Chinese EVs, open-weight AI and battery bans: the export fight reshapes tech

In June a purpose-built LNG-powered bulk carrier, the BYD Zhengzhou, docked at the Port of Melbourne with 5000 newly built electric vehicles on board. It was the largest single EV shipment ever sent to Australia. RNZ published the account on 24 July, crediting ABC News chief business correspondent Ian Verrender.

Australia has no domestic car industry and few trade barriers, so it works as a litmus test for consumer taste. The numbers there are brutal for the old guard. In the first half of this year Mazda sales fell 17 percent, Toyota 21 percent, Subaru and Mitsubishi 25 percent each, and Nissan 32 percent, RNZ reported. Chinese brands went the other way: BYD up 124 percent, Chery up 77 percent, Geely up 495 percent. All three started from a much smaller base, which inflates the percentages, but the direction is not in dispute.

Last month almost a quarter of cars sold in Australia were fully electric, against 7 percent a year earlier.

The pitch is straightforward. Riz Akhtar, founder of the research group carloop, told RNZ he drives a Chinese EV that costs around A$40,000 and is packed with technology. An equivalent Japanese or European car, he said, would cost more than 25 percent more. "Consumers aren't stupid. They're not going to spend money they don't need to," he said.

The incumbents know the ground is moving. Toyota vice-chairman Koji Sato told an annual supplier meeting in March: "Unless things change, we will not survive." Volkswagen chief Oliver Blume has foreshadowed lay-offs of up to 100,000 workers, RNZ reported, with VW's share price halved since his predecessor was removed four years ago.

The AI front

The car fight is the visible one. The quieter contest is over the software and models that sit on top of the hardware. MIT Technology Review reported on 20 July that a free, open source model called Kimi, released by Chinese company Moonshot, appeared to rival models from OpenAI and Anthropic, which are not free. The release set off public insults among current and former Trump AI advisers.

David Sacks, the president's AI and crypto czar until March, called Anthropic's models "lobotomized" and "woke." Emil Michael, a senior Pentagon official, called OpenAI's new head of strategic futures a "supreme village idiot." Sacks also argued that Chinese models are popular because they come with fewer restrictions on use, putting aside built-in state censorship, according to MIT Technology Review. He no longer holds a formal advisory role.

The Guardian reported on 1 August that Chinese open-weight models have split the US industry. On one side are chipmakers and companies worried about OpenAI and Anthropic's dominance. On the other are those two labs, arguing that Chinese models pose security risks. Microsoft, Nvidia, Palantir and Meta signed a letter urging lawmakers not to restrict open models, and Nvidia chief executive Jensen Huang went to Capitol Hill to lobby. Treasury secretary Scott Bessent floated sanctions over alleged intellectual property theft; commerce secretary Howard Lutnick received letters from startup founders asking him not to cut off access.

Trump's own position was ambivalent. "We have to be careful in both ways. We don't want to restrict them when all of a sudden we come in second to China," he said, according to the Guardian.

Who writes the licences matters as much as who writes the code. The Conversation reported on 15 August that Chinese models such as Qwen, DeepSeek, Kimi and GLM are generally open-weight rather than fully open source: the software and final parameters are released, but not the training data or methodology. Some carry extra obligations. Moonshot's Kimi K3 licence requires licensees above certain revenue or user thresholds to display the model name prominently. The Mulan Permissive Software License 2.0 and the OpenAtom Model License 1.0 are bilingual, with the Chinese text prevailing in cases of conflicting interpretation, especially in Chinese jurisdictions.

OpenAI and Anthropic have accused Chinese open-weight model makers of large-scale unauthorized distillation, copying proprietary models through their outputs. In April the Trump administration announced efforts to curb the practice.

Batteries and the ban

In late August the Trump administration declared a national emergency that effectively bans Chinese batteries from grid-scale energy storage, MIT Technology Review reported on 10 September. The order bars installation of "any foreign-produced bulk-power system electric equipment" deemed a national security risk and names battery storage systems, inverters and transformers.

Shan Tomouk of Benchmark Mineral Intelligence called an outright ban "a bit of a surprise" that creates concern for US developers. BloombergNEF expects near-term delays, and analyst Isshu Kikuma said some projects may need alternative cell sources that cost more, with cancellation the worst case. Import taxes on batteries rose to 25 percent in January from 7.5 percent, and from 2026, 55 percent of the cost of materials for new storage projects must come from outside China and other restricted countries to qualify for tax credits.

The US could have enough domestic capacity by about 2030, though some plants may not run at full capability, pushing real supply past that date.

Hardware pressure runs alongside legal pressure. Huawei's US trial opened in Brooklyn federal court on Wednesday, the Guardian reported on 11 September. Justice Department attorney Taylor Stout told jurors: "Theft, lies, cover-up. For 20 years, that's how Huawei, a massive Chinese telecommunications company, victimized American companies and abused the American financial system." The government alleges conspiracy to steal trade secrets from five US companies, including router operating system source code from Cisco and a phone-testing robotic arm from T-Mobile.

Defence lawyer Brian Heberlig countered: "It's about competition, not conspiracy. Innovation, not theft." He said the Cisco and T-Mobile incidents were "actions by individual employees" and that management acted once it learned of them. The trial is expected to last three months.