If there is one universal truth about Silicon Valley investors, it’s that they will find a way to make a profit, and this year the richest among them found a new place to invest: inside the marble halls of Congress. As AI models grow more capable, lawmakers have increasingly floated regulation aimed at protecting children from manipulation and the public from fraud, including the TAKE IT DOWN Act criminalizing AI-generated intimate imagery, the proposed GUARD Act holding chatbot makers accountable for encouraging harmful behavior, and a White House national policy framework built around protecting parents, communities, and creators. Fearful that any of it might slow their companies down, AI executives and investors have mounted an extraordinary campaign to keep Congress friendly.

Over the past year, the tech world has donated nearly $400 million to bipartisan super PACs built to protect its interests. Marc Andreessen and OpenAI cofounder Greg Brockman each gave $50 million to the new Leading the Future PAC, explicitly organized to elect candidates it calls “pro-innovation.” Meta has donated $65 million over the past year to two pro-tech super PACs, including one named, with a straight face, Mobilizing Economic Transformation Across (Meta) California. And Fairshake, the crypto-industry PAC that Andreessen and his partner Ben Horowitz founded in 2023, is sitting on a $193 million war chest it intends to spend this cycle.

Together, these organizations have reshaped American campaign finance in a way that would have seemed unthinkable a few years ago. Lobbying is as old as Congress itself, but campaign spending used to be comparatively blunt — donations and PACs built around electing specific candidates or parties. The 2010 rulings in Citizens United v. FEC and SpeechNow.org v. FEC established that independent political spending is a form of speech, letting individuals spend unlimited sums as long as it isn’t coordinated with a campaign. Super PACs exploited that opening, but for over a decade, single-issue versions rarely moved the national needle: in 2022, the largest were the League of Conservation Voters, at $33 million, and an AIPAC-affiliated group that spent $26 million. Fairshake blew past both by an order of magnitude, collecting $68 million from Coinbase and $45 million from Ripple alone en route to an unprecedented $260 million raised for the 2024 cycle.

That money bought Fairshake the power to make and break careers. It dropped $10 million into the California Senate primary to attack Katie Porter, a Democrat who had built her campaign around corporate accountability — ads over her treatment of former staffers helped sink her Senate bid and lingered over her failed 2026 run for governor. The contrast with Fairshake’s preferred candidate could not be starker: Donald Trump, who upon returning to office declared his support for crypto, established a strategic Bitcoin reserve, and began publishing government data on blockchain currencies. Grateful investors even erected a golden statue of him holding a Bitcoin outside the Capitol.

The lesson of Fairshake was simple — support us and we’ll support you, cross us and we’ll end your career — and it’s a lesson that travels. NBC News reports that Leading the Future was explicitly modeled on Fairshake, down to sharing its major funders and advisors. Meta’s political operation is quieter but cut from the same cloth, with its PAC’s director stating that safety regulation “could stifle innovation, block AI progress.” As midterms approach, that targeting strategy is already in motion: Leading the Future has spent millions on attack ads against Alex Bores, a former Palantir executive whose New York AI regulations are, according to Politico, now the toughest in the country. If it succeeds, the message to the rest of Congress writes itself: cross Silicon Valley, and this happens to you.

But there’s a second act to this story. While Bores absorbs a barrage of attack ads, his campaign is being propped up by a different wing of Silicon Valley altogether. In response to Leading the Future, former congressmen Brad Carson and Chris Stewart formed a super PAC called Public First Action, built around transparency requirements, safety guardrails, and export controls meant to preserve American AI leadership. It has backed candidates in both parties, including Republican Senators Marsha Blackburn and Pete Ricketts, both of whom have pushed for AI safeguards.

Public First Action has raised $50 million this cycle, $20 million of it from Anthropic, the maker of Claude and OpenAI’s chief rival. On its face, that’s a strange move — companies like OpenAI, Perplexity, and Meta bankroll anti-regulation candidates precisely because regulation makes their businesses harder to run. Anthropic looks, at first glance, like it’s shooting itself in the foot.

The difference is that Anthropic has staked its identity on safety. It operates as a public benefit corporation overseen by a Long-Term Benefit Trust meant to keep it accountable to the “benefit of humanity,” a structure that traces back to why siblings Dario and Daniela Amodei founded the company in 2021, reportedly out of concern that developments at OpenAI were becoming difficult to control. The company has published its model’s constitution and releases system cards detailing where its models fall short — the kind of transparency Congress has been asking the industry for voluntarily. Its self-imposed safety work also slows it down relative to less cautious rivals, which gives Anthropic a very practical incentive to want those same constraints imposed on everyone else by law. Supporting the candidates who’d write that law, and banking the goodwill that comes with it, is as much a strategic bet as it is a values statement — one that, in this case, happens to serve the public alongside the company.

One of those bets is close to home: Public First Action has committed at least $500,000 to Scott Wiener, the California state senator running for Nancy Pelosi’s San Francisco House seat. Wiener authored SB 1047, the vetoed bill that would have required safety testing at large AI labs, and SB 53, the law that did pass, imposing disclosure and whistleblower protections on major AI companies. He’s running against two well-credentialed progressives, Saikat Chakrabarti and Connie Chan, in one of the bluest, most contested seats in the country — and Public First Action’s bet is essentially that not all Democrats are alike on tech, and that Wiener’s track record of actually moving legislation is worth defending.

For now, most of the candidates Public First Action backs are ahead: Wiener leads his race, Bores sits near the front of a wide-open field in New York, Blackburn has already secured the Republican nomination for Tennessee governor, and Ricketts holds a narrow lead against an independent challenger. If that holds, it would give AI-safety advocates something they’ve lacked — a working caucus of legislators willing to write serious rules for the industry.

The threat that Silicon Valley’s money poses to that outcome is real: if Leading the Future and its allies repeat what Fairshake pulled off in 2024, pro-regulation candidates get squeezed out and AI development proceeds with even fewer guardrails. But that ending isn’t written yet. Public First Action is the smaller side in this fight by a wide margin, and it’s backing its candidates anyway — a bet that, even against $400 million, a determined minority in Congress can still hold the industry accountable to everyone else.