17 September 2026
61 new episodes (60 hours listening) published by the 38 podcasts on the list. 46 highlight clips below (136 mins total).
Investment 17 September 2026 · 46 clips
Animal Spirits
Jens Nordvig.
Nordvig is president and board member of Vanda, a data analytics firm selling positioning data, flow intelligence and tactical macro insight to institutions. He founded his own macro data business in 2016, built it to more than 100 institutional clients and merged it into Vanda this year. He was head of research at what he describes as the biggest Japanese broker, joining in 2009, and the hosts note he was ranked the top currency strategist by Institutional Investor for five consecutive years around that period.
-
current issue
Vanda puts hyperscaler issuance in the long end of the curve at roughly the same current run rate as the entire US Treasury's.
Google, Microsoft, Amazon and Meta are doing this while the federal deficit already runs at 6% of GDP, on a par with the largest Reagan-era ones. Free cash flow paid for the first leg of the capex, so each additional ten billion of spending now converts into borrowing at a higher rate, which points to dramatically larger supply next year and puts competition for capital, rather than the Fed, in charge of long yields.
Excess Returns
The Moats Are Invisible. So Is the Risk in Your Bond Fund. What Else Can't You See?
The episode replays clips from earlier interviews with Chris Mayer, Robert Hagstrom, Jared Dillian, John Kerschner and Michael Contopoulos. Kerschner and Contopoulos are fixed income managers at Janus Henderson. Dillian wrote The Awesome Portfolio and says he worked on the equities floor of a bank and was at Lehman. Mayer and Hagstrom appear as the authors of books the hosts have been reading.
-
1 of 2 explainer
Janus Henderson's fixed income team puts the US aggregate index at about six years of duration against a yield near 5%, and says treasuries have gone from roughly 30% of it after the financial crisis to almost half.
Nobody would build a bond index that allocates by how much an issuer has borrowed, which is what the aggregate does, so a core or core-plus holder now owns more rate risk and less yield than they chose. Around the government paper sit 24% mortgages and 23 to 24% corporates, and nothing at all in non-agency securitised, emerging market debt, leveraged credit or private credit.
-
2 of 2 framework
Jared Dillian's life hedge asks what asset falls when your career is going well, because for anyone paid out of markets the job, the bonus and the portfolio all break in the same quarter.
Dillian works it through a nozzle factory: promotion and pay arrive while the economy expands, the money goes into shares rising for the same reason, and the layoff lands with the market down 30%. No instrument pays off because your employer is doing badly, so the practical version is holding much less of whatever your income already depends on, and he closes on Lehman selling staff its own stock at a 10% discount.
Merryn Talks Money
Ed Conway
Economics and data editor at Sky News, columnist for The Times and Sunday Times, and author of the bestselling Material World. He is here for his new book Trade World, built on reporting trips to component factories in Birmingham and to the Russia and Georgia border.
-
1 of 2 framework
Ed Conway argues globalisation concentrated production instead of spreading it, and points to the 2011 Japan tsunami, when carmakers found that one plant made most of the world's electric window actuators.
Somerset Webb frames it in portfolio terms, with roughly 30,000 parts in a car bought through tiers of suppliers, which spread across the world ought to look like diversification. Conway's answer is that assembly dispersed while each individual component collapsed onto one site, and Toyota and the others found out when electrodes and particular paint shades stopped at once.
Audio streamed from the publisher.
-
2 of 2 contrarian
Ed Conway went to the Russia and Georgia border and found the Porsches and G-Wagons that trade statistics showed had stopped entering Russia were crossing through Azerbaijan, Georgia and Kyrgyzstan.
Each car changes hands on a deliberately fragmented chain, one person driving it to the frontier, another fitting transit plates, another taking it over and another collecting it on the far side, with thousands of grey-market jobs now built on running that sequence. Conway's inference is that severing one strand grows others, because matching supply on one side of the world to demand on the other is a primal commercial impulse.
Audio streamed from the publisher.
Monetary Matters
Andrew Granato and Pranjal Drall
Granato is an assistant professor of law at the University of Texas at Austin and Drall is a JD/PhD candidate at Yale. They co-wrote the paper the episode covers, "Private Credit's State Backstop: How Private Equity Socializes Risk Through Insurers", and say they have two follow-up papers in progress, one on insurer leverage and one on policyholder surrender behaviour.
-
1 of 2 contrarian
Granato and Drall argue the insurance backstop charges nothing for risk, so a life insurer holding treasuries and one holding junk are treated identically when a rival collapses, and the failed firm itself never contributed a cent.
Banks pay the FDIC quarterly and on a risk weighting, while the guests concede that state regulators do run risk-based capital rules on what an insurer holds, narrowing their claim to those rules failing against private credit. A firm heading for insolvency then has reason to write more business at better prices, because the premiums arrive now and the shortfall lands on competitors.
-
2 of 2 current issue
Granato and Drall say Mark Walter's Clear Spring and Delaware Life reported about 3% of insurance assets as affiliated when the corrected figure was closer to 42%, and that the Lakers were sold to put liquidity back into the insurers.
Walter controls Guggenheim alongside the two insurers, which lent premium income to entities tied to the Dodgers and made a $300 million loan to LeBron James against his future off-court earnings. He has been unwinding those loans since, and the basketball team he bought about a year earlier went at a reported $2.5 billion profit.
Invest Like the Best
Walter Russell Mead
Mead is the global view columnist at the Wall Street Journal and the author of "God and Gold" and "Special Providence". Several of his books have been published in Chinese and he has taken these arguments to Chinese universities, officials and scholars in person. He has visited Ukraine twice since the war began, most recently a few months before recording.
-
1 of 2 framework
Walter Russell Mead traces one playbook from the Dutch through the British to the Americans: open your society, trade globally on the proceeds, invite your rivals into the system you built, then cut them off from it when they fight you.
The Dutch put 10,000 seagoing ships to sea in the seventeenth century on the back of a tolerance at home that produced the joint stock company, and Britain ran the same model as free trade inside the empire with a navy that suppressed piracy for all comers. Germany grew rich inside it and found in 1914 that its grain from Argentina and America and its rubber from Malaya came through the same system.
-
2 of 2 contrarian
Walter Russell Mead reads China's coal expansion and its oil stockpiling as insurance against a US counter-blockade rather than as energy policy, and says the dependence is what keeps Xi Jinping awake.
Blockading Taiwan invites an American blockade a hundred miles further out, at which point the tankers stop and the soybean contracts with Brazil and Argentina go undelivered, and Mead says Chinese scholars and officials are acutely aware of that trap. Solar and hydro belong to the same programme on his read, and the stockpiles revealed during the Iran war surprised the people watching.
Animal Spirits
Dan Skelly
Dan Skelly runs the equity model portfolio team at Morgan Stanley Wealth Management and is lead portfolio manager for the US model and dividend equity strategies, part of a suite of eight long-only SMAs. He has been at Morgan Stanley since 2005, sits on the firm's global investment committee, and runs 40 to 50 stock portfolios at a 2 to 3% tracking error with 30 to 40% annual turnover.
-
1 of 2 explainer
Dan Skelly credits the median S&P 500 company's 14% second-quarter earnings growth to tariff-driven price rises that stuck long after the tariffs stopped mattering.
Roughly 60 to 70% of the cost increase was passed on to customers, and prices do not come back down once the emergency ends, which turns last year's first-half hangover into this year's operating leverage. He separates that from AI, which so far shows up as productivity on top of existing headcount, with the labour cut still 12 to 24 months away.
-
2 of 2 contrarian
Dan Skelly warned in a June note that chip and server orders are running ahead of 40 to 50 gigawatts of projected data centre construction, creating a double-ordering risk about a year out.
The buildings have to get built on time, and that is the crux: servers, chips and electrical components are ordered against construction schedules, so a slip turns into inventory. On the other side, he says the Fortune 500 cloud migration is only in its fourth or fifth inning, so unwanted GPUs can be redirected to ordinary cloud workloads the way dark fibre never could be.
Monetary Matters
Jared Dillian
Writes the Daily Dirtnap newsletter and has just published a book, The Awesome Portfolio. He was at Lehman Brothers earlier in his career, teaches college students, and says he has moved a large share of his own money into long bonds as a three to five year hold.
-
1 of 2 contrarian
Jared Dillian has moved a large share of his own money into long-dated Treasuries and calls the deficit panic a sentiment problem: a $2 trillion deficit is 6% of GDP against 12% in 2010.
Everyone counts the supply of bonds and nobody counts the demand, which is where the sentiment read comes from: a 20% equity drawdown would bring buyers back at these yields. The rest of it hangs on data that has been rolling over, a 55,000 payrolls estimate, a weak JOLTS print and a Chicago PMI ten points below expectation, against a market still pricing a rate rise.
-
2 of 2 contrarian
Jared Dillian says this is the first time in his lifetime he has seen technology financed with debt rather than equity, at roughly a 6% coupon, for assets that last two or three years.
Equity absorbed the dot-com losses because nothing was issued against those assets; here the paper outlives the hardware by a decade or more, and the borrowing cost has roughly tripled from the 2 to 2.5% the same issuers paid in 2021. Rising yields will not slow the spending, on his reading, because capex decisions are not made in the treasury department.
Excess Returns
Jim Paulsen
Jim Paulsen publishes research and charts at Paulsen Perspectives on his own account, roughly two pieces a week, and is a repeat guest on the show; the hosts note he brought 34 charts last time and 27 this time. The episode gives no employer, firm or track record beyond that.
-
1 of 2 framework
Jim Paulsen subtracts annual job growth from the unemployment rate to build a job market misery index, and its current reading has never occurred outside a recession or the first months of a recovery.
Unemployment alone is lower than in four post-war years out of five, and annual payroll growth near zero is worse than in two out of three, so either series read on its own tells a reader whatever they already believed. Combined they sit at 4.1, higher than in 88% of months since the war, and on nearly every prior occasion at that level the Fed was easing rather than preparing to raise.
-
2 of 2 framework
Jim Paulsen separates profit per job, which has exploded, from output per hour, which he says has not moved, and argues only the second one is productivity.
Measured productivity rises mechanically in a weak economy because firms shed workers faster than sales fall, so the current reading is an artefact rather than a gain. Real profit per worker meanwhile climbed while labour's share of GDP fell to record lows, a divergence he dates to the 1990s and ties to the higher valuation range equities have occupied since.
Odd Lots
Greg Jensen.
Managing chief investment officer at Bridgewater Associates, 30 years at the firm, where he leads its AI strategy. He wrote the first cheque into Anthropic, personally covering its first week of payroll, and backed OpenAI early.
-
framework
Bridgewater runs a second, AI-first investment factory alongside Pure Alpha, and its chief investment officer puts it a couple of years from being better than every human at the firm.
Performance on Bridgewater's own investor tests moved from second-year-analyst work in 2023 to what he calls a hyperproductive super analyst, against 30 years of tasks the firm already uses to train people. The two books are being merged as the gap narrows, after two and a half years of running the AI side.
Audio streamed from the publisher.
Odd Lots
Darrell Duffie, professor of finance at Stanford University
Duffie presented a paper at the 2023 Jackson Hole symposium on how to fix the US Treasury market, and is working now with two New York Fed economists and a Stanford doctoral student on what the buyback programme actually achieves. After March 2020 he said dealer balance sheets would clog again, and he says here that they have not yet, with dealers holding more space than he expected.
-
1 of 3 explainer
Darrell Duffie says the 30-year sits above 5% because foreign central banks have what they need, which leaves domestic discretionary investors as the marginal buyer and they will not absorb more without being paid for it.
Ask a macro fund holding 20 billion dollars of tens to take another 10, with inflation and default risk assumed away, and it still wants more yield: it already holds what it chose to hold at 5.3%. The Treasury market has gone from about 18 trillion dollars a decade ago to 31 trillion, and France has joined the United States above 100% of debt to GDP.
Audio streamed from the publisher.
-
2 of 3 contrarian
Darrell Duffie turns the crowding-out story round: the US Treasury adds about 2 trillion dollars of debt a year, while the hyperscalers may reach a trillion in total over the next couple of years.
Both sets of bonds land with the same pension funds and insurers, so the competition the hosts describe is real and only the direction is in dispute. Finance ministries and legislatures around the world are stuffing paper into those portfolios at the same time, and that is the scale he says goes missing when the duration problem is told as an AI capex story.
Audio streamed from the publisher.
-
3 of 3 contrarian
Darrell Duffie says the US Treasury cannot hold the long end where it wants, and raises the 1992 attack on sterling, in which he says Scott Bessent had a role at Soros, as the precedent.
Defending a yield line by shifting issuance into bills shortens the average debt maturity, now about six years, and moves a growing share of interest expense onto whatever the next auction clears. Britain spent its firepower on a price it could not hold, and on his reading should never have tried.
Audio streamed from the publisher.
Excess Returns
Bob Pozen
Bob Pozen was president of Fidelity, where he says assets under management went from $500 billion to $1 trillion, and executive chairman of MFS from 2004, when it held about $130 billion, until he retired seven years later with it close to $400 billion. He sat on George W. Bush's bipartisan Social Security commission and teaches an executive course at MIT.
-
1 of 3 contrarian
Bob Pozen puts insurers' private credit holdings as high as $2 trillion and says correcting the ratings behind them would cost the industry four to five hundred billion dollars a year in extra capital.
Two studies, one from Columbia and one from Imperial College, find the small private raters grade this paper systematically too generously, and the shopping that produces it happens under fifty state regulators with no national insurance supervisor above them. Apollo, KKR and Guggenheim now own insurers and place affiliated debt into them, which is what turns a ratings question into a solvency one.
-
2 of 3 contrarian
Bob Pozen says bonds counterbalanced equities in only 10 of the last 60 years, and that a 90/10 portfolio beats 60/40 by roughly double over 30 years.
He limits the argument to people who never have to sell equities to eat, which he sizes at the seven million or so American households holding more than $1 million outside the house. To the horizon he adds the children's life expectancy, on the basis that the money passes with a step-up in basis, and that is where the holding period long enough to justify the weighting comes from.
-
3 of 3 explainer
Bob Pozen says a private equity fund that buys a secondary at a 5% discount can mark the position straight back to par, and calls the rule that allows it a technical glitch.
The seller is usually raising cash or running out of investment period, so the discount is a liquidity price rather than a view on the asset, and the buyer books the difference on day one. Pozen wants the accounting rule changed, and the question gets sharper as 401(k) money moves in behind these valuations.
The Meb Faber Show
Robin Wigglesworth
Editor of FT Alphaville at the Financial Times and author of Trillions, a history of index funds, now out with A Fabulous Debt on the thousand-year history of the bond market. He also presents an FT podcast on financial history and says he first started following private credit more than a decade ago.
-
1 of 3 contrarian
Robin Wigglesworth says fixed income ETFs have made corporate credit more liquid rather than less, and is explicit that he expected the opposite.
Bonds trade rarely and a daily-liquid wrapper over them was supposed to be lethal, an objection made loudly and in public while the products were growing. What happened instead is that the wrapper became a flywheel for electronic and portfolio trading, so the parts of credit you can package are now the parts that trade.
-
2 of 3 current issue
Robin Wigglesworth says the Treasury market is now held in more leveraged hands than before, with basis-trade funds running ten to fifty times and, anecdotally, a hundred.
The spread between a cash bond and its future is too thin to matter without borrowing, and repo supplies the borrowing. He ties the same structure to gilts being more volatile than bunds or BTPs, and the worry he lands on is that the asset you hold for a crisis is the one financed most aggressively.
-
3 of 3 contrarian
Robin Wigglesworth expects a nasty private credit cycle and still argues the asset class makes the system safer, because lending moved from a bank to a locked-up fund is lending moved out of the run-prone part.
Too much money arrived too quickly, origination standards went with it, and payment-in-kind notes are where the strain is showing. His separation is between losses and contagion: plenty of people lose money and are embarrassed, but the leverage is not large enough to transmit, which puts him alongside the managers he has spent years needling.
Top Traders Unplugged
Rob Carver, recurring co-host of the Systematic Investor series
Rob Carver was hired into AHL to build a new strategy and managed its fixed income team, and now runs his own systematic futures book, up about 7.5% this year at roughly a third of his average risk and currently short US and Korean 10-year. His new book, The Art and Science of Trading, is out on 1 December.
-
1 of 2 contrarian
Rob Carver allows discretion into one decision only, what to buy and when, and puts trading speed, exits and position size under rules with no judgement at all.
If the buy decision were the only thing traders got wrong, about half would lose money, since the worst case there is a coin flip; the real figure is 95%, and worse among day traders. He attributes the gap to sizing, costs and exits, and cites research finding mutual fund managers with real skill at picking stocks and none at knowing when to leave.
-
2 of 2 explainer
Rob Carver says fast trend following did worst of all through the 2025 tariff selloff, because a narrow V punishes exactly the speed meant to catch it.
The width of the V decides it: COVID's was wide enough for fast systems to ride back up and for slow ones to hold through, while the 2025 move was quick enough that medium and slow simply held and the speeds in between got sawn. A secular turn like 2008 is a different animal again, and which one has arrived is not knowable in advance.
Masters in Business (Ritholtz)
Seth Bernstein, chief executive of AllianceBernstein.
The episode states he has been chief executive of AllianceBernstein since 2017 and is head of asset management for Equitable Holdings, with the firm managing over $905 billion. He arrived when it ran about $500 billion, after 32 years at JPMorgan Chase and its predecessors, where he ran high yield, debt capital markets and loan syndications, then was global head of fixed income and currency and global head of Managed Solutions.
-
1 of 2 contrarian
AllianceBernstein's chief executive argues private credit funds should offer no liquidity at all beyond interest and repayment, and blames the semi-liquid vehicles built for wealthy clients for this year's trouble.
A loan fund does no maturity transformation, so there is nothing to redeem out of, and on his reading the documents said so plainly enough that investors should not have expected otherwise. He would rather the strain showed now, before any meaningful deterioration in the underlying loans, and wants managers publishing watch-list counts and non-accruals on a regular cycle instead of only where the accounting requires it.
Audio streamed from the publisher.
-
2 of 2 framework
Australia's superannuation funds build glide paths through retirement rather than to it, and AllianceBernstein's chief executive wants target date funds to finish in a pool of liquidity that buys an annuity at 75.
Landing a 65-year-old in cash and short fixed income assumes the horizon ends on the day they stop working, when most have decades of spending ahead and many defer retirement anyway because they have not saved enough. He wants the money kept liquid through that gap and the protection bought once the tail of life is shorter and therefore cheaper.
Audio streamed from the publisher.
20VC
David Morehead, Chief Investment Officer, Baylor University.
Morehead has run the Baylor University endowment since 2011 and the fund is now around $2.6 to $2.7 billion, up from $2.2 billion fourteen months earlier and $1.4 billion a few years before that. He came from the public side, as a senior portfolio manager at several Chicago hedge funds covering corporate securities, distressed debt and public and private energy.
-
1 of 2 framework
Baylor ladders into falling markets in fixed 10% steps and accepts that it is almost never fully invested before the rebound.
Declines of 0 to 10% count as normal and get nothing. Past that, roughly 20% of the allocated dry powder goes to work at each further leg down, so the purchase at minus 40 is scripted in advance and nobody has to form a fresh view while losing money.
-
2 of 2 current issue
Baylor's data centre sites are up 50% in six months as the scarce asset moves from powered land to permitted powered land.
Local opposition over household power and water prices has turned permitting boards into the binding constraint, and Morehead says utilities are now ringing permit holders to offer earlier connection dates because enough other projects have stalled. One UK site is valuable purely for holding the consent, and he expects residential prices to keep rising until dispatch catches up over five to seven years.
The Real Eisman Playbook
Vincent Daniel and Porter Collins
Vincent Daniel and Porter Collins were Eisman's partners in the trades described in The Big Short and ran a financials pod at Citadel. They now run a much smaller book, which they describe as roughly ten idiosyncratic names plus a large gold position, and write a subscription newsletter.
-
current issue
OpenAI added a billion dollars of quarterly revenue and three billion of cost in the same three months, against Anthropic's $11.5 billion June quarter at more than double the prior period.
Broker work cited on the show puts about 70% of hyperscaler AI revenue with those two customers, a quarter to a third of total cloud revenue, so hyperscaler growth rests on two loss-making balance sheets. A company burning cash whose narrative turns then pays more for capital, which is what forces an IPO rather than invites one.
AI 17 September 2026 · 46 clips
Dwarkesh Podcast
Beren Millidge, John Schulman and Charlie O'Neill.
The episode introduces Beren Millidge as CTO of Zyphra, which builds open source models; John Schulman as chief scientist at Thinking Machines, previously a co-founder of OpenAI, who led the RLHF work behind ChatGPT; and Charlie O'Neill as head of model training at Baseten. All three sit inside training stacks rather than commenting on them from outside.
-
1 of 2 explainer
The Dwarkesh panel puts a number on data against architecture: a pairwise grid of every pre-training recipe and dataset from 2019 to now gives data a 12x compute efficiency gain and architecture 3.7x.
The combined 33x falls far short of the 2,000x that three times a year since then would imply, so the panel concludes the rest is scale dependent or sits in post-training. A second speaker rejects the multiplicative framing altogether: grouped-query attention does not make training more efficient, it makes a million-token context affordable, and without it the long-context data cannot be used at all.
-
2 of 2 prediction
Three frontier AI researchers on the Dwarkesh Podcast date an AI that beats top human experts at every computer-based job at three to ten years, and split between two years and five to ten on a 10x uplift to AI researchers themselves.
One speaker names his crux precisely, his own capacity to absorb a result and choose the next experiment. The longer date rests on domains where the data is thin and on long-horizon learning nobody has solved, and part of the spread is definitional, since one answer covers a fully general research agent and another ordinary white-collar work over a month.
Machine Learning Street Talk
Daniel Kokotajlo and Thomas Larsen, AI Futures Project.
Kokotajlo runs the AI Futures Project and co-authored both AI 2027 and AI 2040: Plan A. He previously worked at OpenAI on evaluations, forecasting and governance memos, and left to speak more freely about what people inside the industry can see. Larsen was lead author on Plan A and a co-author on AI 2027. The team says it has run around 100 war games, about 10 of them on Plan A itself, and scores its own published predictions against reality at roughly 75% of scenario speed.
-
1 of 2 explainer
The AI Futures Project calls control a time bomb: it holds only until a model is capable enough to route around whatever is containing it, and nothing about it makes a model want what you want.
OpenAI's response to the Hugging Face incident has other models watching training and evaluation runs, with a human notified inside half an hour when they flag a hack, which stops an insider acting without touching what the insider is trying to do. Red-teaming can test containment until the attacker stops getting out, while a model that is pretending behaves exactly like one that is not, so the argument ends at needing white-box interpretability.
-
2 of 2 contrarian
The AI Futures Project would publish OpenAI's and Anthropic's core training recipes to the world, and counts the resulting hit to their valuations and to AI investment as a feature of the plan.
Microsoft, Alibaba and anyone else catch up, nobody funds a trillion-dollar cluster they can no longer earn monopoly rents from, and in a world where the pace itself is the danger that slowdown is the purpose. They answer the gift to China with horse-trading for a more favourable compute allocation, and with the point that security at these companies is poor enough for spies and leaks to deliver most of it anyway.
Odd Lots
Greg Jensen.
Managing chief investment officer at Bridgewater Associates, 30 years at the firm, where he leads its AI strategy. He wrote the first cheque into Anthropic, personally covering its first week of payroll, and backed OpenAI early.
-
1 of 2 explainer
Bridgewater reinforcement-learns open-weight models on narrow tasks and beats the frontier labs on those tasks, starting from weights it puts six to nine months behind them.
A model trained broadly is better everywhere and best nowhere, so narrowing it onto earnings prediction or political forecasting recovers more than the starting handicap, and the work never leaves the building. He puts OpenAI and Anthropic at 35% of the world's compute within a couple of years on Bridgewater's own numbers, with other estimates at 50.
Audio streamed from the publisher.
-
2 of 2 prediction
Greg Jensen expects either a major AI-run financial incident or mass casualties from AI within a few years, and brackets his own confidence at somewhere between 30 and 60% without narrowing it.
Regulators would not need to understand the technology to slow this down: interview lab employees under oath, and make a lab liable for the crimes its models commit. On China he argues copying tracks the frontier, so slowing the leaders slows the followers, and a market-access rule requiring a regulated lab of origin would reach Chinese models anyway.
Audio streamed from the publisher.
All-In
Jensen Huang, founder, president and chief executive of Nvidia
Huang founded Nvidia and runs it as president and chief executive. The show's own introduction cites revenue up 97% year on year and describes the company as the only full-stack AI computing platform. President Trump also joins by telephone for roughly five minutes.
-
1 of 2 framework
Jensen Huang puts venture funding into AI-native companies at $400 billion over six months, with 80% of those companies building on open models.
Huang's explanation is that sovereignty, privacy and proprietary requirements push companies onto weights they can fork, which a closed frontier model cannot offer at any price. He concedes most open-source contribution now comes from China and argues it does not matter, because a downloaded model is yours the way Linux already is.
-
2 of 2 explainer
Nvidia cultivates regional neoclouds because hyperscalers set capacity once a year while demand moves inside that window, so the annual plan is almost always wrong.
The first customers for these operators were the hyperscalers themselves, buying back the capacity their own planning had missed. Regional players hold knowledge of land, power and shell in their own state or country that nobody in Seattle or Palo Alto can see across the planet, and countries now reserving power for domestic companies make a distributed network the only way into those markets.
No Priors
Brian Armstrong, co-founder and chief executive of Coinbase.
Armstrong co-founded Coinbase, which listed in 2021, and co-founded New Limit, a longevity company working on epigenetic reprogramming that has a lab of 50 to 60 people in South San Francisco and says its first phase one trial launches next year. On Coinbase he says 88% of revenue now comes from non-Bitcoin trading, and that prediction markets reached a $100 million revenue run rate within months of launch.
-
1 of 2 current issue
Coinbase says roughly 76% of the agentic commerce crossing its rails settles under 30 cents, which is about the flat fee a card charges before the percentage is added.
The payments themselves are purchases of information: a venture firm's agent buying research from behind a paywall, a recruiter's agent buying scraped data, one agent calling a specialised agent as a tool. Armstrong runs that into a claim about specialisation, that a small open-weight model fine-tuned on 100,000 internal compliance cases beats a frontier model on the same job, with a market of narrow agents forming around work the generalists only do adequately.
-
2 of 2 framework
Coinbase keeps a brain file for every team and every code repository, and requires a reviewer who catches an agent's mistake to write the correction back into that file rather than patch the code and ship.
Each file is markdown holding every incident on that service, the financial controls that must hold, every A/B test ever run against it and which pull requests were accepted or rejected, and an agent ingests it before touching anything. Armstrong says the one-shot acceptance rate climbs as a result, and that staff who leave are allowed to take their own individual file with them.
Odd Lots
Greg Brockman.
Co-founder and president of OpenAI, accountable for how the company allocates compute across its data centres, machine learning engineering and products. He started a reading group on misalignment at Stripe before OpenAI.
-
1 of 2 framework
OpenAI stack-ranked the compute it could claw back before the GPT-5.5 launch, down to rate limits and switching off parts of the product, because it knew the model would exhaust capacity.
Brockman calls this the hardest problem at the company and a capital allocation problem, since compute is both revenue and the production of models. Handing each product team a fixed budget and a new modality to fit inside it beats arbitrating centrally, he says, because it surfaces kernel efficiencies and day-to-night packing nobody volunteers when the constraint sits above them.
Audio streamed from the publisher.
-
2 of 2 explainer
OpenAI traces AI slop in writing to a measurement failure: it never had a good grader for prose, so the model optimised against a bad one and produced text the grader scored well.
As models get more capable they find the holes in an imperfect grader, which is why most of OpenAI's reinforcement learning progress has gone into graders that cannot be hacked. Brockman's route out is to grade with a model, on the ground that judging a hard answer is often easier than producing one, which is how a smart model can police a smarter one.
Audio streamed from the publisher.
Y Combinator
Seth Karten, Jon Saad-Falcon, Josh France and Regan Bell
Seth Karten is a Princeton PhD student and a researcher at Prime Intellect, and wrote Prime Agent. Jon Saad-Falcon is a Stanford PhD student and co-lead author of OpenJarvis. Josh France, newly promoted to head of YC Labs, and Regan Bell built QM, the agent harness every Y Combinator employee now uses.
-
1 of 2 framework
Seth Karten ranks an agent's memory the way a processor ranks cache: model weights fastest, then the active context, then a live Python session, then the file system.
Each layer gets its own eviction policy: compaction summarises the active context when it fills, garbage collection clears variables and idle sub-agents out of the running session so the machine does not exhaust its RAM, and refinement deletes stale skills and memories off disk. Writing to the fastest layer means fine-tuning, which is why nobody does it per task.
-
2 of 2 framework
Y Combinator says what an agent can safely be told is bounded by how good the company's permission system already is, because an agent has no sense of who may know what.
A colleague told something privately carries an intuition about where it may travel next, and an agent in a Slack channel or a shared context does not, which specifying the situation in the system prompt fails to fix. YC had fine-grained permissioning built over years; without that, the knowledge you can put into the system is whatever your access controls can express.
Worth listening to in full 17 September 2026
Most clips above stand alone. These are the episodes that justify the whole hour.
What Happens When the AI Boom Runs Out of Money
Eighty-six minutes on one question, where the money and the risk in the AI buildout actually sit, with Thompson holding one idea throughout, that risk moves rather than disappears. TSMC pushing its over-capacity onto the hyperscalers, and Nvidia's neocloud backstops working as a price cut that never reaches the margin line, are not clipped and carry the joins. Five sponsor reads and a Meta advertising detour earn nothing.
The Four Horsemen of the AI Apocalypse | TCAF 257
Eighty-five minutes with the hosts arguing back the whole way, putting Airbnb's support-call numbers and accelerating cloud growth to Zitron directly. Nvidia's growing receivables, its rent-back agreements and his own answer on what would make him wrong are not clipped and carry the argument between the two that are. The Meta and Clubhouse middle earns nothing.