Cold shoulder: Why Beijing is freezing Nvidia’s access to the Chinese market
The Nvidia logo appears on a smartphone reflecting the flags of China and the U.S.
Nurphoto | Nurphoto | Getty Images
Beijing has reportedly halted purchases of yet another AI chip from Nvidia, freezing it out of the market completely — a move industry experts say reflects the country’s growing confidence in domestic chip makers and an attempt at gaining trade leverage.
It was only a few months ago when Jensen Huang announced, from China, that the U.S. would allow it to resume sales of its made-for-China H20 graphics processing units, reversing a previous halt on their exports.
At the time, Huang had also revealed the company’s new RTX Pro GPU for the Chinese market, which had been tailored for AI smart factories and logistics.
But Nvidia’s fortunes flipped in August, when it was reported that regulators in China had begun mandating tech firms to halt purchases of Nvidia’s H20s pending a national security review.
Now that the mandate has been expanded to Nvidia’s RTX Pro 6000D chip, rendering the company unable to sell any products to Chinese customers, according to a report by the Financial Times on Wednesday.
That comes after Chinese regulators on Monday said that Nvidia had violated the country’s anti-monopoly law, as per a preliminary probe, adding they would continue their investigation.
While the exact motives of China’s actions against Nvidia remain unclear, tech and geopolitical analysts say the developments show China has become more confident in its own ability to make AI chips and is wielding that as a leverage against the U.S.
Another Nvidia chip crumbles
The reported reasons for Chinese regulators’ intervention in the H20 had been the need for a national security review over concerns that Nvidia chips could be outfitted with certain tracking systems — an idea proposed by American lawmakers.
Experts had characterized the move as part of Beijing’s efforts aimed at encouraging Chinese AI companies to explore domestic alternatives, though they forecast that exports would eventually be cleared due to high demand from Chinese AI players.
Meanwhile, some Chinese AI companies had indicated they would order tens of thousands of the RTX Pro 6000D, and had started testing and verification work with Nvidia’s server suppliers up until they were asked to cease such activities, according to FT’s reporting.
The country’s regulators, however, blocked access to those Nvidia chips after summoning domestic AI chip makers and concluding they had reached performance comparable to the U.S. company’s made-for-China products, according to the FT.
However, performance isn’t the only challenge facing China’s AI chips. Analysts contend that capacity is also a major barrier, with the domestic industry still unable to produce enough chips at scale.
Reporting from the FT suggests Beijing has also become more confident in this area, with local chipmakers seeking to triple the country’s total output of AI processors next year.
“All these recent actions show that China has much more confidence in their domestic sector than they used to,” said Qingyuan Lin, a senior analyst covering China semiconductors at Bernstein.
China’s chip progress
There are signs that China’s AI ecosystem has been progressing.
Chinese tech giant Huawei announced Thursday new AI compute infrastructure using its in-house Ascend chips, claiming they would be the “world’s most powerful.”
Research firm SemiAnalysis found in April that Huawei’s latest-generation CloudMatrix system was able to outperform Nvidia’s competing AI compute system on some metrics — despite each Ascend chip delivering only about one-third the performance of an Nvidia processor. Huawei built its advantage by having five times as many chips linked together.
Meanwhile, Chinese AI start-up DeepSeek had hinted last month that its latest AI model would be compatible with the country’s “next generation” homegrown AI chips. China’s Alibaba and Baidu have reportedly started using internally designed chips to help train their AI models, partly replacing those made by Nvidia.
Still, analysts are skeptical about China’s ability to cut its dependence on Nvidia chips.
“In terms of China’s domestic chip preparedness, I believe it is misleading to suggest the country can advance AI at a current level solely with domestic alternatives and without NVIDIA’s systematic offerings,” Ray Wang, research director for semiconductors, supply chain and emerging technology at Futurum Group, told CNBC.
Seeking leverage?
Experts, while agreeing that China has become more confident in its ability to counter Nvidia, suggest the country could be using access to its market as leverage in trade negotiations with the U.S.
“The move is likely to be a negotiation tactic as part of a broader set of discussions involving other topics including tariffs. China is no doubt trying to encourage and establish silicon self sovereignty, while also getting the best possible chip in the meantime,” AJ Kourabi, analyst at SeminAlysis, told CNBC.
The U.S. has indicated that it might allow even more advanced Nvidia chips than the H20 into the Chinese market.
Under the Joe Biden administration, export controls on advanced chips had been increasingly tightened with the aim of blocking China’s access to the best American technology. That trend after accelerating initially under the Trump administration is now reversing.
According to Reva Goujon, director at Rhodium Group, by rejecting the H20 and RTX Pro, Beijing could be looking to create an opportunity to negotiate access to more advanced GPUs.
She added that it’s likely not a coincidence that it comes amid other leverage-building by China this week, referring to its recent anti-dumping investigation into imports of certain analog chips from the U.S.
“As Beijing tests Trump’s transactionalism, it has to build up leverage of its own,” she said.
Lisa Su, chair and chief executive officer of Advanced Micro Devices Inc. (AMD), during a Bloomberg Television interview in San Francisco, California, US, on Monday, Oct. 6, 2025.
David Paul Morris | Bloomberg | Getty Images
AMD CEO Lisa Su said on Tuesday that the company’s overall revenue growth would expand to about 35% per year over the next three to five years, driven by “insatiable” demand for artificial intelligence chips.
Su said that much of that would be captured by the company’s AI data center business, which it expects to grow at about 80% per year over the same time period, on track to hit tens of billions of dollars of sales by 2027.
“This is what we see as our potential given the customer traction, both with the announced customers, as well as customers that are currently working very closely with us,” Su told analysts.
Ultimately, Su said that AMD could be able to achieve “double-digit” share in the data center AI chip market over the next three to five years.
AMD shares fell 3% in extended trading.
The AI chip market is currently dominated by Nvidia, which has over 90% of the market share, according to some estimates, and which has given the company a market cap of over $4.6 trillion, versus AMD’s roughly $387 billion valuation.
AMD is holding its first financial analyst day since 2022, as the company has found itself at the center of a boom in data center spending for AI.
While companies are spending hundreds of billions of dollars in total on graphics processing unit (GPU) chips to build and power artificial intelligence applications like OpenAI’s ChatGPT, they are also looking for alternatives to increase capacity and control costs. AMD is the only other major developer of GPUs aside from Nvidia.
In October, AMD announced a partnership with OpenAI in which it would sell the AI startup billions of dollars in its Instinct AI chips over multiple years, starting with enough chips in 2026 to use 1 gigawatt of power.
As part of the deal, OpenAI could end up taking a 10% stake in the chipmaker. Su also highlighted long-term deals with Oracle and Meta on Tuesday.
AMD shares have nearly doubled so far in 2025.
OpenAI is also helping AMD set up its next-generation systems based around its Instinct MI400X AI chips, which ship next year.
AMD has said that its chips will be able to be assembled into a “rack-scale” system where 72 of its chips work together as one, which is essential for running the largest AI models.
If AMD succeeds at its rack, it will catch up with Nvidia’s AI chips, which have been offered in rack-scale systems for three product generations.
Su said that the company now sees the total market for AI data center parts and systems hitting $1 trillion per year in 2030, representing 40% annual growth per year. AMD reported $5 billion in AI chip sales in its fiscal 2024.
That’s up from the company’s previous forecast of a $500 billion market in 2028 for AI chips. But the updated AMD figure also includes central processors (CPU), an important kind of chip that sits at the heart of a computer, but isn’t a pure AI accelerator like the GPUs made by Nvidia and AMD.
AMD’s Epyc CPUs are still the company’s most important product by sales. It primarily competes with Intel and some smaller Arm-based processors in the CPU market. AMD also makes chips for game consoles, networking parts, and other devices.
On Tuesday, although AMD focused much of its focus on its growing AI business, it told shareholders that its older businesses were growing too.
“The other message that we want to leave you with today is every other part of our business is firing on all cylinders, and that’s actually a very nice place to be,” Su said.
Technology
CoreWeave CEO responds to data center delays as stock plunges. Core Scientific shares fall
CoreWeave shares sank 13% on Tuesday after CEO Mike Intrator addressed delays at a third-party data center developer that hit full-year guidance in its latest earnings report.
“Quite frankly, every single part of this quarter went exactly as we planned, except for one delay at a singular data center,” Intrator told CNBC’s “Squawk on the Street” on Tuesday.
He then clarified that a “singular data center provider” is more accurate.
“Some people might think it’s one complex, but when I go over the numbers, we’re talking about multiple places,” CNBC’s Jim Cramer said. “And it just so happens that the places are all connected to an outfit called Core Scientific that you tried to buy.”
Cramer noted delays at complexes in Texas, Oklahoma and North Carolina.
Intrator said the companies have been working together on infrastructure for a long time a would continue work to bring it online. He did not directly confirm that Core Scientific is the third-party provider.
CoreWeave tried to acquire Core Scientific for $9 billion earlier this year. Core Scientific shareholders voted against the proposed deal. Core Scientific shares sank 7% Tuesday.
During CoreWeave’s quarterly earnings call on Monday, JPMorgan Securities analyst Mark Murphy asked if the delay was related to Core Scientific, but Intrator declined to name the company. At another point in the call, the CEO suggested that just one data center, not multiple sites, were affected.
“There was a problem at one data center that’s impacting us, but there are 41 data centers in our portfolio,” Intrator said.
At a different point in the call, CoreWeave’s CFO Nitin Agrawal said the delays stem from “a single provider, data center provider partner.”
When reached for comment about how many sites were affected, CoreWeave did not provide a number and pointed to Intrator’s statements on the earnings call and during his “Squawk on the Street” interview.
CoreWeave, which provides infrastructure for artificial intelligence companies, reported third-quarter results on Monday that showed $1.36 billion in revenue for the period, up 134% from $583.9 million a year ago. But CoreWeave now sees 2025 revenue coming in between $5.05 billion and $5.15 billion, below the average analyst estimate of $5.29 billion.
Intrator told CNBC on Tuesday that CoreWeave has teams of employees working with contractors and Core Scientific at those sites “every single day” to get things back on track.
“It became apparent to us in Q3 that there were delays at the facility,” Intrator said. “CoreWeave responded by deploying our own boots on the ground to ensure that everything was being done in order to move those facilities along as quickly as possible.”
Intrator told analysts on Monday that the delays would not affect its backlog or get the full value from contracts.
Core Scientific did not immediately respond to a request for comment.
CoreWeave has been on a deal-making blitz as big tech companies and AI startups race to build out their computing infrastructure.
The company announced in September that it agreed to provide Meta with $14.2 billion of AI cloud infrastructure, just days after expanding its contract with OpenAI to $22.4 billion.
-
Sports2 years agoStory injured on diving stop, exits Red Sox game
-
Sports3 years ago‘Storybook stuff’: Inside the night Bryce Harper sent the Phillies to the World Series
-
Sports2 years agoGame 1 of WS least-watched in recorded history
-
Sports3 years agoButton battles heat exhaustion in NASCAR debut
-
Sports3 years agoMLB Rank 2023: Ranking baseball’s top 100 players
-
Sports4 years ago
Team Europe easily wins 4th straight Laver Cup
-
Environment2 years agoJapan and South Korea have a lot at stake in a free and open South China Sea
-
Environment1 year agoHere are the best electric bikes you can buy at every price level in October 2024