The 12 Best Invest Like The Best Episodes, Summarized

The best Invest Like The Best podcast episodes from the past year, ranked and summarized: key ideas, frameworks, and quotes from Shyam Sankar, Brian Chesky, John Arnold, Gavin Baker, Mitchell Green, and more. Every summary is free.

Patrick O'Shaughnessy has recorded hundreds of Invest Like The Best conversations, so the honest answer to "where do I start" depends on what you want out of the show. This guide ranks the twelve strongest episodes in our summary library: the conversations where an investor or an operator hands over a repeatable way of thinking, with enough specifics to use on Monday.

We went back through every Invest Like The Best conversation we have summarized, mostly from the past year, and scored them on how much durable thinking survives the hour. What earned a spot: Palantir's CTO on how talent actually compounds, Airbnb's founder on the gap between founding a company and running one, Anthropic's CFO on budgeting inside an exponential, and the energy trader who built the best seat in his market before he built his track record.

Each entry explains why the episode earns its rank, the takeaways worth acting on, and a quote that captures the conversation. Every one links to our full free summary, and most include the official YouTube upload so you can listen in place. Three ideas recur across the list, and we pull those threads apart after the ranking.

1. Why Great Builders are Heretics | Palantir CTO

Patrick O'Shaughnessy with Shyam Sankar · 1h 29m · March 2026

The single most useful episode on this list for anyone who manages people or manages themselves. Sankar supplies three frameworks in ninety minutes: heretics as the actual engine of institutional change (Billy Mitchell inventing air power and dying penniless, Rickover building the nuclear navy out of a converted restroom), gamma-ray talent development where growth tracks your tolerance for pain, and the superpower-versus-kryptonite audit that tells you which weaknesses to route around permanently. Most leadership content gives you vocabulary. This gives you a hiring test and a self-diagnosis.

Key takeaways

  • Your maximum rate of learning sits exactly at your maximum tolerance for discomfort, so structured career ladders produce the least growth.
  • Your real superpower is what feels effortless to you and impossible to smart people around you. Effort is a poor signal of excellence.
  • Kryptonite weaknesses run six standard deviations below the mean. Route around them permanently and build a culture where people can admit them safely.
  • The hiring question Sankar uses: give this person an inch and can they turn it into a mile? The worst bet is the person who turns a mile into an inch.
  • Institutional innovation comes from obsessives who fight the bureaucracy. Systems designed to produce innovation reliably deliver very little of it.
Your maximum rate of learning will be coincident with your maximum ability to tolerate pain. — Shyam Sankar

Read the full episode summary

2. How Brian Chesky Is Redesigning Airbnb for the AI Era

Patrick O'Shaughnessy with Brian Chesky · 1h 23m · May 2026

Chesky has given a lot of interviews, and this is the one where he is most specific about the mechanics of the job. He separates founding from running a company (founders learn by doing, which is exactly the habit that wrecks a CEO), then names the delegation sequence most leaders get backwards: start hands-on, audit the work, teach everything you know, and give ground grudgingly as the muscle memory forms. The Hawaii system is the operational half: Airbnb spent sixteen years failing to launch a second business because it kept starting global, and the fix was one city, a team of ten to twelve, then ten cities, then industrialize.

Key takeaways

  • Founders are largely innate; CEOs are made. Almost every CEO instinct is counterintuitive, and trial-and-error learning is expensive once an empire-builder is already hired.
  • Delegate in the correct order: work intensely with a new team first, audit their output, then release control gradually as they develop judgment.
  • The Hawaii system: shrink a new business to one city and a team of ten to twelve, perfect it, expand to ten cities, then industrialize.
  • Pure people managers are the most exposed role in the AI era. Everyone becomes a hybrid who manages through the work, and layers of management collapse.
  • Industrial design is unusual training for product leadership because a design succeeds only when it sells, which forces empathy and commerce into the same decision.
No one is born a good CEO. I think people are basically born good founders or said differently it's innate. You don't have to learn how to be a good founder. The job of CEO is completely counterintuitive and almost all of your intuition about what to do is wrong. — Brian Chesky

Read the full episode summary

3. How Claude went from $9 billion to $45 billion in one year | CFO explains

Patrick O'Shaughnessy with Anthropic CFO Krishna Rao · 1h 22m · May 2026

The rare finance conversation that changes how you plan, whatever business you run. Rao's job is to allocate capital inside a company whose revenue went from $9 billion to north of $30 billion run rate in four months, and his answer is a working method: plan in scenarios across a cone of uncertainty, keep the bar for updating assumptions deliberately low, and revisit forecasts continuously. The compute discussion is the concrete version of the same tension. Overbuy and you are bankrupt; underbuy and you fall off the frontier, so Anthropic runs three chip platforms fungibly to keep the option open.

Key takeaways

  • Exponential environments break linear planning. Build scenarios across a cone of uncertainty, work backwards from each, and update assumptions on a low bar.
  • Model intelligence is multi-dimensional: long-horizon task capability, tool use, speed, and agentic performance. A model that finishes in a day beats one that takes a week by roughly seven times.
  • Compute flexibility is the strategic asset. Anthropic allocates dynamically across Amazon Trainium, Google TPUs, and Nvidia GPUs, and between training, internal use, and customer serving.
  • Enterprise returns to frontier intelligence stay extremely high because each model generation unlocks new TAM and previously impossible products.
  • Recursive self-improvement is already operational: over 90% of Anthropic's code is written by Claude Code, including much of Claude Code.
If you buy too much compute, you go out of business. If you buy too little compute, you can't serve your customers and you're not at the frontier. — Krishna Rao, CFO of Anthropic

Read the full episode summary

4. Gavin Baker on Orbital Compute, TSMC, and Frontier Models

Patrick O'Shaughnessy with Gavin Baker · 1h 22m · May 2026

Baker makes the best available argument about why the AI buildout has held together, and it has one name: Taiwan Semiconductor. His claim is that TSMC's deliberate restraint on wafer capacity is the brake on a bubble, because if it produced everything Nvidia could sell, the market would absorb $2 to $3 trillion of GPUs annually and overbuild on debt. The orbital compute segment is the speculative bookend, and Baker is precise about it: racks of GPUs linked by lasers with solar arrays and radiators, from the company already operating the overwhelming majority of satellites. He also names his own worry, a diversity breakdown where everyone turns bullish at once.

Key takeaways

  • TSMC's supply discipline forces efficient capital allocation across the AI buildout and is the main structural check on an infrastructure bubble.
  • Economic value at the model layer concentrates at the frontier. Models that felt remarkable months ago become intolerable as the frontier moves.
  • Anthropic added $11 billion of ARR in a single month, roughly what Palantir, Snowflake, and Databricks built collectively over a decade.
  • Orbital compute is GPU racks in space connected by lasers, with solar for power and radiators for cooling, built on capabilities Starlink already proved.
  • Distinguish the two kinds of drawdown: a broken thesis you crystallize, and price action you disagree with in companies you know deeply, where you can lean in.
Anthropic added their combined businesses in one month. That's just nothing like that has ever happened in the history of capitalism. Forget my career. Just the flatout history of capitalism, the history of business. — Gavin Baker

Read the full episode summary

5. What 100 Years of American Finance Tells Us About Today

Patrick O'Shaughnessy with Alan Waxman · 1h 6m · April 2026

The most useful macro episode of the year because Waxman refuses to argue about symptoms. He organizes a century of American finance into three systems (Glass-Steagall stability from 1933 to 1999, deregulation and leverage through 2008, Basel 3 plus a private capital market that grew from $2 trillion to $15 trillion), then identifies the behavioral change he thinks matters now: the factory model, where firms industrialize both fundraising and deployment. The diagnostic he leaves you with is portable. Crises come from mismatched assets and liabilities plus leverage, so look at the liability structure first and the headlines last.

Key takeaways

  • Read any financial system through three lenses: incentive structures, regulatory guardrails, and market structure.
  • Nearly every crisis shares the same DNA. Leverage plus an asset-liability mismatch forces investors out before the thesis can play out.
  • The factory model industrializes fundraising, which then forces industrialized deployment and looser underwriting to place the capital.
  • Asset manager valuations moved from 10-15x fee-related earnings in 2010 to 25-30x today, which rewards asset gathering over investment quality.
  • Wealth channel capital is the cheapest to raise in good times and the most procyclical in bad ones, because those investors ask for liquidity under stress.
Everything that is covered in the media is just talking about the symptoms and not actually getting to the root cause. And again, when you think about history, like people talk about the symptoms, but when you start to diagnose what happened and how we got there, it had to do with the root cause. — Alan Waxman

Read the full episode summary

6. The World's Greatest Energy Trader on Markets, China, and AI

Patrick O'Shaughnessy with John Arnold · 1h 22m · March 2026

Arnold's explanation for his own record is structural, which makes it transferable. He charged 3% management and 35% performance when the industry charged 2% and 20%, and he spent the difference on the best fundamentals team, proprietary data, and trading systems. That is the whole thesis: build the seat with the best information flow and excellence compounds on its own. The China section is the second reason to listen. His account of factories going from groundbreaking to production in seventeen months, with every supplier inside a 200 mile radius, is the clearest picture of agglomeration advantage in any recent episode.

Key takeaways

  • Cultivate the best seat in your industry: the position with the best information flow, perspective, and systems. Superior economics fund that seat.
  • Being the largest market maker in natural gas produced three edges at once: spread capture, low-slippage anonymous positioning, and visibility into who was doing what.
  • Chinese factories reach first production roughly seventeen months from groundbreaking, powered by supplier proximity, flexible skilled labor, and state support.
  • Total immersion built the edge, and Arnold is candid that it cost him in relationships and health.
  • Energy policy balances five competing goals (affordability, reliability, emissions, security, jobs) that shift every four to eight years while infrastructure takes decades.
I wanted to cultivate and build the best seat in my industry, the seat with the best perspective, with the most information, with the best systems. — John Arnold

Read the full episode summary

7. How to bet on yourself (without venture capital)

Patrick O'Shaughnessy with William Hockey · 1h 17m · March 2026

The sharpest counterargument to default startup financing on this list, from someone with the receipts. Hockey built Column, a software company that owns a regulated bank, while owning 100% of it and growing out of earnings. His point is about time horizon: full ownership let him buy a bank and invest in it for two to three years before taking a single client, which a company raising every eighteen months could rarely justify. He is also unusually clear-eyed about geography, arguing that San Francisco's consensus is simultaneously an operating advantage and a blind spot, which is why he goes to Kinshasa to find what the consensus missed.

Key takeaways

  • Constraint drives financial innovation. Africa built mobile payments long before Venmo existed because the infrastructure forced a leapfrog.
  • Venture funding creates a dependency loop: each raise optimizes for the next raise, so strategy drifts toward whatever is fashionable that year.
  • Early-stage founders typically surrender 50-75% of their equity value to dilution alone, plus 10-80% more to preference stacks, and most learn this years too late.
  • Self-funding allows decade-long bets. Column bought a regulated bank and invested in it for two to three years before onboarding clients.
  • Emerging markets hold differentiated talent. In places with no local frontier lab competing for engineers, world-class people are working at banks and breweries.
VC money is kind of like heroin. It like feels good. It's amazing, but like you got to keep shooting up. Like it's very challenging to get off. — William Hockey

Read the full episode summary

8. Clay’s CEO Thinks Differently About Everything

Patrick O'Shaughnessy with Kareem Amin · 1h 0m · June 2026

The most philosophically interesting hour of the year, and the one that will annoy people who prefer tactics. Amin's central claim is that capitalism pays for risk above hard work and above skill, where real risk means genuine uncertainty plus exposure to shame. The corollary is the part worth sitting with: he argues you take bigger risks from wholeness, because a founder operating from lack builds to fill a void and does collateral damage on the way. His meditation retreat story, where he recognized constant future-orientation as a quiet death drive, is the kind of thing almost no operator says on a business podcast.

Key takeaways

  • Capitalism compensates risk more than effort or skill. Real risk requires genuine uncertainty about the outcome and exposure to shame in failure.
  • Creating from wholeness produces more courage and less collateral damage than creating from a chip on the shoulder.
  • Self-respect is the durable metric. External validation from funding rounds and revenue milestones fades quickly, and your own judgment knows everything you did.
  • Arrangements where one side dominates another stay unstable however powerful they look, so fairness is a precondition for long-term progress.
  • Constant future-thinking is a form of escape. The only certainty ahead is death, which makes present-moment attention the more honest default.
I think capitalism rewards risk more than anything else. It's not meritocracy. There are a lot of people who are working really hard and not making a lot of money. So it's not hard work. It doesn't just reward skill. I do think it rewards risk. — Kareem Amin

Read the full episode summary

9. Why Now is the Best Time to Buy Public Software Companies

Patrick O'Shaughnessy with Mitchell Green · 1h 0m · March 2026

The most process-heavy episode on the list, which is exactly its value. Green gives away Lead Edge's actual screening rule (revenue today must exceed cumulative cash burn to date), explains why 800 executive LPs function as a sourcing and diligence machine, and then talks about the sell side, which almost nobody in growth equity discusses publicly. His argument about software moats is the part worth arguing with: he claims the advantage was always distribution, sales, and client services, so Workday-style switching costs survive AI in a way that R&D advantages would have.

Key takeaways

  • The screen that avoids disasters: revenue today should be greater than all the cash the company has ever burned, a 1:1 ratio or better.
  • 800 executive LPs function as infrastructure: warm introductions when companies go quiet, customer back-channels during diligence, and business development afterward.
  • Software moats come from distribution, sales, and client services. Microsoft could rebuild most niche products in a month with 500 engineers.
  • Sell discipline separates firms. Lead Edge runs a disposition committee once or twice a month and has exited a third of its positions via secondaries.
  • Entry price is survivable when the exit multiple assumption is conservative. The 2020-2021 error was underwriting 20-25x revenue exits as permanent.
If you tell an entrepreneur that you're going to actually do something, then actually do it. And in a I think that's actually true of like life. Um there are so many people that say they'll do they do things that just like never do them. — Mitchell Green

Read the full episode summary

10. Uber CEO on AI, Autonomous Vehicles, and the Future of Transportation

Patrick O'Shaughnessy with Dara Khosrowshahi · 1h 11m · June 2026

Khosrowshahi is the best available case study in taking a job because it is hard, and Daniel Ek's line about impact over happiness is the hinge of the whole conversation. The operating content is equally good: he reframes Uber as a supply company where adding drivers, restaurants, and retailers manufactures latent demand, which points the strategy at sparse markets and smaller cities. His observation that magic normalizes within a day (a five minute wait becomes a grievance at six and a half) is the most quietly useful thing anyone said about autonomous vehicles this year.

Key takeaways

  • Marketplaces like Uber are supply businesses. More drivers, restaurants, and retailers create demand that materializes on its own.
  • Magic normalizes in under a day, so new technology has to compete on safety, efficiency, and affordability once the novelty burns off.
  • Break chaos into component dimensions, set an initiative against each, and solve them independently before recombining.
  • AI adoption appears in unpredictable pockets. Developers in India drove 10x code commits with autonomous agents, so find those internal rebels and promote what they learn.
  • Overcoming difficulty is where satisfaction comes from, which is why solving every problem for your kids removes the thing that forms them.
Since when is life about happiness? It's about impact. — Daniel Ek (to Dara Khosrowshahi)

Read the full episode summary

11. The Two Harvard Dropouts Who raised $800M to take on NVIDIA

Patrick O'Shaughnessy with Gavin Uberti and Rob Munro · 1h 32m · June 2026

A masterclass in finding leverage by auditing the assumptions everyone else inherited. Etched's founders noticed that the semiconductor industry designs every component with general-purpose buffer, including the requirement that chips run at full speed at 0 degrees Celsius, which is meaningless for AI data centers that stay above 80. Removing that buffer compounds: they run at under half the voltage of other AI chips (power scales quadratically with voltage), and they built interconnects with 5x lower latency so eight chips behave like one memory pool. Listen for the method even if you will never design silicon.

Key takeaways

  • General-purpose buffer is hidden cost. Pin down your actual operating constraints and every default configuration becomes negotiable.
  • Voltage scales quadratically with power, so running at under half the voltage of competing AI chips allows far more compute in the same silicon area.
  • Ask about bandwidth across the cluster, since interconnects with 5x lower latency let scaling from one chip to eight deliver close to 8x throughput.
  • Vertical integration buys velocity. Etched builds its own chips and racks in parallel, with thermal test chips validated before silicon returned.
  • Token generation today resembles pre-industrial manufacturing, and whoever mass-produces tokens efficiently makes frontier AI as universal as consumer electronics.
I've never seen an AI data center with ice in it. So, you know, we can feel pretty confident that our chips don't need to run at full speed at 0 degrees CC. In fact, like they're never really going to be running below 80° C anyway. — Gavin Uberti

Read the full episode summary

12. Investing a $120 Billion Balance Sheet with No Outside Investors

Patrick O'Shaughnessy with Liberty Mutual's CIO · 1h 16m · June 2026

The quietest episode here and a useful corrective to forecasting culture. Liberty Mutual runs $120 billion by preparing for eventualities, and the mechanism is exposure-first, product-second: decide what exposures the total portfolio needs, then pick the best route to each one, whether that is an LP commitment, a direct investment, a co-invest, a club deal, or a partnership. Pair it with the Waxman episode and you get both halves of the same argument, since the mutual structure removes the quarterly pressure that drives the factory model everywhere else.

Key takeaways

  • Prepare for eventualities in place of predicting them. Build a portfolio across credit, private equity, real estate, and infrastructure that survives several futures.
  • Exposure-first, product-second: define the exposures you want, then choose the optimal vehicle for each one from a wide toolkit.
  • The mutual structure removes shareholder pressure for quarterly results and buybacks, which enables patient capital and real investment hygiene.
  • Insurance balance sheets do two jobs at once: syndicating risk so others can take it, and investing premiums into growth and infrastructure.
  • Entrepreneurial culture inside a large institution has to be engineered through incentives, curious hiring, and governance, because deal flow dries up once you stop looking.
We're not in the business of predicting the future. We're in the business of being prepared for all its eventualities. — Liberty Mutual CIO

Read the full episode summary

What these episodes have in common

Theme 1: Build the seat, then take the shot

The investors on this list agree about something that rarely makes headlines: returns follow structural position more reliably than they follow insight. John Arnold says it most directly, charging 3% and 35% so he could fund the best fundamentals team, the best proprietary data, and the best systems in natural gas. Mitchell Green's version is 800 executive LPs who open doors, back-channel customers, and run business development after the check clears. The Liberty Mutual CIO gets it from the mutual structure itself, which removes shareholder pressure and permits genuinely long holding periods.

Alan Waxman supplies the negative case and makes the whole theme legible. His factory model is what happens when position gets optimized for asset gathering: industrialized fundraising forces industrialized deployment, underwriting standards slip to place the capital, and the liability structure quietly diverges from the assets. Same variable, opposite sign. Arnold and Green engineered position to make good decisions easier; the factory model engineers position to make volume easier, and the bill arrives later.

The portable version for anyone outside finance: audit the seat before you audit the effort. Who sees the flow first, who has the data, whose capital can wait. Those questions predict outcomes better than a sharper opinion does.

Theme 2: Compute is now a capital allocation problem

Three episodes approach the AI buildout from three sides and converge. Anthropic's CFO Krishna Rao frames it as a budgeting knife edge: overbuy compute and the company dies, underbuy and it falls off the frontier, so the resolution is fungibility across Trainium, TPUs, and GPUs plus dynamic allocation between training and serving. Gavin Baker looks at the same question from the supply side and credits TSMC's wafer discipline with holding the whole system together, on the logic that unlimited supply would let the market absorb $2 to $3 trillion of GPUs a year and overbuild on debt. Etched attacks it from physics, stripping general-purpose buffer out of chip design once you accept that AI data centers stay above 80 degrees Celsius.

The shared conclusion is that constraint is doing the useful work. TSMC's restraint forces efficient allocation across the industry. Anthropic's compute budget forces scenario planning. Etched's narrow use case unlocks performance the general-purpose design could never reach. Every one of these advantages comes from accepting a limit and optimizing hard inside it.

Where they diverge is worth noting. Baker's worry is a diversity breakdown, the moment everyone turns bullish simultaneously and stops pricing risk. Rao is building as though the exponential continues. Both can be correct for a while, which is the uncomfortable part.

Theme 3: Growth comes from removing the safety rails, with one loud disagreement

Shyam Sankar, William Hockey, and Kareem Amin are describing the same mechanism in three vocabularies. Sankar throws people into problems they are unqualified to solve because learning rate tracks pain tolerance, and calls structured career ladders comfortable and inert. Hockey refuses venture capital so he can make ten-year bets, and travels to Kinshasa to get outside San Francisco's consensus. Amin argues that capitalism pays for risk above effort and skill, where risk means genuine uncertainty plus exposure to shame. All three treat discomfort as the input rather than the cost.

Brian Chesky disagrees, and the disagreement is the most useful thing in this guide. His delegation advice is to start hands-on, audit the work, teach everything you know, and give ground grudgingly, because he watched trial-and-error learning at the CEO level produce empires that took years to unwind. Sankar accepts a 50% failure rate as the price of building someone extraordinary; Chesky considers that rate unaffordable when the mistake compounds through an org chart.

The reconciliation sits in the stakes. Sankar's gamma-ray doses land on individuals inside a system that can absorb the failure. Chesky's caution applies to hiring decisions that shape hundreds of people and take years to reverse. Use the deep end on yourself and on contained problems, and audit carefully wherever a bad call compounds.

Every episode referenced

Frequently Asked Questions

What is the best Invest Like The Best episode?

From the show's recent run, the strongest all-around episode is Why Great Builders are Heretics with Palantir CTO Shyam Sankar (March 2026). It delivers three usable frameworks in ninety minutes: heretics as the real engine of institutional change, talent development through deliberate deep-end exposure, and the superpower-versus-kryptonite audit for deciding which weaknesses to route around. For operating craft, Brian Chesky's Airbnb episode is the standout. For markets, John Arnold's conversation on building the best seat in your industry.

Which Invest Like The Best episodes should I start with?

Start from what you need. For managing people and yourself, Shyam Sankar of Palantir. For running a company, Brian Chesky of Airbnb. For planning inside fast growth, Anthropic CFO Krishna Rao. For macro and credit cycles, Alan Waxman on a century of American finance. For investment process, Mitchell Green of Lead Edge. For the AI hardware stack, Gavin Baker and the Etched founders. Every episode in this guide has a full free summary, so you can preview the ideas in a few minutes before committing to the hour.

What is the Invest Like The Best podcast about?

Invest Like The Best is a long-form interview podcast hosted by Patrick O'Shaughnessy, part of the Colossus network, that explores how investors and operators actually think. Guests range from public market investors and private equity leaders to founders, CFOs, and technologists, and conversations dig into capital allocation, competitive advantage, company building, and how technology reshapes industries. Recent episodes have featured Palantir CTO Shyam Sankar, Airbnb CEO Brian Chesky, Uber CEO Dara Khosrowshahi, and energy trader John Arnold.

How long are Invest Like The Best episodes?

Most Invest Like The Best episodes run about one hour to an hour and a half. The twelve episodes in this guide range from 1h 0m (Kareem Amin of Clay, and Mitchell Green of Lead Edge) to 1h 32m (the Etched founders on taking on Nvidia), with the typical episode landing near 1h 20m. Because the conversations are dense with numbers and frameworks, reading a structured summary first is an efficient way to decide which hour deserves your commute.

Where can I find Invest Like The Best episode notes and transcripts?

Official transcripts and show notes live on the Colossus site and in the podcast player of your choice, and full episodes are published on the Invest Like The Best YouTube channel. 1% Better takes a different angle: we publish a free structured summary of each episode in our library, with the single key takeaway, five or more detailed insights, and the quotes worth keeping. Every episode ranked on this page links straight to its summary, and our library covers hundreds of episodes across shows including Acquired, All-In, and a16z.

Keep exploring