The collapse of the federal AI training moratorium has rekindled a bidding war for a few hundred senior machine-learning researchers, with multiple U.S. hyperscalers extending compensation packages above $20 million a year and, in a handful of cases, north of $40 million for the most sought-after specialists, recruiters and pay consultants said.

The escalation, which has accelerated sharply since the House Ways and Means Committee killed the Sanders-Ocasio-Cortez moratorium on April 22, has unsettled academic departments, startups and the boards of the hyperscalers themselves, several of whom have privately warned shareholders that compensation expense in their AI divisions will run materially ahead of guidance through the second half of the year.

“It is the most dislocated labor market I have seen in twenty-two years of doing this work,” said Priya Ramanathan, founding partner at the executive search firm Linden Strand, which places senior researchers at large technology firms. “Three months ago, when the moratorium looked like it would pass the House, we had clients quietly trimming offers. Last week we had a client raise an outstanding offer twice in forty-eight hours.”

Ramanathan said her firm had placed two researchers in the past ten days at total compensation above $35 million annually over four-year vesting schedules, both moves involving departures from frontier labs to rival hyperscalers. She declined to name the firms or candidates.

The figures cited by recruiters could not be independently verified against company filings, which typically aggregate research-staff compensation, and several pay consultants cautioned that headline numbers can be inflated by retention grants tied to specific multi-year milestones. But the pattern is consistent across interviews with eight industry recruiters, two academic department chairs and four startup chief executives contacted over the past week.

The driver, those people said, is the same surge in data-center capital expenditure that hyperscalers laid out in late April. Four of the largest U.S. cloud and AI firms have collectively guided to more than $210 billion in 2026 capex, much of it directed at training infrastructure. With construction now unconstrained at the federal level, and with state-level moratorium efforts in California, New York, Connecticut and Oregon still months from any binding vote, the firms are racing to staff up the research and engineering teams that will train models on the new clusters.

“You cannot spend a hundred and ninety billion dollars on GPUs and then lose the bidding war for the twelve people who know how to use them,” said Marcus Whelan, head of compensation advisory at the consulting firm Caldera Group. “That is the math the boards have been forced to internalize.”

Whelan said his firm had reviewed pay structures at three of the five largest U.S. AI employers in the past month and found that the top decile of researchers — roughly the top 200 individuals across the industry — were now commanding total packages averaging $14.3 million a year, up from $8.6 million at the start of 2026. The top quartile of that group, he said, had moved into a range that “five years ago would have been reserved for chief executives of S&P 500 companies.”

The shift has alarmed university departments, which had already lost large fractions of their tenure-track AI faculty to industry over the past three years. Prof. Hannah Liedtke, who chairs the computer science department at Carnegie Mellon, said in a phone interview that two of her senior faculty had received unsolicited offers in the past two weeks and that one had already accepted a leave of absence.

“We have stopped trying to compete on cash, because that is not a fight we can win,” Liedtke said. “We are competing on intellectual freedom and on graduate students. Those are real assets. But the gap is now so wide that I worry about what kind of academic AI ecosystem we have in five years.”

Startups are similarly squeezed. Eight founders of mid-sized AI companies contacted for this article said they had lost at least one senior hire to a hyperscaler in the past month, often after a counter-offer that exceeded their own total compensation packages. Two said they had pivoted to recruiting more heavily from Europe and India, where the dollar gap, while still meaningful, was less extreme.

“We can pay our people very well,” said Daniel Osei-Mensah, co-founder of the model-tooling startup Spindle AI. “But we cannot pay them a number that comes with nine figures in retention grants, and we cannot promise them access to a hundred thousand H-series chips. Those are the two levers, and only four companies in the world can pull both.”

Hyperscaler executives have begun, quietly, to push back against the narrative that the spending is unsustainable. A senior vice president at one of the largest cloud providers, speaking on condition of anonymity because the executive was not authorized to discuss compensation matters, said the top-line figures circulating among recruiters were “real but unrepresentative,” covering perhaps fifty to seventy individuals globally. Aggregate research-staff cost growth, the executive said, was running closer to 30 percent year over year, “which is high but not historically unprecedented for a build-out cycle.”

The talent surge has also begun to draw the attention of regulators. The Federal Trade Commission, which has been studying labor-market concentration in the AI sector, indicated last week that it had opened informal inquiries into the use of no-poach understandings and non-compete clauses among the largest employers. An FTC spokesperson, asked for comment, declined to confirm or deny specific investigations.

Senior congressional aides on both sides said the issue was unlikely to draw legislative attention before the August recess but that the optics of multimillion-dollar pay packages, set against the energy and water debates that animated the moratorium fight, could complicate efforts to reach a compromise AI bill later in the year.

For now, the bidding continues. Linden Strand’s Ramanathan said her firm had four searches active for senior research leads, with anticipated closing dates before the end of the month.

“It is going to get more expensive before it gets less expensive,” she said. “Nobody on either side of the table thinks otherwise.”

Industry analysts said additional capex revisions and updated headcount disclosures were expected when the remaining megacap technology firms reported quarterly results later this month.