Martin Shkreli on All-In: The Quantum Short and Why AI Won't Find Drugs

Martin Shkreli told the All-In Podcast he has lost millions shorting quantum computing, expects no AI drug within ten years, and picks Meta as the AI winner. In this 25 minute live segment, the former hedge fund manager and pharma executive, sentenced to seven years for securities fraud, takes questions from the hosts on stage. He covers quantum stocks, photonic computing, buying shelved drugs from big pharma, Dario Amodei, effective accelerationism, AI valuations, Sam Bankman-Fried and what prison taught him.

TLDW

Shkreli argues that quantum computing has one commercially meaningful algorithm, that today’s machines cannot multiply 300 by 200, and that the stocks are not a business, though he admits the short has cost him millions. He thinks photonic chips will send today’s Nvidia hardware to the scrap heap within 20 years, yet he is long semiconductors and owns OpenAI. He calls Dario Amodei a fear-monger, says AI does not help drug discovery, and expects the gains from AI to flow to consumers the way the internet’s did, with Meta the likeliest company to over-earn along the way. The segment ends with the most personal material: his refusal to admit he was not a great trader, he says, is what put him in prison.

Thoughts

Shkreli’s flattest claim is the one with a clock on it: not “a single AI relevant drug in the next 10 years,” because AI “does not help drug discovery.” He earns the room’s attention by asking who else has taken a drug to FDA approval, and no hand goes up. But his own account of where his money came from cuts the other way. His edge was finding molecules that big companies had shelved and aiming them at a different disease, and the constraint he names was reading: 40 million medical papers, plus patents and university work, which he tried to attack in 2015 with a side company built on IBM’s Watson. Reading everything and spotting the overlooked asset is the job language models are now good at. What he probably means is the slow part, trials and approval, which software does not shorten much. The claim is already under test. Insilico Medicine has published mid-stage trial results for rentosertib, a drug whose target and molecule both came out of its AI platform.

Asked what derailed Sam Bankman-Fried, Shkreli ends up answering about himself. Gifted kids carry an albatross, he says, and his was needing to be seen as a great trader when he was really a good entrepreneur. That fits the record better than the montage the show opens with. His 2017 conviction was for securities fraud, for misleading investors in his hedge funds, not for raising the price of Daraprim from $13.50 to $750 a pill. The hosts do ask about the price increase directly, and he answers with his teenage job at Jim Cramer’s fund and never comes back to it. Nobody on stage mentions that a federal court banned him from the pharmaceutical industry for life in 2022, which matters when the guest’s authority rests on drug development. It is a friendly room. We get real self-knowledge about trading ego and nothing about the pricing.

His market view only holds together as a timing bet. He says AI “has to be like the internet,” where the winners were consumers and businesses and Cisco and a few others “get a little bit out of it.” The same man is long semiconductors, owns OpenAI “through a few different ways,” names Meta as the company that will over-earn, and volunteers that in 2000 he would have been long Yahoo and Cisco too. One host puts the reconciliation to him: the job is guessing who over-earns and for how long, and he agrees. So the bullishness is a trade on duration, not a belief that the suppliers keep the profits. His line that Nvidia’s current chips are scrap in 20 years is the other half of that position.

The quantum short rests on a history that is half right. Peter Shor’s 1994 factoring algorithm made the field famous, but the idea is older: Richard Feynman proposed quantum machines in the early 1980s as a way to simulate quantum systems, which is to say chemistry. That is the application closest to Shkreli’s own trade, and he never mentions it. The business argument is stronger than the history. Slow clock speeds, no quantum memory, tiny inputs, and large companies whose best answer for the spending is “someday” are fair points. The best evidence for caution is his own result. He says he and his wife lost millions on the short, which is what being right about a technology and early on a stock looks like.

The attack on Dario Amodei should be read with the stated bias in mind (“I kind of hate the guy”), and two of its supports are loose. Shkreli says Amodei predicted full unemployment within one to five years. Amodei’s widely reported 2025 warning was narrower: that AI could eliminate half of entry-level white-collar jobs and push unemployment to 10 to 20 percent in that window. Shkreli also says he helped popularize and “start” effective accelerationism. Popularize is plausible. The founding is usually traced to pseudonymous accounts in 2022, the best known being Beff Jezos, later identified as Guillaume Verdon. The charge that Anthropic’s founders walked off with OpenAI’s Pandora’s box is an assertion, and a host pushes back that leaving to start a rival is ordinary in Silicon Valley.

Key Takeaways

Shkreli’s quantum thesis is that one family of problems, integer factorization and the elliptic curve discrete log problem, gets an exponential speedup, and that nothing else runs fast enough to beat a classical machine.
He says today’s quantum computers run at hundreds or thousands of hertz, are effectively single core, have no equivalent of RAM, and need problems with very small inputs.
He and his wife Chelsea have, by his account, tried every publicly available quantum computer and could not multiply 300 by 200 on any of them.
The short has lost them millions of dollars so far. He does not claim the trade has worked.
He thinks of AI as matrix multiplication, which costs a billion operations for two 1,000 by 1,000 matrices on a GPU, and says optical hardware can do the same multiplication essentially for free.
He names photonic startups including Neurophos and Lumai, says he may start a company in the area, and predicts Nvidia’s H100, B100 and B200 chips will be scrap within 20 years.
On pharma, he says large drug companies will sell rights to a shelved molecule for a million or two that can be worth billions if aimed at the right disease.
His example is sparsentan, which Bristol-Myers developed for high blood pressure and his company redirected to the kidney disease FSGS. He says he paid $2 million and that the company is worth $6 billion today.
He is long semiconductors and bought the selloff that followed a Dario Amodei essay, while saying he would have been long Yahoo and Cisco in 2000 as well.
He does not expect a single AI relevant drug in the next ten years and says AI does not help drug discovery.
He sees no valuation trap in AI because demand for intelligence is unlimited. His example is McDonald’s paying $10 billion for 10 million GPUs to decide its next sandwich, if more tokens keep producing better answers.
His analog for AI is the internet: consumers and businesses win, infrastructure suppliers get a little, and Meta is his pick to over-earn because of its distribution and its Muse product.
He says the need to live up to being a gifted kid explains Sam Bankman-Fried, Elizabeth Holmes, Raj Rajaratnam and himself.

Chapters

00:55 Why Quantum Computing Is a One-Algorithm Business

Shkreli credits his wife Chelsea, who he says was an early OpenAI employee and taught quantum computing at MIT with Scott Aaronson, for teaching him the subject. His point is that quantum is not faster computing in general. Asked which machine sorts a deck of cards faster, the kind of problem that opens Introduction to Algorithms, he says the classical one wins easily. The exception is factoring and the elliptic curve discrete log problem, which he calls a great magic trick for a nerd and not much of a business. One host argues the field is at the transistor stage. Shkreli’s reply is that punch cards, vacuum tubes and core memory were once investable too.

06:04 Matrix Multiplication and the Case for Photonic Chips

Asked whether breaking encryption is the natural use, he changes the subject to the problem he thinks matters. Every time he hears “AI” he thinks “matmul,” since multiplying matrices is what a GPU spends nearly all its time doing, at a cost that grows with the cube of the matrix size. He describes reading a Microsoft paper on analog optical computing and noticing that the same multiplication done in light happens almost instantly. He names a few photonic startups, says he and his wife may start one, and concedes that nonlinearity and storage are unsolved.

07:42 From Jim Cramer’s Fund to Buying Drugs Big Pharma Shelved

A host asks about the drug price increase that made him famous. Shkreli answers with his career instead. He says he started at Jim Cramer’s hedge fund at 16, was pointed at genomics, moved to a Tiger cub fund, and then found he was a better entrepreneur than trader. The opportunity was the market for molecules: big companies would sell rights to a drug for a million dollars that could be worth a billion. With sparsentan, he says the move was to take a hypertension drug into FSGS, a kidney disease with no approved treatment and about 30,000 patients. He compares it loosely with Royalty Pharma’s purchase of NYU’s Remicade royalty.

11:11 The AI Capex Buildout: Long Semis, Skeptical of the Endgame

Asked what he would short in the compute buildout, he says it is insane to plan for data centers in space and nuclear plants when a faster kind of computer is possible. He is still long semiconductors and says Jevons paradox usually wins: cheaper compute creates more demand. The risk he flags is a transformer design a thousand times more efficient, which would be bad for Nvidia, since there are not a million times more people to serve. He no longer runs a fund and describes his own investing as nimble.

13:04 Dario Amodei, Effective Altruism and e/acc

Shkreli says he bought the selloff that followed an Amodei essay and calls the selloff ridiculous. He then calls Amodei a fear-monger and objects to Anthropic’s origin, likening it to being shown Pandora’s box and opening a second one down the street. Effective altruism is “even worse.” He claims a role in starting effective accelerationism, defines it as roughly move fast and break things, and says there will be time to panic when AI is physical. For now, he says, you can turn off the computer.

16:01 Why He Says AI Does Not Help Drug Discovery

He lists Amodei’s predictions on unemployment and curing disease and rejects both. He asks the audience to raise a hand if they have taken a drug through FDA approval, and concludes he is the only one. Computers have been part of drug work for a long time, he says. He once built a Cray cluster to analyze drugs and can now use GPUs, but he sees little that AI adds. Sparsentan, sold as Filspari, won its first FDA approval in 2023 for a different kidney disease, nine years after his company’s board removed him.

16:35 Is There a Valuation Trap in AI Revenue?

A host cites a figure of $400 billion of venture money redirected into AI and asks whether investors are funding startups to buy tokens. Shkreli agrees that is happening and denies it is a trap. Sizing the market by knowledge workers undercounts it, he says, because demand for intelligence has no ceiling. He expects AI to start businesses, with whole portfolios of companies inside OpenAI or Anthropic, and mentions a friend’s startup he calls Core Automation. The McDonald’s example carries one condition: it works only if more tokens keep buying better answers.

18:27 AI as the Internet, and Why He Picks Meta

Asked for a historical analog, he picks the internet, where the benefit went to users and suppliers such as Cisco kept a little. Pressed on who over-earns in the meantime, he says Meta, for its distribution and for the product work on Muse that he credits to Alexandr Wang. He thinks chatbots will look like an early curiosity. He would buy OpenAI and Anthropic today, already holds OpenAI, and calls Sam Altman’s move into hardware a smart hedge because delivering the model may be worth more than having the best one.

20:08 AI Safety Fears and Who Has the Spine to Dismiss Them

A question about Elon Musk’s idea of competitors reviewing each other’s models gets an indirect answer. Shkreli says Nick Bostrom’s Superintelligence scared Musk, and that the field would be better off without what he calls the narcissistic self-importance of some effective altruists. He does not think OpenAI believes the warnings and says it does not want to look too different from Anthropic. He credits Meta, the president and Jensen Huang with saying so out loud. His summary: it is just computer science.

21:32 The Gifted-Kid Trap and What Prison Taught Him

The hosts suggest that unconventional thinkers may also struggle to stay inside the rules, and ask about Sam Bankman-Fried, whom Shkreli says he has advised on getting through prison. He says gifted people feel they must live up to something, and that failing and admitting it is not so bad. He was petrified of not being seen as a great trader and calls that a stupid thing to fixate on. What prison showed him is how alike people are: everyone inside was upset about a friend who lost touch or missing someone.

Notable Quotes

“To a nerd, that’s the greatest coolest magic trick ever. But to a business person, unless you were in the business of factoring numbers, there’s not a whole lot to be excited about here.”
Martin Shkreli, on what quantum computers are actually good at

“Unfortunately, we lost millions of dollars shorting quantum computing companies, but I told my wife that we gained billions in insight.”
Martin Shkreli, on how the quantum short has gone

“I’m just wondering why we think this is a sane idea to cover the earth with computers when we can make a faster one.”
Martin Shkreli, on the AI data center buildout and photonic computing

“I want to pay deference to him creating a trillion dollar company, but I kind of hate the guy.”
Martin Shkreli, on Anthropic CEO Dario Amodei

“I don’t think there’ll be a single AI relevant drug in the next 10 years.”
Martin Shkreli, on AI and drug discovery

“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.”
Martin Shkreli, on why he thinks he went to prison

“I didn’t realize how petrified I was of not being seen as a great trader. What a stupid thing.”
Martin Shkreli, on what changed for him in prison

Watch the full segment with Martin Shkreli on the All-In Podcast here. Nothing in it is investment advice, and his positions are his own account.

Martin Shkreli on All-In: The Quantum Short and Why AI Won’t Find Drugs