than to one product.
The validation he leans on is Nvidia's own money. It is investing $2 billion directly into Coherent and has made a multi-billion-dollar commitment to purchase technology from the company.
The earnings picture he showed runs out to 2029. Full-year 2027 estimates have already been revised up 50%, from roughly $6 a share to $9.43; 2028 is $14 a share; 2029, which he called murkier, is $18.
His read of the setup is a split between the two halves of his own process: fundamentals very strong, technicals a little weak because of the pullback in the space, with large inflows all year. The red on the chart is the Situational Awareness hedge fund blow-up forcing shares out.
8. Close to the Bottom
Bennett pressed on the timing: Coherent has pulled back nearly 30% in three months, the fears are still there, and on his own chart three more weak weeks remain. She asked whether this is a good time to buy.
His answer is that the damage in AI names has largely already happened. The space went through a big pullback and compression while energy, health care and gold rallied, which he described as a market rotation.
A broad market selloff would still hurt, but he argued these stocks have already taken the brunt of their selling.
He came back to the diagnosis: this is not an execution, earnings or revenue problem at the companies, it is a sentiment problem, with the signals pointing to tailwinds several years out.
So his answer was yes. "It's hard to catch a falling knife as we all know, but I think these valuations are getting more compelling by the day."
9. Tower's Doubling Earnings
The third name is the one he expects viewers not to know. Bennett confirmed it had barely appeared on the show before.
Tower Semiconductor is a specialized semiconductor foundry that manufactures wafers and semiconductors — and he stopped to define a semiconductor as a material that regulates how much electricity passes through a chip.
The positioning argument is layered against Coherent. Where Coherent gives direct exposure to photonic and optical products, Tower gives exposure to the manufacturing layer underneath the transition, making chips that other companies design.
What that buys an investor is the ability to avoid stock selection inside photonics: the silicon wafers are part of every photonics package, so there is no need to pick the winning transceiver or networking supplier.
The earnings he read off: $3.86 a share by the end of this year, doubling to $6.64, and nearly doubling again the year after.
The pattern in his own data is the same shape as the other two: weak technicals, strong fundamentals, and a run of institutional inflows his firm calls the stairway to heaven, interrupted by the same end-of-July hedge fund liquidation. Earnings surprises show all green.
Asked whether there is runway left after a more-than-200% run this year, he argued the supply constraint is structural. Only so many companies turn pure silicon into tubes, slice them into wafers and print chips on them.
"It's clear as day this is not a demand issue. It's a supply issue." He treats Jensen Huang's own roadmap — grow on copper as far as possible, then move to silicon photonics — as the vote of confidence, and the technology as cutting edge in its infancy.
10. No Way Back From AI
Bennett closed on the risk she had flagged at the start: the worst case in which government stops the growth of AI, demand falls away and the stocks go to zero, or the more moderate version in which regulation triggers a severe pullback.
Bodner rejected both. "We're at a case here where you can't put the genie back in the bottle." AI is on phones, desktops and televisions, and adoption is too far along to reverse.
His anecdote for where the capital is: a private equity partner told him over dinner, "We don't even look at businesses if there isn't AI involved in them."
His comparisons are to earlier technology panics — the internet, Y2K, and the industrial revolution's fear of replacing manual laborers. He put the pattern down to humans worrying about the unknown.
On the second half of the question, he said the warnings are being misread. "They want to slow down how fast these models update and they want to ensure safety of the models." Growth in AI use is not what anyone proposed slowing.
His conclusion is that safety regulation increases demand rather than capping it — a regulated, safer product gets adopted more widely and is accretive to people's lives.
The reductio he offered: if the argument is that people stop using AI, the same argument says they stop using computers and software.
Bennett's closing frame was the aviation one again — regulation is what lets people board the plane knowing the aircraft will not collide in the sky.
Bonus Insights
Bodner's framing of how sentiment moves was that it is like a stormy day when the wind could blow from any direction, which is how he explained a market that was celebrating Nvidia's guidance two and a half weeks before selling on safety warnings.
His firm's "frontiers" idea is a deliberate break from sector investing — ten decades of tech stocks and materials stocks, against a world he wants to sort into AI, photonics, robotics and space.
Bennett's own reflection on the airline comparison was about the first passengers, who had to get on the aircraft without knowing it would work rather than crash.
Bodner's bottom line is that the AI safety debate is a distraction from a hardware shortage that is already priced into nobody's expectations: memory and silicon photonics are the binding constraints, Nvidia has said so publicly and backed it with its own balance sheet, and the September weakness is a seasonal sentiment problem sitting on top of 52% quarterly earnings growth.
Products, Companies & Tools Mentioned
Micron (His first pick: the number two high bandwidth memory maker, with forecast earnings of $73.23, $155 and $169 a share across 2026 to 2028 and a forward multiple of 5.9)
Coherent (His photonics pick, supplying lasers, optical components and transceivers; Nvidia is investing $2 billion into it and has a multi-billion-dollar purchase commitment)
Tower Semiconductor (His third pick: the specialized foundry making the silicon wafers underneath photonics, with earnings forecast to go from $3.86 to $6.64 a share and nearly double again)
Nvidia (Guided to 70% revenue growth against a 36% consensus, is approaching $100 billion a quarter, and named memory and silicon photonics as its two bottlenecks)
SK Hynix and Samsung (The other two large memory makers; SK Hynix's chief executive sees a bottleneck to 2030, Samsung shortages through 2027 into 2028)
Anthropic and OpenAI (The two companies whose chief executives called for a slowdown; he reads both as firms with initial public offerings ahead of them and large press operations)
Money Flows (His own research firm's software, which scores stocks from 0 to 100 and flags unusual institutional inflows; Micron scores 79 against a 70 to 85 sweet spot)
Stargate (Announced at $500 billion and mocked at the time; he uses it as the measure of how far the build-out has run)
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: