Five percent now, and another two percent every time the valuation rises by half.
That is the structure Anthropic has committed to a new public benefit fund, and the second clause is the one that matters. A company pledging 5% of its equity to a charitable vehicle is a familiar enough gesture in technology, generous at the margin and quietly shrinking in relative terms as every subsequent funding round dilutes it. A pledge with a ratchet attached does the opposite. The fund’s claim is indexed to the company’s success, so the more valuable the enterprise becomes, the larger a slice of it the public interest holds.
The same week, a rather different model of artificial intelligence era philanthropy was on display, with Elon Musk pledging one SpaceX share to every child in America. The two approaches are now sitting side by side and they are not variations on a theme. They are competing theories of what an AI company owes the world, and they disagree about almost everything except the fact that something is owed.
Consider how the two would behave if the companies in question became twenty times more valuable. The children’s shares would appreciate, each holder a little better off, the aggregate gift much larger in dollars but unchanged as a proportion of the enterprise. The benefit fund would mechanically acquire additional equity at each threshold, ending up with a materially bigger stake in a materially bigger company, and a correspondingly larger voice. One structure distributes upside. The other accumulates it on the public’s behalf.
The espresso parallel is about how you buy the farm’s output rather than how you drink it. Most buyers agree a price per kilo for the harvest, and when the coffee turns out to be exceptional and sells at four times the expected price in the cup, that windfall sits entirely with the roaster. A smaller number of buyers write the contract differently, so the farm takes a defined share of whatever the coffee eventually fetches, and in a great year the people who grew it get rich alongside the people who sold it. The first contract is a purchase. The second is a partnership, and the difference only becomes visible in the years when something extraordinary happens.
Anthropic has written the second contract, with itself, on behalf of the public.
What the coverage understated is the governance implication rather than the generosity. A fund whose stake grows with valuation becomes, in the scenarios that matter most, a shareholder of real consequence. If the transformative AI case is correct and these companies reach valuations that reorder national economies, the ratchet means the public benefit vehicle arrives at that moment holding a position large enough to be consulted rather than thanked. That is a very deliberate piece of design. It takes the uncomfortable question of who holds power if this technology works and answers it in advance, in a document, while the answer is still cheap to give.
It also sets a benchmark that competitors will now be asked about. The useful consequence of any public commitment of this kind is not the commitment itself but the question it creates for everybody else in the room, and that question is no longer whether a lab intends to do some good. It is whether the lab’s good intentions are indexed or fixed.
Five percent, growing with the valuation, documented before anybody knows how large the number gets. Two philanthropic models, one betting on the generosity of a very rich man and one betting on the arithmetic of a contract.
The contract does not need to be in a good mood.
Source: Reuters, “Anthropic to commit 5% stake to new public benefit fund”