US to Allow Nvidia H200 Chip Shipments to China, Trump Says 

A Nvidia logo appears in this illustration taken August 25, 2025. (Reuters) 
A Nvidia logo appears in this illustration taken August 25, 2025. (Reuters) 
TT

US to Allow Nvidia H200 Chip Shipments to China, Trump Says 

A Nvidia logo appears in this illustration taken August 25, 2025. (Reuters) 
A Nvidia logo appears in this illustration taken August 25, 2025. (Reuters) 

The United States will allow Nvidia's H200 processors, its second-best artificial intelligence chips, to be exported to China and collect a 25% fee on such sales, US President Donald Trump said on Monday.

The decision appears to settle a US debate about whether Nvidia and rivals should maintain their global lead in AI chips by selling to China or withhold the exports, though Beijing has told companies not to use US technology, leaving it unclear whether Trump's decision would lead to new sales.

Nvidia shares rose 2% in after-hours trading after Trump made the announcement on Truth Social, following a 3% rise during the day on a report by Semafor.

Trump said in his post that he had informed President Xi Jinping of China, where Nvidia's chips are under government scrutiny, about the move and that he "responded positively."

He said the US Commerce Department was finalizing details of the arrangement and the same approach would apply to other AI chip firms such as Advanced Micro Devices and Intel.

Trump's post said the fee to be paid to the US government was "$25%", and a White House official confirmed he meant 25%, higher than the 15% proposed in August.

"We will protect National Security, create American Jobs, and keep America’s lead in AI," Trump wrote on Truth Social. "NVIDIA’s US Customers are already moving forward with their incredible, highly advanced Blackwell chips, and soon, Rubin, neither of which are part of this deal."

Trump did not say how many H200 chips would be authorized for shipment or what conditions might apply, only that exports would occur "under conditions that allow for continued strong National Security."

Administration officials consider the move a compromise between sending Nvidia's latest Blackwell chips to China, which Trump has declined to allow, and sending China no US chips at all, which officials believe would bolster Huawei's efforts to sell AI chips in China, a person familiar with the matter said.

"Offering H200 to approved commercial customers, vetted by the Department of Commerce, strikes a thoughtful balance that is great for America," Nvidia said in a statement.

Intel declined to comment. The US Commerce Department, which oversees export controls, and AMD did not respond to requests for comment.

A White House official said that the 25% fee would be collected as an import tax from Taiwan, where the chips are made, to the United States, where the chips will undergo a security review by US officials before being exported to China.

FEARS OF CHIPS STRENGTHENING CHINA'S MILITARY

China hawks in Washington are concerned that selling more advanced AI chips to China could help Beijing supercharge its military, fears that had first prompted limits on such exports by the Biden administration.

The Trump administration had been considering greenlighting the sale, sources told Reuters last month. Trump said last week he met with Nvidia CEO Jensen Huang and that the executive was aware of where he stood on export controls.

"It’s a terrible mistake to trade off national security for advantages in trade," said Eric Hirschhorn, who was a senior Commerce Department official during the Obama administration. "It cuts against the consistent policies of Democratic and Republican administrations alike not to assist China’s military modernization."

According to a report released on Sunday by the non-partisan think tank, the Institute for Progress (IFP), the H200 would be almost six times as powerful as the H20, the most advanced AI semiconductor that can legally be exported to China, after the Trump administration reversed its short-lived ban on such sales this year.

The Blackwell chip now in use by US AI firms is about 1.5 times faster than H200 chips for training AI systems, the IFP said, and five times faster for inferencing work where AI models are put to use. Nvidia's own research has suggested Blackwell chips are 10 times faster than H200 chips for some tasks.

Several Democratic US senators in a statement described Trump's decision as a "colossal economic and national security failure" that would be a boon to China's industry and military.

Republican Representative John Moolenaar, who chairs the House China Select Committee, said in a statement to Reuters that China would use the chips to strengthen its military capabilities and surveillance.

"Nvidia should be under no illusions - China will rip off its technology, mass-produce it themselves and seek to end Nvidia as a competitor," he said.

CHINA EYES POTENTIAL SECURITY RISKS

The approval, however, comes as China is strengthening its resolve to wean the country off its reliance on Nvidia's chips. China's cyberspace regulator in July also accused Nvidia's H20 chips of potentially carrying backdoor security risks, an allegation Nvidia has denied.

In recent months, Beijing has cautioned Chinese tech companies against buying chips that Nvidia downgraded to sell to the Chinese market, which are the H20, RTX 6000D and L20, two sources said.

"Chinese firms want H200s, but the Chinese state is driven by paranoia and pride," said Craig Singleton, a senior fellow at the Washington think tank Foundation for Defense of Democracies. "Washington may approve the chips, but Beijing still has to let them in."

The H200 change of stance comes the same day that Trump's Justice Department announced it had cracked a China-linked chip smuggling ring that in late 2024 and early 2025 exported and attempted to export at least $160 million worth of controlled Nvidia H100 and H200 chips.

Chris McGuire, an expert on technology and national security who served at the US State Department until this summer, said Chinese firms would likely still buy H200s, given that the chip "is better than every chip the Chinese can make."

China's domestic AI chip companies now include tech giant Huawei Technologies, which in September released a three-year product roadmap, as well as smaller players such as Cambricon and Moore Threads.

China's SSE STAR Chip Index and the CSI Semiconductor Industry Index both dropped more than 1% at market open on Tuesday but soon recovered most of the losses.



Trump Says Will Not Tolerate Attempts to Slow AI Growth

An AI (Artificial Intelligence) sign is seen at the World Artificial Intelligence Conference (WAIC) in Shanghai, China July 6, 2023. (Reuters)
An AI (Artificial Intelligence) sign is seen at the World Artificial Intelligence Conference (WAIC) in Shanghai, China July 6, 2023. (Reuters)
TT

Trump Says Will Not Tolerate Attempts to Slow AI Growth

An AI (Artificial Intelligence) sign is seen at the World Artificial Intelligence Conference (WAIC) in Shanghai, China July 6, 2023. (Reuters)
An AI (Artificial Intelligence) sign is seen at the World Artificial Intelligence Conference (WAIC) in Shanghai, China July 6, 2023. (Reuters)

US President Donald Trump on Saturday lashed out at critics of artificial intelligence and data centers, saying he would not allow the "destruction" of the industry and announcing the creation of a new government "AI Force."

Trump -- who finds himself out of step with growing public skepticism over AI and dire safety warnings from AI industry leaders -- said liberals want the "decimation, or destruction, of AI."

"I, as President of the United States, will not stand by and let this happen," he wrote on his Truth Social platform.

"We will not in any way hinder or stifle the Growth of this incredible Industry. Rather, we will cherish it, help it, and watch over it, as it grows!"

Trump added, without explaining, that he was creating an "AI Force" that would be tasked with "looking for BAD" actors, and would soon announce an AI coordinator or "Czar."

Trump's latest full-throated defense of artificial intelligence came even though polls show Americans are largely opposed to allowing the energy-guzzling data centers that are the engines of AI to be built in their communities.

As the war against Iran drags on and Americans grapple with the inflation it has fueled, Trump's approval rating is low.

Surveys suggest his Republican Party could lose control of the House and perhaps the Senate in November's midterm elections.

Trump is also firmly embracing the technology despite widespread warnings, even from top industry leaders, that AI development is going too fast and the technology could soon threaten our survival as ever-smarter AI agents escape the control of their human designers.

Jacob Coxon, a former AI researcher who worked at Anthropic before resigning, set off the latest round of alarm by describing how those building the technology "earnestly believe that it could kill us all by the end of the decade."

Trump has repeatedly dismissed all the doomsday talk, calling it a "hoax" -- a term he often uses for crises or issues he cannot control.

The president said the United States could not afford to lose a technological race with rival China and he insisted the only guardrails necessary were a "STRONG AND SMART (High IQ!) PRESIDENT, and the U.S.A. has that, in spades!"

In another social media post on Saturday, Trump mused on whether artificial intelligence is an accurate term, offered three alternatives in what he called a poll, and asked people to weigh in.

The three are superior intelligence, extreme intelligence and supreme intelligence, Trump wrote.

"This is a Poll, and I would appreciate everybody voting! Which is the best name for this ever growing "Revolution?"


Google's Gemini AI Carried Out Cyberattacks, Guessed Passwords

Google's AI model 'Gemini' (Reuters)
Google's AI model 'Gemini' (Reuters)
TT

Google's Gemini AI Carried Out Cyberattacks, Guessed Passwords

Google's AI model 'Gemini' (Reuters)
Google's AI model 'Gemini' (Reuters)

Google's consumer AI model Gemini hacked multiple systems through guessing login credentials, the company told AFP on Friday, the latest case of rogue AI cybersecurity transgressions, which have generated safety concerns.

The hacks, first reported by the Wall Street Journal, took place in May and were discovered by Google in July.

"In a standard evaluation, the model found public information online and guessed credentials to access websites it thought were part of the test," Heather Adkins, Google's vice president of security engineering, told AFP in a statement.

"In all three of these instances, the model stopped," Adkins added, without specifying which organizations were breached.

"We ensured the three entities were made aware, and we worked with our training partner on the changes they've now made to their testing processes."

In July, two OpenAI models escaped the closed environment they were meant to stay in, got onto the internet on their own and broke into the internal systems AI platform Hugging Face.

The incident fed worries that AI giants cannot keep their own models under control, as similar episodes have been reported at Anthropic and China's Moonshot AI.

"These events highlight the importance of training powerful AI models to act responsibly," Adkins emphasized.


Founder of Chinese Startup Spirit AI Says Robot Brains Set for 2027 Breakthrough

FILE PHOTO: The words "AI Artificial Intelligence," a keyboard and a robotic hand are shown in this illustration created on June 5, 2026. REUTERS/Dado Ruvic/Illustration/File Photo
FILE PHOTO: The words "AI Artificial Intelligence," a keyboard and a robotic hand are shown in this illustration created on June 5, 2026. REUTERS/Dado Ruvic/Illustration/File Photo
TT

Founder of Chinese Startup Spirit AI Says Robot Brains Set for 2027 Breakthrough

FILE PHOTO: The words "AI Artificial Intelligence," a keyboard and a robotic hand are shown in this illustration created on June 5, 2026. REUTERS/Dado Ruvic/Illustration/File Photo
FILE PHOTO: The words "AI Artificial Intelligence," a keyboard and a robotic hand are shown in this illustration created on June 5, 2026. REUTERS/Dado Ruvic/Illustration/File Photo

Humanoid robot brains are likely to achieve a breakthrough as soon as mid-2027, but their deployment in homes could take at least eight years as model development faces a data bottleneck, leading Chinese embodied AI firm Spirit AI said this week.

Despite impressive hardware advances allowing Chinese humanoids to sprint, dance and do backflips on command, Chinese robot firms are increasingly focusing on the software that determines robots' intelligence and their economic productivity in real-world settings.

"The brain is indeed the weakest link in the complete robotics stack," Gao Yang, co-founder and chief scientist of Spirit AI, told Reuters at its Beijing offices on Thursday.

An industry breakthrough comparable to ⁠OpenAI's landmark GPT-3.0 model, ⁠which powered ChatGPT, could come around mid-2027, said Gao. Spirit AI's robots have achieved a 90% success rate for simple tasks in structured living-room environments.

"The next one to two years mark the initial window for industrial applications. Two years from now, we'll see robots deployed in commercial service settings doing simpler tasks. Entering homes is far harder than both," said Gao, who is also an assistant professor of robotics at Tsinghua University.

Spirit AI ⁠currently has tens of its own Moz1 wheeled humanoid robots deployed on production lines at battery maker CATL and retailer JD.com, which is also an investor.

The 300-person startup has raised over $670 million since its 2024 founding, making it one of China's most rapidly capitalized embodied intelligence firms. It is currently valued at 20 billion yuan ($2.9 billion). Gao declined to comment on any plans for an initial public offering.

"Progress is extremely fast. When Spirit AI was founded, a robot could perform only one isolated task well, like pouring water or folding a piece of clothing," said Gao.

"Today, robots operate across large spatial areas and execute continuous complex workflows."

Difficulties remain in perfecting fine-motor actions like unscrewing a bottle cap ⁠and dealing with ⁠unseen tasks, he said. Spirit AI overwhelmingly relies on real-world data to train its robot brains instead of virtual simulations, which many competitors use to reduce model training costs.

"Simulators handle rigid bodies well, but flexible objects like deformable electric cables remain a problem," said Gao.

The company employs around 1,000 contractors nationwide using wearable data-collection equipment in households and on production lines.

Its Beijing offices include a robot data training center where a Reuters witness saw dozens of young people fitted with sensors who repeated motions like opening fridges, unlocking safes, and cutting vegetables with knives to train the humanoid robots.

In other robot-training facilities in China, operators may need to repeat a movement more than 50 times to get one "clean" movement with the required precision.

However, Spirit AI found that using "dirty data" with a more diverse range of motions enabled its models to improve faster, Gao said.