Martin Shkreli: Big Pharma's Blind Spot, Betting Against Quantum & What Prison Taught Him
Martin Shkreli argues quantum computing is a business dead end, photonic computing may displace GPUs, AI fear-mongering is overblown, and his own prison term stemmed from refusing to admit a limit.
Watch on YouTube ↗Charlie's summary
- ▶ 2:01
Quantum computing is not faster computing: it has essentially one business-relevant trick, so Martin bets against it
Martin says quantum was invented around Peter Shor's insight that one algorithm, integer factoring or the elliptic curve discrete log problem, runs in polynomial time. Everything else, like sorting, is not faster. Machines run at hundreds or thousands of hertz, are single-threaded and have no quantum RAM. He says that “they'll never beat an Nvidia machine.” He and his wife tried every quantum computer and couldn't multiply 300 by 200.
- ▶ 4:32
Quantum is a bad business because big companies fund it on the hope that “someday there might be something”
Martin compares quantum to punch cards, vacuum tubes and mercury delay lines, technologies that went bankrupt. Asked why they spend hundreds of millions a year, Google, IBM and Microsoft say something might come out of it, which he doesn't call a great business. Google has a few prototypes and IBM a small cloud. Martin admits his short lost millions, but says it “gained billions in insight.”
- ▶ 8:08
Pharma's blind spot is an inefficient market for molecules, where big companies sell valuable drug rights cheaply
Martin says buying and repositioning drug rights offered huge alpha. He bought a drug from Myers for $2 million, and that company is now worth $6 billion. Big pharma will sell assets for a million that are worth a billion. With Sparsentan, he moved a Bristol hypertension drug into FSGS, a disease that had no approved drug. The bottleneck was finding people to vet assets, which led him to start an NLP company.
- ▶ 6:04
Photonic matmul could make today's GPUs scrap, and Martin sees insanity in covering the earth with computers
Martin says AI is mostly matrix multiplication, which costs n-cubed operations on GPUs. A Microsoft analog optical computing paper showed matrix multiplies in light are free and instant. He names startups like Neurophos, Opticore and Lumai. He thinks H100s and B200s end up on the scrap heap in 20 years. He still holds semis, citing Jevons paradox, but warns a 1000x more efficient transformer would hurt Nvidia.
Give your email to unlock the rest of the key points — for every episode.
Do more with Charlie
Free — sign in with your email, no password.