Was the AI executive order a mistake?
Letting Congress achieve a bipartisan deal would've been better for the AI industry
Welcome to Cautious Optimism, a newsletter on tech, business and power.
📈 Trending Up: Waymo, after an SF power outage … the value of precious metals … Binance’s inability to follow the law … TPU talk … self-driving cars
📉 Trending Down: EU-China trade relations … the Chinese real estate market (still) … lifespan expectations for Russian generals … Sequoia … demand for local chips in China? … dynamic grocery prices
Things That Matter
China’s AI sector is soaring: Here in the United States, we’re beset by warnings that the AI economy is a bubble about to burst. Imminently. But over in China, investors are going all-in.
Two recent IPOs from chipmakers MetaX Integrated Circuits and Moore Threads soared after listing, rising 700% and 400%, respectively. The two IPOs are just the start: AI model makers MiniMax and Zhipu AI intend to list, and have shared early financial records. At the same time, another chip company (Shanghai Biren Technology) also plans to go public.
And why not? Chinese investors are rewarding AI IPOs. MiniMax and Zhipu appear to be small companies with revenue in the eight figures, and valuations in the single-digit billions. As they might say in Silicon Valley, how quaint.
Still, Chinese retail investors are being offered an avenue into domestic AI leaders at a far earlier point than the average American investor.
No idea who MiniMax and Zhipu are? MiniMax’s open MiniMax-M2 model is impressive and has seen more than 140,000 downloads on Hugging Face alone. Zhipu AI’s (also known as Z.ai) GLM family of models was tasked with around 20 billion daily tokens on OpenRouter, before the holiday slowdown began to bite.
There’s a non-zero chance that we see OpenAI or Anthropic list next year. If they do, they will price in the hundreds of billions, if not trillions. Very cool, but I am a bit jealous that Chinese retail investors are getting to bet so early on local companies that could become their nation’s AI champions.
The (Un)Free Press: 60 Minutes had a segment on the El Salvadoran CECOT prison — you can read more about the facility in the HRW piece subtitled “Torture and Other Abuses Against Venezuelans in El Salvador’s Mega Prison” here — that was cut at the last minute. Social accounts for the news magazine announced the change in programming yesterday.
What happened? Piecing together a host of reporting, it seems the Trump administration refused to comment or otherwise engage with the CECOT piece, which CBS News’s new editor-in-chief, Bari Weiss, said was worth shelving the production for a later date.
CNN reports that Weiss “suggested that the program try to interview White House deputy chief of staff Stephen Miller, and provided Miller’s number.”
60 Minutes correspondent Sharyn Alfonsi said shelving the CECOT story was a “political” decision done to “shield [the Trump] administration.”
Setting aside the journalistic nuances of asking for comment, Weiss isn’t in a comfy spot, albeit the situation is of her own making. Paramount, the company that purchased The Free Press and installed her atop CBS News, has business with the White House.
POTUS intends to weigh in on the fate of Warner Bros. Discovery and its suitors, Netflix (which won the bidding war for WBD’s assets) and Paramount (which owns CBS News/60 Minutes/The Free Press). If the 60 Minutes crew put out a story that Trump doesn’t like, it could directly endanger Paramount’s hopes of buying WBD. (Here’s the latest on that saga.)
The President has already lashed out at 60 Minutes after it was acquired by a company run by a regime-aligned family. Another story that would bother the White House would be a bridge too far, some might think.
But why would a piece on CECOT upset the Trump administration? Because the prison is a human rights incinerator, a hellish place where legal rights do not exist, and journalists have “documented systematic physical beatings, torture, intentional denial of access to food, water, clothing, [and] health care.”
The current administration has used that prison as a dumping ground for people rounded up in the United States and expelled. Perhaps we should know more about what happened?
Was the AI executive order a mistake?
The recently signed executive order designed to dissuade states from enacting most AI-related regulations is at best a temporary patch. Without action by Congress, it may last as long as the current administration.
But AI companies are far from getting over their worries that they will have to comport with every state’s rules. Preferring a single, federal standard, these companies lobbied Congress to ban states from regulating AI themselves, but those efforts have failed so far.
Enter the EO. But governors from both major political parties are not amused; New York even passed an AI bill after the EO was signed.
The latest is that there was a chance at a bipartisan compromise on limiting state-level AI regulation. After the infamous 99-1 vote in the Senate to abandon an effort to preempt state-level AI regulation this summer, Politico reports:
[T]ech lobbyists began pulling together the contours of a deal on preemption that they believed could attract Democrats and some concerned Republicans on Capitol Hill. The tech lobby would get its ban on most state AI laws, while Democrats and tech-skeptical Republicans would receive new child safety protections as well as rules on frontier AI models. The defense bill eventually emerged as the most likely vehicle for that compromise. […]
But that representative, as well as other people familiar with the preemption talks, said Sacks’ reluctance to pressure House Majority Leader Steve Scalise (R-La.) or other lawmakers to compromise helped sink a major opportunity for the AI sector.
In fast-moving legislative maneuvers like this one, the White House often plays a key role by pressing Congress to compromise. But people familiar with the talks said Sacks had little interest in granting concessions to Democrats or skeptical Republicans — preferring instead to jam Congress by unilaterally preempting state rules with an executive order.
Following the EO, the chances of Congress passing a White House-friendly (and, therefore, tech industry-friendly) limitation on state-level AI rules appear low. It turns out that one coequal branch of the government doesn’t like being end-run.
The AI industry is running uphill right now, in PR terms. Two-thirds of the US population thinks it’s important to regulate the AI industry, even to the detriment of competitiveness with China. And major AI labs are underwater if we add up their critics and boosters.
Mix in rising local criticism of data center projects that do little to create local jobs while creating problems (rising power prices, for one), and you have a public primed to support aggressive AI regulation. A compromise set in law might have been more AI-friendly in the long-term than a limited EO with an expiration date.
Hindsight and all that, but I wonder if tech folks view the executive branch like the C-suite of a company and Congress akin to employees: You tell the staff what to do, right? And if they don’t, you cut them loose.
Government, fortunately, is not a business.