This is Morgan Stanley's 'Hard Lessons' series interviewing legendary macro investor Stan Druckenmiller. He ran Duquesne Capital from 1981 to 2010, compounding at roughly 30% annually with not a single down year.

He first tells a case that is 'unsexy and has nothing to do with AI' yet best represents his investment process: last summer and fall, AI was 'overheating in an unsettling way and starting to rhyme with 1999-2000,' so his team turned to look for other opportunities and dug up Israeli generics maker Teva — on the surface a boring 6x-P/E generics company, but in fact new CEO Richard Francis (who had run the same playbook at Sandoz) was pivoting it from generics toward biosimilars and innovative drugs. The beauty of it: value investors dumped it because it was 'chasing growth,' while growth investors didn't want it because the pivot wasn't yet proven, so the stock stayed stuck at 6x P/E. Over six months the stock rose from 16 to 32, its valuation re-rating from 6x to about 11-12x. He uses this to make his methodological point: 'Looking at today, you can't make money; you have to look at what will change ahead and how investors will re-evaluate it.'

He also explains why he pushed heavily into biotech: from 30 years on the board of Memorial Sloan Kettering Cancer Center, he knows deeply that 'one of AI's best real-world applications is biotech' (drug discovery, diagnosis, monitoring), while the biotech sector has 'been flat for 4 years' and its technical momentum is starting to reverse, and the fear AI provokes may itself spark a rotation of sector leadership. He stresses he doesn't need to understand the details of gene sequencing, gene editing, or proteins at all — 'thank God, the answer is an emphatic no' — what he needs is an internal expert he trusts, plus his own feel for how the market will embrace such a change; his analysts' 'enthusiasm' matters to him as much as the facts, because he's 'not smart enough to understand those facts.' He rates his edge as not IQ but 'trigger pulling.'

Macro view: the US economy is already strong and will get stronger (the Big Beautiful Bill + heavy stimulus), and the Fed will most likely not hike and will cut; but valuations are in a historically high range and not cheap. If he had to 'come down from Mars and build a portfolio from scratch,' he'd lean toward a basket of 'compromise stocks' — over the past three years the portfolio was very AI-driven, and now AI is still there but 'no longer the driving engine'; he still holds large Japan and Korea positions (part AI, part not); he's short the dollar (its purchasing power is in a historically high range and foreigners are severely overweight the dollar); long copper (no meaningful new supply for 8 years, compounded by AI and data-center demand, holding mainly copper futures rather than miners); holding gold (mainly a geopolitical trade rather than a monetary one); and because he holds these risk assets, he shorts bonds as a hedge — he says shorting bonds won't necessarily make money, but if the economy is strong and it's 'disinflationary growth' it could pay off big, covering both ends of the matrix.

The most compelling part is the Nvidia 'hard lesson': in early 2022, the young people at his firm and his partners' AI circles started buzzing about AI, and he also noticed Stanford students shifting from 'half crypto, half AI' toward more of them going into AI ('watching where the kids go' is his old VC habit — he bought Palantir in '08-09 too because it was then the coolest company kids most wanted to join). His partner brought people in to explain AI to him; 'most of it flew over my head,' but he knew this was big, so he asked 'what should I buy,' and the answer was 'Nvidia — that's the way to bet on AI.' He first built a position that was small but 'big enough to hurt or to matter'; about two weeks later ChatGPT burst onto the scene, and once he saw even its crude capabilities he understood, so he doubled up; then a Morgan Stanley analyst who'd come from the tech world publicly woke up a room of macro types who could only pontificate about worldviews — 'you can't see the forest for the trees' — so he doubled again. He admits that three months earlier he 'probably couldn't even have spelled Nvidia' and couldn't articulate its earnings. He held all the way from 150 to 800 (he'd publicly declared 'I can't imagine selling in the next two or three years'), but ultimately 'couldn't handle the success' and cut at 800 — five weeks later it ripped to 1,400, and he was 'absolutely miserable.' He blames it on 'getting more timid with age,' mocking himself as 'Mr. Taco' — no, 'Daco (Druckenmiller always chickens out).'

On mindset: he thinks 'contrarian investing is overrated,' quoting Soros that 'the crowd is right 80% of the time; you just can't get trapped in the other 20%'; what he truly enjoys is the state of 'having extremely strong conviction when no one believes me.' The most valuable lesson he learned from Soros wasn't macro but 'position sizing' — 'what matters isn't being right or wrong, but how much you make when right and lose when wrong.' He also cautions that old tricks like technical analysis and 'a stock failing to react to good news often foreshadows bad news' have lost much of their edge because, after 2000, smart people flooded in and everyone knows them ('loved to death'). Finally he confesses to '15 years of imposter syndrome,' relying on 40 years of accumulated 'scars, successes, and pattern recognition,' and he tells young fund managers: during drawdowns, 'don't torture yourself — turn the page and move on.'

(For AI investors: a top-tier macro maestro proactively 'trimmed his AI exposure and rotated into neglected biotech and copper' at a time of AI overheating, and candidly shares the pain of 'selling a big winner too early' — the core point being that 'investing in AI ≠ buying overpriced AI stocks,' and positioning AI as an underlying force that drives other sectors (pharma, copper/power).)