Bridgewater's CIO Pairs an 18% Job-Loss Forecast With a Tax on AI Compute

Bridgewater Associates is the world's largest hedge fund, a macro-focused firm whose clients include sovereign wealth funds, pensions, and central banks. Last week its managing chief investment officer, Greg Jensen, told a podcast audience that AI could displace as much as 18% of US jobs within five years, and published a tax proposal in The New York Times that would put a price on every answer a model produces. Together those moves convert the "AI extinction" warning into a number, a date, and a tax form, and Jensen is asking readers to grade policymakers against a Covid-19 parallel: the question, he says, is whether the same complacency that missed the 2020 pandemic will be allowed to miss an even larger labor shock.

The number and the lever arrived in the same week. On Odd Lots, Jensen, who leads Bridgewater's AI strategy and was one of the earliest outside backers of both OpenAI and Anthropic, said the firm has been modeling AI's labor impact and concludes that "as much as 18% of US jobs might be displaced in five years." The phrase is a forecast, not a probability, and Jensen hedged it as such. What makes it unusual is the speaker. A man whose firm trades on behalf of sovereign wealth funds, pensions, and central banks has a reputation that depends on getting five-year macro scenarios right. If 18% is wrong, Bridgewater's macro book will show it before the policy debate does.

Days earlier, Jensen had put a price tag on the alternative. In a New York Times op-ed dated August 14, 2026, he proposed a "token tax" levied on the compute that produces each AI answer, with the revenue used to cushion workers and fund retraining. A token is the small chunk of text a model reads or writes; taxing them means taxing the act of generation, not the act of employment, which puts the bill on the companies building frontier systems rather than on the payrolls being automated. Jensen's argument is that the current tax code subsidizes AI capital expenditure and leaves labor holding the cost of displacement, so the code should be rebalanced before the displacement actually arrives.

Bridgewater's own research-and-insights page frames the policy choice as "likely the most important policy decision of our lifetime," language usually reserved for wartime fiscal policy or central-bank regime change. The page is short on the tax mechanism and long on the diagnostic test: it asks whether the US will treat AI labor disruption as a transition to be managed or a force to be absorbed. Jensen has put his version of the answer in the Times; the firm's own commentary is the question he wants the reader to ask.

The Covid-19 test is the lever Jensen is pulling. In the Odd Lots conversation, he argues that the current AI policy debate resembles the months before Covid-19 took over in 2020, when intelligence and public-health circles had warning signals and the political system did not. The parallel is not that AI is a virus. It is that the cost of moving late is paid by people who were not in the room when the warning was issued, and the political system has no natural incentive to internalize that cost before it arrives. A token tax, in Jensen's view, is what pre-pandemic preparedness funding was supposed to be: a budget line that exists only to make the political system act earlier than it otherwise would.

Andrew Yang's AI Commission released materials in August pushing for a payroll-tax offset funded by AI revenues, which overlaps with Jensen's revenue-use argument but lands on a different base (Odd Lots show notes). Yang's version is easier to administer, since payroll systems already exist; Jensen's is harder, since it requires a new audit on compute rather than on wages, but it taxes the right base if the goal is to slow the rate of automation rather than just compensate for it. The two proposals are not yet in dialogue, and the federal window for either is narrow.

What makes the next twelve months worth watching is not the rhetoric. It is the falsifier. Bridgewater has now published a number, a date, and a policy lever in the same week, and Jensen has named the historical test. If US job displacement does not approach 18% by 2030, Bridgewater's forecast fails and the policy case for the token tax has to be rebuilt on a different number. If it does, the question stops being whether Jensen was right and starts being whether anyone moved early enough to matter.