First, Silicon Valley’s founders believed they didn’t need government. Then they believed they could demolish it. Now they’ve decided to use it where they can and appease it where they must. This latest turn may usher in a new era of AI industrial policy.
The disdain came first. Robert Noyce famously ran Intel (founded in 1968) as an experiment in flat, anti-hierarchical management and viewed Washington as a distant irrelevance. The style became known as the “Californian ideology,” quasi-hippie bohemianism mixed with libertarian contempt for the state. That convenient self-image required ignoring an inconvenient history that the federal government was the semiconductor industry’s indispensable early customer. The Apollo program bought the chips that made Silicon Valley possible, the Defense Department built the first version of the internet, and the Reagan administration came to Noyce’s aid when Japan threatened to displace Intel. The disdain was real, but the independence was mythology.
By the 2000s, the anti-state mythology had grown more ambitious. If government was unnecessary, why not build civilization without it? Peter Thiel bankrolled the Seasteading Institute to construct floating city-states beyond the reach of any flag, and Tony Hsieh spent $350 million trying to conjure a new urban order out of downtown Las Vegas. Balaji Srinivasan preached the “network state.” The people who built software could build a society, too, and the state was legacy code. These ventures either never got off the ground or collapsed in spectacular fashion.
In the 2010s, the tech companies began styling themselves as the primary forces of progress in American capitalism. Their products, the story went, were connecting the world, fueling righteous revolutions, and giving the marginalized a voice. The companies became the champions of woke capitalism. Racial justice trainings and pronoun protocols were paired with free kombucha and dry cleaning. Meta stocked tampons in the men’s rooms.
Successful industrial policy requires coherent policy, clear lines of authority, and fair administration.
Then came the counterrevolution. Hypermasculinity—performed through million-dollar watches, gold chains, and the inevitable appearance on Joe Rogan’s podcast—was in, and diversity was decidedly out. Mark Zuckerberg medaled at a jiujitsu tournament. Elon Musk and he planned, then abandoned, a cage match. Employees who objected to the new culture were invited to leave. Brian Armstrong declared Coinbase a “mission focused” company, by which he meant employees would need to focus on crypto adoption over social justice. Any employee who didn’t like it should leave.
But this wasn’t a return to the old libertarianism. It was a new desire to sideline critics and openly exercise the political power that had always lurked behind Silicon Valley’s public service rhetoric. When Lina Khan’s FTC tried to hem in the giants, they took the familiar route available to concentrated wealth—they replaced her boss. The industry’s leading men wrote checks and hosted podcasts in support of a president they had previously disdained—and then took their seats on the inaugural dais.
They are now discovering that the best way to extend their power is to partner with the state—the libertarian’s worst nightmare.
Intel CEO Lip-Bu Tan chose to hand the government a 10 percent stake in his company—converting subsidies into shareholding—and won a rescue for a struggling “national champion.” Tech companies of all stripes, particularly AI ones, are getting large government contracts. Antitrust enforcement against the largest AI-adjacent platforms has grown conspicuously quiet, while the White House expedites permits for favored infrastructure.
If you don’t play nice, trouble is in store. In February, the Pentagon designated Anthropic—a rare company led by a CEO who declined to back Trump—a “supply chain risk,” a label previously reserved for the likes of Huawei, after the company refused to drop contractual limits on military use of its models in Trump’s war on Iran. The president ordered every federal agency to stop using Anthropic’s technology. Sam Altman seized the opportunity to ingratiate himself to Trump, announcing OpenAI’s own Pentagon deal for classified deployments, and accepting “any lawful purpose” language his rival had refused.
In June, the Commerce Department went further, invoking export control authority of contested legality to force Anthropic to shut down its most capable models, Fable 5 and Mythos 5, worldwide, just three days after launch.
A June executive order established a “voluntary” framework under which frontier AI developers give the government up to 30 days of prerelease access to their most powerful models. The order disclaims any mandatory licensing regime, but when the government asked OpenAI to delay the full rollout of GPT-5.6, OpenAI complied.
The clearest sign of the new order is how Anthropic’s standoff over its flagship products, Mythos and Fable, ended. The company spent 18 days with those models dark. Commerce lifted the export controls after Anthropic agreed to work with the government on safety protocols for its current and future models and to report malicious activity it detects. Commerce Secretary Howard Lutnick’s letter pointedly reserved the right to reimpose restrictions “should circumstances change.” The company now describes the arrangement as a possible “template for effective global coordination.” The administration’s most defiant AI critic has become a model collaborator, because it had to. Everyone got the message.
Yet Washington is only half of Silicon Valley’s political problem. The physical build-out that the AI business model requires is colliding with state and local government. In the first quarter of this year alone, local opposition blocked or delayed at least 75 data center projects worth roughly $130 billion—as much as in all of 2025. The number of organized opposition groups has more than doubled since December, to over 800 across 49 states. Maine’s legislature passed a statewide moratorium (vetoed only because it snagged a favored project), and the New York legislature just sent a moratorium of its own to the governor’s desk. Major tech lobbying efforts to reverse these restrictions are already afoot.
None of this should surprise us. Frontier AI looks like a naturally oligopolistic industry with staggering capital requirements, relatively low marginal costs, and hence winner-take-most dynamics. Industries with that structure have always ended up entangled with the state, from railroads to aerospace to banking. Entanglement is not necessarily bad. Some of American history’s most successful industrial policy—the Reconstruction Finance Corporation, the Defense Department’s seeding of the chip industry, Operation Warp Speed—came from exactly this kind of public-private fusion.
But successful industrial policy requires coherent policy, clear lines of authority, and fair administration. What we have instead is rule by impulse and deal—a supply chain designation for the recalcitrant, a contract for the compliant, and an equity stake for the favored.
Which is why the most interesting proposal in recent days deserves more than a shrug. Writing in the Financial Times, Altman called for an American-led international body—he has compared it to the International Atomic Energy Agency—to set global standards for AI safety testing. Skeptics will note, fairly, that the CEO proposing the referee also owns a team. But the instinct is still right. An international governing coalition could set reasonable standards and shift the center of regulatory power out of an increasingly politicized Washington. That could elevate the role of expertise and make policy more impartial.
Either way, the American government will continue to shape the AI industry, and now Silicon Valley is looking to take advantage of that fact. In the startup vernacular, they’d call this a pivot. The rest of us should call it what it is: the most powerful industry in America casting off its libertarian instincts because it sees how much there is to gain from collaborating with a government that won’t publish the rules.
