Alex Karp said the quiet part with the volume of a man who has never been on the wrong side of a disruption. Speaking about what AI will do to the pharmaceutical industry, he reached for the language of ordnance — it would land, he said, like a fucking nuclear weapon. The glee was unmistakable. The spectator's delight at a detonation that will not touch his own body. I watched this from the fire escape on the Fourth of July, fireworks still drifting south over the harbor, and I thought: this is how empire has always narrated itself.
Not as a choice someone engineered. As weather. As inevitable physics. As a bomb that lands without anyone having lit the fuse.
But let me be more careful than Karp was, because the argument beneath his showmanship deserves better than the costume he put it in. What he is describing is real, and it is not trivial. The pharmaceutical industry built its competitive moats out of information scarcity — proprietary compound libraries, paywalled clinical data, decades of molecular labor locked inside institutional vaults that charge admission by the patent year. Now a machine arrives that can process, correlate, and synthesize that information at a speed no human team can match. If you are Palantir — if your business is making infrastructure legible to itself — then what you see is genuine: a structural transformation in how drugs get discovered, tested, brought to market.
Karp's deeper point, stripped of its ordnance metaphor, is that incumbency in pharma rests on accumulated data advantage, and AI collapses that advantage by making the data operationally accessible to anyone with compute. The old moats drain. The knowledge democratizes. This is not fantasy. It is a plausible, even generous, reading of where the technology points.
I will give him that. The physics is real. But physics alone has never explained who gets burned — or who sells tickets to the burning.
Every sharecropper in the Mississippi Delta could have told Eli Lilly how this story ends. You pour your labor into someone else's land. The land learns your methods. Then the land announces it doesn't need you anymore.
What Karp narrates as democratization is, in structure, extraction. The data that trains these models did not materialize from compute cycles. It was produced by thousands of researchers — graduate students running gels at midnight, postdocs rewriting grant applications for the fourth time, clinical trial participants whose bodies bore the toxicity risk so that a molecule could be validated or discarded. Their labor became papers. The papers became training data.
The data got uploaded into infrastructure they do not own, cannot audit, and will never govern.
Here is where I want to surprise myself, because the extraction frame — while true — can become its own kind of comfort. It can let me rest in indictment without reckoning with the stranger question: What if Karp is partly right that the moat deserved draining? What if Eli Lilly's data hoarding genuinely killed people — kept molecules locked behind pricing walls, kept generic development slow, kept the bodies waiting? Then the disruption Karp celebrates might carry within it a real liberation — cheaper drug discovery, faster pipelines, diseases of poverty finally becoming economically interesting to solve. I have to hold that.
I have to hold that the sharecropper's land was also sometimes fallow under the old master, and that a new claimant arriving with better tools does not automatically make the situation worse for the person who was never going to own the harvest either way. But this is precisely where the question tightens: liberation for whom? Decided by whom? If the moat drains into Palantir's reservoir instead of a public commons, then what has changed is not the structure of extraction but merely its beneficiary. The grammar shifts.
The body remains on the table.
They keep saying your data as if ownership were the contested ground. Ownership is the decoy. The real question is governance — who decides what the data becomes, who converts collective labor into a product that enriches a single entity, who sets the price on the molecule that someone's body made legible. The slaveholder did not primarily argue he owned the slave's body; he argued the slave was property, which made the body a category error rather than a crime. The platform does not argue it owns your research; it argues that once your research became data, it entered a commons the platform merely organized.
In both cases the trick is jurisdictional — move the contested thing into a frame where the taking is not taking but taxonomy.
What interests me most is not the technology. It is the narrative posture. When Karp says nuclear weapon, he is not mourning. He is not warning. He is entertaining.
He is the man in the bleachers watching the demolition — close enough to feel the shockwave thrum his sternum, far enough that his house still stands on the next block. And this posture — the aestheticization of someone else's displacement — is older than silicon. I have watched it my entire life. I watched it when they called urban renewal progress. When they called mass incarceration a tough-on-crime era.
The grammar is always the same: turn a decision into physics, turn a taking into weather, then marvel at the storm you yourself seeded.
The light always arrives before the sound. By the time you hear the explosion, the burn is already working its way through the skin. Karp knows this. That is why he describes the weapon with delight rather than dread. He has already seen the flash.
He is just waiting for the rest of us to hear it.
