Washington says it won't accept rules that slow AI while it writes new ones for who can buy the chips. If you build on frontier models, the second set of rules will cost you more than the first.
President Donald Trump told leaders at the APEC summit in Shenzhen on Thursday, November 19, that the United States will not allow regulation to stifle the AI industry, Bloomberg reported. According to SBS, he called the technology almost entirely good but said it has bad elements that countries have to manage carefully.
The Indian news agency ANI reported that he framed the argument around what he calls superintelligence, and said the US opposes rules that would control it. Outlets quote the adjective before 'regulation' differently, so we won't put a specific one in his mouth. The 'stifle' line itself is consistent across the coverage.
On Monday, November 16, the administration announced a new rule restricting shipments of advanced AI chips to data centers in Singapore, Thailand, Malaysia, Indonesia and Brazil, according to Reuters reporting by Karen Freifeld. Chinese companies have used those facilities to rent computing power and train models. Overseas data centers that supply significant volumes of computing power to China-based firms need US export licenses within 90 days of the rule taking effect, or face further restrictions.
Put the two events side by side and the administration's position is easy to read. A rule about who can buy a chip is national security. A rule about what a model can do is regulation, and that's the kind Trump says he'll block.
Nvidia's restricted AI chips are selling at 2x to 3x US prices on China's gray market, with B200 racks at 50% premiums and over $1 billion smuggled in a single quarter. Jensen Huang has called US export controls "largely backfired" as Nvidia fell from 95% to zero China market share, while Huawei credited the bans for accelerating China's domestic... - nvidia banned chips black market price - export controls china ai accelerator market
That's the split.
What the remark actually targets #
Frankly, nobody should treat the Shenzhen line as a policy. None of the reports we reviewed tied it to state AI laws, federal preemption or safety testing requirements. It was a speech at a leaders' plenary, delivered in China, with Fortune describing a trade truce deadline, new export controls and Iran as the main stakes in the room, and CNN's live coverage tracking talks with Xi Jinping and Vladimir Putin. The audience was other governments. The target looks like international frameworks for governing advanced AI.
That matters if you've been waiting for Washington to settle the domestic argument. This remark doesn't settle it. A founder deciding whether to budget for model audits, or a bank deciding how much to let an AI agent touch, still has the same open questions it had on Wednesday.
What does change is the cost of the compute side. A cloud provider in Kuala Lumpur or Bangkok that has Chinese customers now has a licensing process to run, and the ones that fail it lose access to the chips their business depends on. Frontier labs that rent capacity in those countries, and startups that route inference through them, should ask their providers where they stand.
Existing data centers get 90 days from the rule's effective date to secure a license. The deadline will fall in early 2027.
Also read: CoreWeave's McBee says GPUs beat buying back its 15% yield debt • Waymo s Bay Area robotaxi service after floods strand cars in San Francisco • How to set max iterations for AI agent loop before it drains your budget
This article is posted in AI News, check it out for more related stories.
Join the discussion #
Open in the community → Almost there. Sign in and your reply posts straight away.