Martin Shkreli: Quantum Computing Fraud, Pharma Deals

Martin Shkreli owns, by his own admission, the only existing copy of a Wu-Tang Clan album he has never played. He also spent seven years in federal prison for securities fraud, and before that he raised the price of a life-saving drug by 5,000 percent (00:00). None of this makes him an obvious authority on quantum computing. And yet.

On the All-In podcast, Shkreli walked through a surprisingly technical argument for why quantum computing, the hyped alternative to ordinary chip-based computers, is mostly a business mirage. His case rests on a detail most people miss: a quantum computer, which stores information in quantum bits instead of ordinary binary bits, runs at a clock speed of hundreds or even a few thousand hertz (03:10). An iPhone runs at billions of cycles per second. A quantum machine, by comparison, is single core and single threaded, good at almost nothing a business actually needs done quickly.

"The modern quantum computers run at a clock speed of hundreds of hertz or thousands of hertz, not at the gigahertz that we're used to." - Martin Shkreli [03:10]

There is a real exception, and Shkreli is careful about it. A quantum computer can solve integer factorization and something called the elliptic curve discrete log problem in polynomial time, meaning the calculation stays manageable as the numbers grow instead of exploding the way it does on an ordinary computer (03:29). That is the one trick, discovered by mathematician Peter Shor, that justifies the entire field. It happens to matter a great deal for cryptography. It does not happen to matter for sorting a deck of cards, running a search engine, or training a language model. There is also no quantum equivalent of RAM, so any problem has to be shrunk down to fit a machine with almost no memory (04:06). Shkreli and his wife, Chelsea, who studied quantum computing at MIT under theorist Scott Aaronson, have tried every quantum computer currently available, he said, and none of them can multiply 300 by 200 (05:37). He has shorted several quantum computing stocks and lost money doing it, though he maintains the position has been worth the insight even where it has not been worth the money.

The Thing That Might Actually Replace Nvidia

If quantum computing is a magic trick with one application, Shkreli thinks the more interesting rival to Nvidia's dominance is sitting in plain sight: photonic computing, which performs calculations using light instead of electricity. Nearly every modern AI model spends almost all its computing time on one operation, matrix multiplication, the repeated multiplying and adding of large grids of numbers (06:17). On a standard graphics chip, doubling the size of that grid makes the job many times slower, a scaling problem computer scientists call N-cubed time (06:37). Shkreli described reading a paper in the journal Nature from Microsoft researchers on what they called Analog Optical Computing, and noticing that light-based matrix multiplication is close to free.

"Matmuls in light are free. They happen in instant time, which is a really remarkable thing." - Martin Shkreli [07:20]

He named three early-stage startups working on photonic chips, NeuroPhos, OptiCore, and Lumi, and said he thinks today's Nvidia chips, the H100 and the newer B100 and B200, could be in the scrap heap within twenty years (11:22). He was careful to note this is not a reason to short Nvidia now. He still holds the stock long, alongside semiconductor names, because he thinks the Jevons paradox, the idea that making a resource more efficient tends to increase total demand for it rather than reduce it, will keep demand for compute rising even as individual chips get more efficient (11:58). The one thing that would hurt Nvidia, he said, is a transformer architecture a thousand times more efficient than today's, because that kind of leap would outrun demand growth rather than follow it (12:18).

What gives this critique weight is where Shkreli learned to spot mispriced assets in the first place. He started working at television personality Jim Cramer's hedge fund at sixteen years old (08:00), a fund that, by his account, earned 23 percent a year net of fees through the 1980s and 90s by trading constantly (08:20). Shkreli said he eventually realized he was a better entrepreneur than a trader, and the business he built instead was a strange, overlooked corner of the pharmaceutical industry: buying the rights to drugs that big companies had stopped caring about. He once paid pharmaceutical company Bristol-Myers $2 million for a drug; the company built around it is now worth $6 billion (09:05).

"Pfizer would sell you a drug for a million bucks that was worth a billion dollars. They didn't care." - Martin Shkreli [09:05]

The specific drug, later named sparsentan, had been developed for hypertension, or high blood pressure, a crowded market nobody needed another entry in. Shkreli's team repositioned it for FSGS, a kidney disease known formally as focal segmental glomerulosclerosis, which affects roughly 30,000 people in the United States and had never had an approved treatment (10:30). It later won FDA approval. The arbitrage, as he described it, was not scientific insight so much as attention: reading through a pile of roughly 40 million published medical papers that nobody at a big, distracted pharmaceutical company had time to sift for overlooked value (09:48).

That background is also why Shkreli is skeptical that artificial intelligence is about to transform drug discovery, a claim he thinks Anthropic chief executive Dario Amodei has overstated. Shkreli said he doubts there will be a single AI-discovered drug on the market within ten years, noting that computational tools, supercomputers included, have been part of drug discovery for decades without producing the kind of leap Amodei has predicted (16:05). He extended the same skepticism to the broader AI safety conversation, which he called fear-mongering, and to effective altruism, the philosophy of trying to do the most measurable good, which he said is shot through with a narcissistic kind of self-importance among some of its most visible adherents, including people he considers genuinely brilliant, such as Amodei and FTX founder Sam Bankman-Fried (20:29).

What Brilliance Does Not Protect Against

The conversation's most unguarded moment was not about chips or molecules. Asked what sent him to prison, Shkreli did not describe prosecutorial overreach or a technicality. He described a psychological failure to let go of an identity.

"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 [23:03]

He drew a line from his own story to Bankman-Fried and to Theranos founder Elizabeth Holmes, arguing that intellectually gifted people often carry a private terror of being seen as merely ordinary at the one thing they have staked their identity on. Sometimes, he said, a person can be really smart at one thing and really dumb at another (21:06), a line he applied first to himself.

It is a strange note for an episode that spends most of its runtime on clock speeds and molecule arbitrage, but it is the note that explains the rest. Shkreli's method, in quantum computing, in pharmaceuticals, in his reading of the AI boom, is to ask what a thing is actually for, stripped of the story told about it. Quantum computing is a magic trick for one narrow mathematical problem, not a faster version of ordinary computing. A pharmaceutical patent sitting unused at a big company is not worthless, it is unattended. And a hedge fund manager's reputation, it turns out, was never the thing worth protecting.