Special Guest Bill Ackman on Markets, AI and Concentrated Investing

ELI5/TLDR

Bill Ackman runs a fund that owns just 8 to 12 companies and never sells in a panic, because his investors can’t yank their money out — that “permanent capital” lets him buy great businesses when short-term traders dump them cheap. He thinks the AI infrastructure boom (the chips, data centres, cloud) is mostly sound spending by cash-rich giants, but the companies actually selling AI models — OpenAI, Anthropic — face a brutal price war that could crush their economics. His newest bet copies Warren Buffett directly: use an insurance company’s cash pile to buy stocks and compound it for decades. He also throws in opinions on Trump, tariffs, and Iran along the way.

The Full Story

Why he locks the doors on his own fund

Most money in markets today is impatient. Ackman points at two culprits. First, “pod shops” — firms that split capital across dozens of small teams, each paid on short-term results and forced to dump a stock the moment it dips. Second, index funds, which now own so much of the market that they shrink the pool of shares actually changing hands, so each remaining buyer and seller swings prices harder. The result is more volatility, and periods where genuinely good companies trade at silly discounts.

The problem: if your investors can pull their money overnight, you can’t take advantage of those discounts — you’re forced to sell at the worst moment. So Ackman built his firm around capital that can’t run away. Pershing Square is structured as a closed-end vehicle, meaning investors who want out sell their stake to someone else rather than forcing him to liquidate the underlying companies.

“If you pick a good one, it’s an infinite life entity. And the value of that business is the present value of the cash you can generate from it… to own that business for a month or two, really, you’re going to miss out on significant opportunities.”

He keeps a roster of companies he’d love to own — he calls it a “library” — ready to deploy when prices fall. And he says right now is one of those buying windows.

Concentration: why 8 to 12 and not 50

The logic is almost embarrassingly simple. If you’ve ranked your best ideas, why put money into idea number 20 when you could buy more of idea number 1, 2, or 3? Diversification smooths the ride, but it also drags your returns toward the average.

So why doesn’t everyone concentrate? Because concentration guarantees you’ll sometimes look very different from the crowd — and looking different is how money managers get fired.

“The way to get fired in asset management is to have performance that’s adverse to others… if your money can disappear overnight, then it’s a much more prudent strategy to not hug the index.”

Most managers, he argues, secretly hug the index and only make small bets at the margin, because being down when everyone else is down is survivable, but underperforming a rising market gets you sacked. His permanent-capital structure is what lets him ignore that fear. Over 22 years it’s produced roughly 6 to 7 percentage points of outperformance per year — though with stretches (including the last few months) where he badly trails.

From bullhorn to backroom

Ackman built his name as an activist — buying a chunk of a company and publicly pressuring management to change. He says that style has softened, for two reasons. One, today’s volatile markets throw up well-run, high-quality businesses at cheap prices more often, so he doesn’t need to fix them — he can just buy them and target mid-to-high-20s percent returns. Two, after 20 years and a track record, he no longer has to fight his way into the boardroom.

“Activism is only needed if you can’t get into the boardroom… that’s really changed. And so the result today is we still make investments in companies which there are opportunities for the businesses to be improved, but we can have those conversations behind closed doors as opposed to hosting people in a hotel ballroom.”

Politics, tariffs, and a detour through Iran

Asked about the political climate, Ackman notes a more interventionist White House. When Biden gave Intel a grant to shore up US chipmaking, Trump’s instinct was to demand ownership in return — so “you don’t want to get adverse to the president these days.” He credits Trump on Iran (calling the higher oil prices a short-term political cost worth paying to prevent a nuclear Iran) but was openly critical of the tariff rollout, which he found dangerously abrupt. He claims credit for the “90-day pause” idea that the president then adopted.

The AI bubble question, split in two

Ackman pushes back on the idea that consumers dislike AI — adoption has been the fastest of any product in history, and people are weaving it into health, work, daily life. The real fear is job loss, sharper than with any prior technology, though he thinks the net effect on jobs will be positive.

On the “bubble or not” debate, he draws a sharp line. The hyperscalers — Microsoft, Google, Meta, Amazon — spending $150 billion-plus on data centres are not betting the company; they’re cash-rich, debt-light, and responding to genuine demand in a race where the first to “superintelligence” wins enormously. Crucially, this is growth CapEx, not the maintenance kind that just keeps a business standing still.

“Warren Buffett always said the best business in the world is one that generates high returns on capital and you can reinvest that capital at high rates on new projects. These are clearly examples of that.”

The danger sits elsewhere — in the model makers themselves.

“I worry more about the LLM business model than I do about the people building the infrastructure.”

The reasoning: most users don’t need the absolute best model to find a restaurant or build a spreadsheet. Open-source and free models are good enough for the majority of tasks, and cheap Chinese models are already eating into Silicon Valley. Pricing is shifting from all-you-can-eat to per-token, making corporate buyers far more price-sensitive. He compares the cash-burning model makers to Webvan — the dot-com grocery firm that only worked while it could keep raising money. That’s why OpenAI and Anthropic are now going public: they need the capital. And if a giant like Google decided to slash token prices using its balance-sheet muscle, it could make life genuinely miserable for them. He thinks OpenAI is more exposed than Anthropic, which has leaned into the enterprise market.

Why he sold Google and bought Microsoft

Pershing Square owned Alphabet, made roughly 3.5 to 4 times its money, then exited this quarter — not because it’s a bad business, but because at 31–32 times earnings the capital could work harder elsewhere. He was surprised Alphabet then issued $80 billion of stock it didn’t need, reading it as an opportunistic grab of cheap capital while competitors are forced to raise. He rotated into Microsoft, which he frames as the gateway to AI in the workplace: roughly half a billion users paying about $20 a seat for a bundle that would cost $50-plus piecemeal, with Copilot steering each task to the cheapest model that gets the job done. No sane CIO wants to build their own Word or Excel — so Microsoft becomes the toll booth.

The Buffett move: insurance float

The most concrete new idea is Howard Hughes. It’s an odd duck in his portfolio — a small-cap real-estate company that owns what are effectively small cities, spun out of the General Growth bankruptcy he restructured during the financial crisis (his most successful investment ever). Pershing Square now controls about 47% of it.

The plan is to turn it into a mini-Berkshire. Through a specialty insurer called Vantage, he’ll take the insurance “float” — the premium cash an insurer holds before paying out claims — and park it in short-term treasuries (no risk), while investing the surplus equity into stocks using the Pershing Square playbook.

“Mr. Buffett took 100% of the float generated by insurance, invested in treasuries, and took the surplus or the equity invested in common stocks. And he was a good stock picker. That drove the value of Berkshire.”

Why doesn’t every insurer do this? Because, Ackman says bluntly, great investors don’t take jobs at insurance companies — the pay and prestige aren’t there. Buffett’s edge was being a brilliant investor who happened to own an insurer and could manage its money himself for free. Pershing Square will manage Vantage’s portfolio for free too, and since they own so much of the company, the incentives line up. Compound that float at a high rate over decades — Buffett did 20% a year for 60 years — and you have, in Ackman’s words, one of the great investment stories of all time.

Key Takeaways

Permanent capital is the moat behind the strategy. A closed-end structure means investors exit by selling their stake to someone else, never by forcing the manager to liquidate holdings — which is what lets him buy when others are forced to sell.
Index funds increase volatility. By locking shares out of the tradeable float, they shrink the active market, so the marginal buyer/seller moves prices more — which paradoxically creates the mispricings an active investor needs.
Concentration’s logic: with ranked ideas, adding more positions just dilutes you toward your weaker bets. The only reason to diversify is career safety, not returns.
“The way to get fired in asset management is to have performance that’s adverse to others” — so most managers closet-index and only bet at the margin.
Growth CapEx vs maintenance CapEx is the key distinction in the AI debate. Hyperscaler data-centre spending earns returns far above cost of capital; it’s not the value-destroying upgrade-or-die kind. That’s why Ackman doesn’t see a bubble in the infrastructure layer.
The vulnerable layer is the LLM business model, not the infrastructure. Most tasks don’t need the frontier model; open-source/free/cheap-Chinese models are “good enough,” pricing is moving to per-token, and a giant could weaponise its balance sheet to slash token prices.
Insurance “float” is the cash an insurer holds between collecting premiums and paying claims — it can be invested. Buffett’s trick was holding the float in treasuries and the surplus equity in stocks, then compounding it for 60 years while issuing almost no new shares.
The hidden Buffett advantage: a great investor who owns the insurer manages the float for free, instead of paying a hedge fund — a structural cost edge competitors can’t easily copy because top investors don’t work at insurers.
Valuation discipline in action: sold Alphabet at ~31–32x earnings after a 3.5–4x gain, rotated into Microsoft — exit driven by opportunity cost, not a bad-business call.
The Webvan analogy: a product can have real demand and still fail if its economics only survive while capital keeps flowing — relevant to cash-burning AI model makers.

Claude’s Take

This is a clean, jargon-light hour of a sharp investor explaining how he actually thinks, and the interviewers (both Hargreaves Lansdown analysts) ask real questions rather than softballs. The high-value content is the framework stuff: permanent capital as the structural enabler of contrarianism, the career-risk explanation for why concentration is rare, and the genuinely useful distinction between growth and maintenance CapEx for thinking about the AI spending wave. The float/Berkshire walkthrough is the clearest plain-English explanation of why insurance is an investing machine that you’ll hear.

Where to keep the BS filter on: this is a fund manager talking his book on a broker’s podcast, so the framing is self-serving. The “600–700 basis points of outperformance” figure is cited from memory and unaudited here; his early activist record was also bumpier than the smooth retelling suggests (Valeant and Herbalife don’t come up). The AI-layer thesis — infrastructure safe, model makers exposed — is plausible and well-argued, but it’s a widely-held view by now, not a unique insight, and “Google could just slash token prices” assumes Google wants a price war more than it wants margins. The Trump commentary is the weakest stretch: brief, flattering, and not really analysis.

A 7. Genuinely useful mental models delivered cleanly, slightly discounted for being a promotional setting where the speaker’s own record goes conveniently unchallenged.