A Chip Ban Could Leak Just Miles Outside The Border
The hosts raise it as a possibility, not a finding, that offshore hardware could still serve Chinese work.

The Twenty Minute VC (20VC): Venture Capital | Startup Funding | The Pitch · 20VC: Jensen Huang Investigation: What Happens Now | Sequoia Leave London, Close Scout Program and End Evergreen Fund | NASDAQ Loses $2.5TRN in 24 hours | Anthropic targets $1.4TRN IPO valuationThe United States has sought to limit sales of powerful AI chips to China, but investors on the show raised a possible leak in that approach. Restricted chips in a nearby country could still do work for China based developers reached over the internet.
Those chips, called GPUs, both train AI models and run finished models for users, a job called inference. A team does not need the hardware on its own desk if it can rent the computing from a data center elsewhere.
Investor Jason Lemkin was being sarcastic about that loophole, not describing a proposed rule, when he spoke of letting firms train models if they would “keep the chips a few miles over the border.” He said firms could sidestep a China sale by putting the hardware in hyperscalers, the giant cloud companies, or other data centers just outside the country.
The hosts named Singapore and Thailand as examples of nearby places for that hosting. They did not present them as findings about capacity, only as places raised in the discussion where the chips might sit.
The motive is commercial, investor Rory O'Driscoll said. From Nvidia's point of view, he said, the company wants broad sales in part so Chinese rivals do not gain ground with alternative chips.
That talk came up during a separate discussion about a reported investigation involving Nvidia chief executive Jensen Huang and whether he would get a pass. O'Driscoll said a pass could not be assumed because national security and government handling questions were involved.
For the chip rule itself, the legal picture was uncertain in their account. Lemkin believed a congressional effort had passed the House but not the Senate, so he was describing a debated restriction rather than settled law.
In Lemkin's guess, the controversy would make two things harder, both direct GPU sales to China and hosting those chips across the border. O'Driscoll and Lemkin treated that as prediction, not as something that had already happened.
Questions people ask
What is the difference between training and inference?
Training builds a model by processing large amounts of data. Inference is using the finished model to produce answers, text, images, or video. Both steps can use rented computing, which is why the hosts discussed remote access.
What restriction were the hosts actually discussing?
The hosts talked about limiting GPU sales to China. They then asked whether placing the chips outside China, but renting their computing to China based work, would sidestep that kind of limit.
Why did Singapore and Thailand come up?
They named Singapore and Thailand as nearby examples where restricted GPUs might be hosted. The discussion did not include detail about data center size or networks there.


