Limited Liability

There was a time when the list of the world’s largest companies read like an inventory of things, oil, steel, automobiles, commodities that weighed, rusted, and occasionally exploded, but whose utility any child could understand without the aid of an analyst. Today’s list reads like a prophecy. The top spot belongs to Nvidia, valued at around 4.7 trillion dollars, followed by Apple, Alphabet, Microsoft, and Amazon, and whoever reads down to the tenth name discovers that nearly 40 percent of the entire value of the S&P 500 rests on ten variations of the exact same wager. Diversification, a word consultants pronounce with the solemnity of someone prescribing vegetables, has come to mean owning ten companies dreaming the same dream and sleepwalking toward the same abyss. Almost half of the retirement savings of the Ohio firefighter and the Haifa schoolteacher depend, without either having been consulted, on half a dozen boards of directors getting the future right.

We have seen this movie before, and it was in black and white. In 1929 the darling of the stock market was RCA, the radio company, the technology that was going to change everything, and which in fact did, transforming politics, war, and the solitude of kitchens without any of it sparing its shareholders from ruin. The stock lost 98 percent of its value, the Dow Jones index took twenty five years to revisit its own record, one in four Americans was left jobless, and the shockwave crossed the Atlantic and reached Berlin just in time to help shape the twentieth century. Days before the disaster, Irving Fisher, the most respected economist in America, assured the public, from the heights of his Yale professorship, that prices had reached a permanently high plateau. The phrase survived. His reputation did not.

After 1929 America learned. It created the SEC and separated the banks from the casinos, writing the new rules while the blood was still fresh. The lesson lasted exactly as long as collective memory does. In the nineties the grandchildren of that catastrophe, convinced that financial engineering had abolished gravity, repealed the separation to bipartisan applause, and in 2008 the rerun arrived in technicolor and full sound. Lehman Brothers fell on a Monday in September, global credit froze before Friday, and the remedy, hastily administered by a government that swore it loathed interventions, was to transfer to the American taxpayer the bailout of the very same banks that had privatized the profits of the party, while the rubble of the crisis asphyxiated the economy on a global scale. Hardly anyone went to prison. The losses were nationalized with an efficiency any Soviet commissar would envy. The precedent that interests us here was established for anyone willing to see. When the gamble of a few fails, the bill arrives, punctually, at the doorstep of all.

Optimists will say, with some reason, that the comparison is unfair, that today’s giants have real profits and paying customers. Let us concede the point. The RCA radios also worked admirably well, and that did not save anyone. But there is a more uncomfortable difference, and it resides in the product. At RCA, the danger was that the promise might fail. Here, for the first time in the history of bubbles, the danger is that the promise might be fulfilled. Notice the symmetry of the trap. If the promise fails, the most concentrated wager ever placed on a single idea implodes, the world gets its new 1929, and 2008 has already taught us the address where the bill will be delivered. If the promise is fulfilled, an intelligence greater than our own is born, exactly as announced by the executives building it, with the serene candor of someone who is a billionaire. Heads, we pay. Tails, we lose.

Daniel Kokotajlo, a young researcher who left OpenAI in 2024, refused to sign the non disclosure agreement the company demanded upon his departure and was willing to forfeit nearly two million dollars in stock, practically everything he owned, in exchange for an asset that appears on no balance sheet, the right to speak his mind. He estimates at 70 percent the probability that the current race will end badly for the species, and the species, it is worth remembering, is us. He describes companies founded on oaths of caution in which safety has been demoted to a mere launch pitch, executives who accelerate because they fear a rival might arrive first, and he proposes, with the tranquility of someone who no longer has a position to lose, a plan of almost scandalous sobriety, postponing superintelligence to 2040, opening the laboratories to scientific transparency, preventing power from settling into a few hands.

Kokotajlo gave up almost everything to be able to issue a warning. The owners of the wager cultivate a more discreet, and far more real estate oriented, prudence. A social media founder erects a compound in Hawaii with an underground shelter and its own supplies. A renowned investor secured citizenship in New Zealand, which is the end of the world with good vineyards. The head of OpenAI has already enumerated in an interview, with disarming frankness, what he hoards for an emergency, weapons, gold, antibiotics, gas masks purchased from the Israeli army, and a remote plot of land in California to fly to. Experts from Silicon Valley calculated that more than half of the region’s billionaires hold some sort of insurance policy against the apocalypse. These are men who rarely miscalculate a risk, and the market rewards them precisely for that. When the one selling the tickets is building his own lifeboat, it is wise to ask what he knows about the ship.

And here we arrive at the question no one asks out loud at a shareholders’ meeting, between the reading of the balance sheet and the customary applause. Can we trust the owners of these companies? Honesty demands a prior question, namely, knowing who they are, and the answer has the uncomfortable grace of mirrors. In part, they are us. Index funds, pension plans, automated savings that buy the entire market without ever looking inside. A market is, after all, an arrangement of trust, the deeply unsentimental expectation that each will pursue their own interest with competence, and that is what makes our case so singular. The machine has not broken. Every gear turns with perfect rationality. The manager of a pension fund in Cleveland with Nvidia in her portfolio merely wants to beat the market before December, and no one will censure her for it, least of all the retirees of Cleveland. It is the sum of individual prudences that produces collective recklessness. Carelessness, multiplied by trillions, dispenses with cruelty.

The only time a board attempted to place the mission above the share price, the experiment lasted five days. In November 2023 the OpenAI board fired Sam Altman alleging a loss of trust. Investors and employees with stock options in their pockets returned him to his chair before the end of the week, and the ones who left were the board members. The market filed the episode away as a crisis of governance, when perhaps it was the only time it functioned in broad daylight. It was the cleanest test ever conducted on who is in charge when prudence collides with price, and the result confirmed that we lose regardless of the outcome.

The regulation of 1933 functioned as a remedy because the economy, though ruined, was still breathing. The 2008 rescue functioned because there was a Treasury larger than the hole. That is the detail the next repetition refuses to guarantee. A financial crash leaves the world poorer, and a central bank patches it up by printing paper. The current wager, if it fails in the way its own engineers admit to fearing, will find no lender of last resort. One cannot print a second species.

Kokotajlo asks for little, in the end. He asks that the future arrive slowly enough to be seen, that companies valued in the trillions accept being worth a little less for a while in the name of all those who do not appear on their spreadsheets. Perhaps they will. History records more improbable conversions. But it is wise to read the contract before signing the proxy. In 1929 the shareholders lost everything. In 2008 we were the ones who paid for them. Next time, if the engineers are right, there will be no one left to pay and no currency that will serve.