Martin Shkreli: Big Pharma's Blind Spot, Betting Against Quantum & What Prison Taught Him

1. Key Themes

Quantum computing is one magic trick, not "faster computing"

Shkreli's core thesis is that quantum computing is widely misunderstood as generic speedup, when it actually offers a dramatic advantage on only a narrow class of problems. He credits his wife Chelsea, who studied at MIT and taught quantum computing with Scott Aaronson, for his education. His argument: "the only thing it's really good at, which indeed is a magic trick, is going from exponential-time algorithm, which is integer factorization or elliptic curve discrete log problem, which I want to mention a little later, shout out to the ECDLP problem, the quantum computers can do those in polynomial time" 00:02:32. For a business person, "unless you're in the business of factoring numbers, there's there's not a whole lot to be excited about here" 00:02:32. He adds structural limits: modern machines run "at a clock speed of hundreds of hertz or thousands of hertz, not at the gigahertz that we're used to. They're single-core, and they're single-threaded," and "quantum doesn't have storage. There's no such thing as quantum RAM. So you have to design a problem that has a really, really small input" 00:02:32. His empirical test is blunt: "Chelsea and I have tried every single quantum computer in the world. We cannot multiply 300 by 200 on any of these" 00:05:37. He also reports a demand-side signal: "Every time somebody pitches OpenAI, from what I've heard, about a quantum computer, they they say, 'No, we're not interested'" 00:02:32.

The AI compute stack may be obsolete in 20 years: photonics as the next substrate

Shkreli argues that the current GPU build-out may be a transitional technology, and that analog optical computing could make the matrix multiplication at the heart of AI nearly free. He frames the economics: "every time somebody uses the word AI, I actually think of the word matmul. These are matmul machines" 00:06:17. GPUs do matmul in n-cubed time, so "if you want to multiply 1,000-element square matrix versus another 1,000-element square matrix on an Nvidia machine, it takes a billion operations" 00:06:33. Then the discovery: "I saw a paper from Microsoft that was called AOC, Analog Optical Computing... what I saw in the corner of my eye was that matmuls in light are free... They happen in instant time" 00:06:33. He names startups "Neurophos, Opticor, Lumai that want to make these photonic matmul cores, which would run circles around what Nvidia does" 00:07:20, and says "Maybe we'll start a company" 00:07:20. On the capex implication: "in 20 years, these H100s are in the scrap heap, B100s are in the scrap heap, B200s are in the scrap heap" 00:11:30. He caveats that "there's a lot of details and nonlinearity and storage" still to solve 00:07:20.

The pharma molecule market: inefficiency as alpha

Shkreli's playbook was buying under-valued drug assets from big pharma and repositioning them. He describes it as "a really inefficient market. Pfizer would sell you a drug for a million bucks that was worth a billion dollars, and they didn't care" 00:08:29. A concrete data point: "I bought a drug from Bristol Myers for 2 million, and that same company is public today. It's worth 6 billion" 00:08:29. The mechanism was indication-shifting: sparsentan was developed by Bristol for hypertension, "a dead, sleepy market," and he "moved into a disease called FSGS... that never had a drug approved for it," and it is now FDA approved and "a blockbuster drug" 00:09:23. The binding constraint was diligence bandwidth: "our biggest bandwidth was problem was was reading these papers," which led him to start "a small NLP company on the side in 2015" to read the roughly "40 million papers" in medicine 00:09:34.

Jevons paradox and the infinite-demand-for-intelligence case

Shkreli stays long semis on the grounds that intelligence demand has no ceiling: "Jevons always wins in the end" 00:11:30, citing Alex Wang's view of "a million-X compute improvement." He pushes back on Brad Gerstner's TAM framing: "if you take the TAM and limit it to knowledge workers, you're really undercutting the TAM... there's no limit towards intelligence" 00:17:03. His illustrative example: "McDonald's can afford to spend $10 billion and say, 'OpenAI, give me 10 million GPUs. I want to know what the best next sandwich to launch is.'" 00:17:20. But he holds the counter-risk in view: "If there's a thousand times more efficient transformer, that's bad for Nvidia" 00:11:30.

AI-native conglomerates: the S&P 500 inside the labs

He describes a vision of AI labs spawning entire companies: a friend who "left OpenAI and started a company called Core Automation... invented the reasoning model at OpenAI" plans "to have within, you know, OpenAI, OpenAI Pharma, OpenAI Motors, OpenAI Aviation... you're going to have an a whole S&P 500 within OpenAI or Anthropic or or wherever" 00:17:20.

Hardware and distribution as the AI moat

Shkreli sees the delivery layer as potentially more valuable than the model: "Sam's hedge with hardware is really brilliant because... he'll always have the best model, I think, but being able to deliver that model is probably going to be more valuable" 00:20:09. On Meta, his pick for who over-earns: "they have the distribution. With Muse, they showed that they still have product chops. I think we're still on day one of of AI" 00:19:30.

The AI-safety/EA fight and e/acc

Shkreli is hostile to Anthropic's safety posture and to effective altruism. He says "I helped popularize and I even think I would say start the e/acc movement, which was this counter-insurgency" 00:14:50, and calls the fearmongering "ridiculous, you know, hyperbole" 00:14:34. On the realism of AI doom: "this is just software. It's not the end of the world... it's not going to kill us all" 00:15:30. He thinks "there'll be a time to panic. It's just not now. Physical AI presents those opportunities" 00:15:11.

Psychology of founders who blow up

Shkreli offers a self-diagnosis applicable to gifted founders: "if you grow up as a intellectually gifted person, you have this big albatross... You have to live up to something" 00:22:35. He applies it to Elizabeth Holmes and Raj Rajaratnam, and to himself: "I had a hard time admitting I was not the best hedge fund manager in the world, but I was a wonderful entrepreneur. That simple refusal to admit that put me in prison" 00:22:35. The takeaway: "I didn't realize how petrified I was of not being seen as a great trader" 00:23:20.

2. Contrarian Perspectives

Quantum computing is a business dead end that will rot like vacuum tubes

Against the consensus that Google, IBM, and Microsoft are building the next computing paradigm, Shkreli says the economics don't work. His historical analogy: "Would you go long punch cards and vacuum tubes, right? Mercury delay lines, core memory. These are all things that disappeared and went bankrupt. Those were stocks at one point" 00:04:59. When he asks big tech why they spend "hundreds of millions of dollars a year on this," the answer is "Well, someday there there might be something that comes out of it. That's not a great business to me" 00:05:20. He backed this with real money: "Unfortunately, we lost millions of dollars shorting quantum computing companies, but I told my wife that we gained billions in insight" 00:05:49.

AI will not meaningfully help drug discovery for a decade

Contradicting the prevailing narrative (including Dario Amodei's claims), Shkreli says: "I don't think there'll be a single AI-relevant drug in the next 10 years. AI does not help drug discovery." His credibility argument: he asked the room to raise a hand if they had gotten an FDA approval for a drug, "I'm pretty sure that's it," and notes "we've had computers for a long time. I built a Cray cluster to analyze drugs" 00:15:11. On Dario's "cure all diseases in 5 years" claim: "Total bullshit" 00:14:37.

Anthropic's origin story is a trade-secrets betrayal, and AI fear is narcissism

Shkreli frames Anthropic's founding as a violation of founder loyalty: "imagine you and I started a business, and I said, 'I'm going to show you Pandora's box'... you said, 'Thanks a lot, Martin. I'm going down the street. I'm going to start Pandora's box 2.'... I would probably sue you for trade secrets" 00:13:31. He attributes the safety fears to nerd psychology: "The nerds don't [know what to do in the spotlight]... they want to draw attention to themselves. They want to say, 'Well, the thing I'm working on is really amazing'" 00:15:30. He credits Nick Bostrom's book Superintelligence with scaring Elon, and says "it would have been better if some of the EA kind of folks never sort of had this like narcissistic kind of self-importance" 00:20:22.

The AI market will resemble the internet: value accrues to consumers, not the infrastructure builders

Rather than a winner-take-most model, Shkreli says "the only winners will be consumers and and businesses. They get this tax-free kind of internet. Cisco and some other guys will get a little bit out of it" 00:19:04. He is honest about the ambiguity: "I probably would have guessed in 2000 that Yahoo and Cisco were longs, too. So it's a tricky place" 00:11:30. His one exception is Meta 00:19:25.

3. Companies Identified

Meta

Social and AI company. Shkreli names it as the likely over-earner of the AI era: "I think it's Meta." Why: "they have the distribution. With Muse, they showed that they still have product chops. I think we're still on day one of of AI" 00:19:30. He also credits it with courage: "Meta has the spine to call it out and say, 'This is nonsense'" 00:20:22.

Nvidia

GPU maker. Shkreli remains constructive despite photonics risk: "that doesn't mean Nvidia's not a long, and I'm long semis right now" 00:11:30. He frames the risk as efficiency, not displacement by quantum: "They'll never beat an Nvidia machine. They'll never beat a million Nvidia machines" 00:02:32. Also notes his wife's father "was one of the first people in Nvidia" 00:01:49.

OpenAI

AI lab. Shkreli says he holds exposure "through a few different ways" and praises Sam Altman's hardware hedge: "he'll always have the best model, I think, but being able to deliver that model is probably going to be more valuable" 00:19:45. He also notes OpenAI declines quantum pitches 00:02:31.

Anthropic

AI lab. Shkreli would own it as an investment but dislikes its posture. "He's a fearmonger" 00:13:30 on Dario Amodei, and notes the "Dario trade" sell-off as a buying opportunity: "I sort of piled in... into this fear" 00:13:04. Also mentioned: SBF was early in Anthropic 00:22:01.

Neurophos, Opticor, Lumai

Photonic computing startups. Why mentioned: "there's a couple of startups like Neurophos, Opticor, Lumai that want to make these photonic matmul cores, which would run circles around what Nvidia does" 00:07:20.

Core Automation

Under-the-radar startup founded by Shkreli's friend who left OpenAI and "invented the reasoning model at OpenAI." Shkreli is "really excited about" it. Vision: "OpenAI Pharma, OpenAI Motors, OpenAI Aviation... a whole S&P 500 within OpenAI or Anthropic or or wherever" 00:17:20.

Microsoft (Analog Optical Computing research)

Mentioned as the source of the "AOC, Analog Optical Computing" paper in Nature that sparked Shkreli's photonics thesis: "matmuls in light are free" 00:06:31. Also named among big tech spending on quantum 00:05:15.

Google and IBM

Quantum computing spenders. "Google has a few prototypes. IBM has a little cloud." Mentioned as the subjects of his quantum critique 00:05:37. IBM also referenced via CTR (Computer Tabulating Recording) historical analogy 00:05:15.

Royalty Pharma

Biopharma royalty buyer. Described by Shkreli as "genius guys, obviously. Their big claim to fame... was getting this big royalty from Remicade, and they bought it for like one-times earnings. And then that royalty 20X-ed" 00:10:52.

Bristol Myers Squibb

Source of sparsentan and of the $2 million asset: "I bought a drug from Bristol Myers for 2 million, and that same company is public today. It's worth 6 billion" 00:08:29. Illustrates big pharma's undervaluation of molecules.

Pfizer

Used as the example of big pharma selling assets cheaply: "Pfizer would sell you a drug for a million bucks that was worth a billion dollars, and they didn't care" 00:08:29.

Roivant

Biopharma holding-company model. Shkreli names it when asked about the asset-flip model 00:10:52.

Cursor, Robinhood

Named as examples from Sam Bankman-Fried's investment portfolio that looked like "genius" 00:22:14.

Muse (Meta product)

Meta's AI product used as evidence of product chops and a "new avenue" 00:20:09.

Waymo

Mentioned in the context of the engineer who "went to jail" for trade-secret theft 00:14:22.

4. People Identified

Chelsea (Shkreli's wife)

Early OpenAI employee, MIT-trained, taught quantum computing with Scott Aaronson. Shkreli: "she's, I think, a big deal in technology. She was one of the first people at OpenAI... studied at MIT and taught quantum computing with Scott Aaronson" 00:01:49. She educated him on the quantum thesis.

Scott Aaronson

Quantum computing theorist, called "the pre-eminent guy in quantum" 00:01:49.

Peter Shor

Inventor of Shor's algorithm, which founded the field: "the whole field was invented because of Peter Shor's thought experiment... a revolutionary idea in computing, but it's not really a revolutionary idea for business" 00:02:32.

Jim Cramer

Shkreli's first boss at 16. "Cramer was actually a tremendously effective and skilled trader... he actually made 23% a year uh net net of fees" 00:08:06.

Sam Altman

OpenAI CEO. Praised for his hardware hedge and noted as having said "publicly that he thinks photonic computing is on a rival podcast. The photonic computing could be the future" 00:11:30; "Sam's hedge with hardware is really brilliant" 00:20:09.

Alex Wang

Meta's AI leader. Told Shkreli "Jevons going to... take care of all this, it'll be a million-X compute improvement" 00:11:30. Credited with "really creative product work" on Muse 00:20:09.

Dario Amodei

Anthropic CEO. Respected for building a trillion-dollar company but disliked: "I want to pay deference to him creating a trillion-dollar company, but I kind of hate the guy... he's a fearmonger" 00:13:20.

Jensen Huang

Nvidia CEO. Cited by Chamath for the $400 billion VC-to-AI redirect and the water analogy, and by Shkreli as having "the spine" to call out the AI-doom narrative 00:21:05.

Sam Bankman-Fried

Early Anthropic investor. Shkreli gave him advice on how to tough out prison, and cites him as an example of intellectual-gift albatross: "you have this big albatross... You have to live up to something" 00:22:35.

Elizabeth Holmes

Theranos founder. Cited alongside SBF as an example of the intellectually-gifted-albatross dynamic 00:22:34.

Raj Rajaratnam

Hedge fund manager. "a decent investor" who fell into the same pattern 00:22:34.

Mark Zuckerberg

Meta CEO. "came out the other day and said, 'We should move fast in in AI'" 00:15:11.

Nick Bostrom

Author of Superintelligence, who "scared the hell out of Elon" 00:20:22.

Elon Musk

Floated "competitor peer review" as a middle ground on AI safety 00:20:09; scared by Bostrom's book 00:20:22.

Warren Buffett

Cited in a quip that he'd "go back in time, he'd go back and shoot the Wright brothers and stop the aviation industry from ruining capitalism" 00:20:22.

Dustin Moskovitz

EA-linked figure named by David Friedberg as a brilliant example 00:21:15.

Bill Gates

Shkreli's childhood hero: "Bill Gates was my hero" 00:08:06.

5. Operating Insights

Bandwidth, not deal flow, was the real constraint: build tooling for diligence

Shkreli's asset-flipping model was limited not by finding deals but by evaluating them. "We saw this, and we built two very similar companies that our biggest constraint was finding people to vet these assets" 00:09:23. His response was to build NLP infrastructure early: "our biggest bandwidth was problem was was reading these papers. So I started a small NLP company on the side in 2015" 00:09:35. The tactic: when your edge depends on reading more than competitors, invest in a reading machine before it's obviously good (he notes "We had Watson. It was great. It wasn't great.").

Reposition the asset, not the product: change the indication, change the market

His sparsentan case shows that the value lever was the target market, not the molecule: "Bristol discovered and developed it for hypertension, which is sort of a dead, sleepy market that nobody really needs... we moved into a disease called FSGS, which is a disease that never had a drug approved for it. So you sort of try to angle it in a way that makes sense" 00:10:09. The generalizable lesson for operators: identify under-appreciated segments where the existing asset becomes the first solution.

Hire and invest around your self-image blind spot

Shkreli's most operationally useful reflection: "I didn't realize how petrified I was of not being seen as a great trader... you can be a great long-term investor, value investor, entrepreneur" 00:23:20. The tactic is to identify the role you're optimizing for ego over fit, and let go of it. "Just failing and admitting that you are not that great at that is not that bad" 00:22:35.

Evaluate technology by what it does for the customer's specific job, not by the scientific novelty

His quantum critique doubles as a framework: ask whether a novel technology wins on the customer's actual workload. "To a nerd, that's the greatest, coolest magic trick ever. But to a business person, unless you're in the business of factoring numbers, there's not a whole lot to be excited about" 00:02:32. Use a concrete benchmark test too: "We cannot multiply 300 by 200 on any of these" 00:05:37.

6. Overlooked Insights

Complexity classes (n-cubed) will become boardroom vocabulary

Buried mid-answer, Shkreli says: "I think things like complexity algorithms are going to be parlance, you know, in the next 5 to 10 years as we move towards a society" 00:06:33, and gives the n-cubed framing (a billion operations for a 1,000-element matmul). The implication is that AI cost structure will be discussed in terms of algorithmic complexity, and the investable edge shifts to anyone who can change the exponent (photonics, better transformers) rather than just add more chips. Combined with his point that a "thousand times more efficient transformer" would be "bad for Nvidia" 00:11:30, it hints that the biggest risk to the AI capex thesis is algorithmic, not demand-side.

The "Dario trade": fear-driven sell-offs on AI narratives as repeatable buying opportunities

In a single throwaway line about portfolio management, Shkreli says "I think this Dario trade was like a great reason to go long. So I sort of piled in, you know, unto this fear" 00:13:04, and calls the prior day's sell-off "ridiculous" 00:13:17. Paired with his belief that AI doom narratives are overblown, it implies a tradable pattern: AI-lab safety or labor-displacement essays can trigger indiscriminate sell-offs in the sector, creating entry points for those who believe the underlying demand is intact.