China's military tried to buy Nvidia chips for AI servers and a robot dog, documents show
BI reviewed records on the Chinese People's Liberation Army (PLA)'s official procurement portal, where military units post thousands of requests for equipment that local contractors can bid on.
Multiple requests over the past year reference both banned and permitted Nvidia hardware, including the H20 chip, on which the US is set to loosen export controls—a move that has sent Nvidia's stock soaring.
An Nvidia spokesperson told BI that China has "more than enough" domestic chips for all its military applications.
"Buying a handful of older products to test the US competition is not a national security concern," the spokesperson said. "Using restricted products for military applications would be a nonstarter, without support, software, or maintenance."
While the Chinese military tries to buy Nvidia's hardware, the chipmaker faces pressure from other parts of the Chinese government, which has long supported efforts to build a domestic alternative.
On Thursday, China's internet regulator said it had summoned the company to explain alleged backdoor security risks in its H20 chips, claims that Nvidia has denied, Reuters reported.
How the Chinese army wants to use Nvidia chips
The military projects reviewed by BI are typically for servers to run AI models, like those built by DeepSeek, for unspecified purposes.
Three requests published this April call for banned Nvidia chips to power such servers. One requests at least eight H20 cards for a system capable of running one of DeepSeek's largest and most powerful models, DeepSeek-R1 671B, at full capacity.
Another, tied to an "intelligent decision-making" support system that can also run DeepSeek, calls for four RTX 6000 graphics cards—also banned. Although the contract hasn't been finalized, a supplier was provisionally selected in June. DeepSeek didn't respond to requests for comment.
Another server request from the same month asks for an H100 graphics card— banned from export to China since 2022—which can cost tens of thousands of dollars apiece. The listing specifies that the cards must be delivered in their original packaging and installed on-site. Another AI server request published in July 2024 also calls for H100 GPUs.
The Chinese military isn't only calling for AI servers.
One project from April, for example, shows a military unit requesting a 33-pound "robot dog" with an Nvidia Jetson computing module as part of an unspecified training project. This particular request was later canceled, and Jetson modules are not barred from being exported to China for most use cases.
It's unclear whether the Chinese military actually obtained these chips. Ryan Fedasiuk, a former State Department advisor on China tech policy, says the Chinese military has many ways to acquire them.
"There are ample cutouts, subsidiaries, and shell companies that the PLA can and does use to source chips illicitly from American companies, including Nvidia," he told BI.
Nvidia plans to sell H20 chips to China again
Some national security policymakers and professionals have expressed concerns about making it easier for China to get some Nvidia chips.
Twenty of them sent a letter to the Commerce Department on Monday asking the US to reimpose the H20 ban, saying the chips would "support China's military."
"This reversal will certainly open the floodgates," said Craig Singleton, a senior director at the Foundation for Defense of Democracies who co-signed the letter.
The Commerce Department told BI that the current administration has tightened controls compared to its predecessor.
"The Trump Administration will consider any H20 license applications carefully, accounting for both the benefits and the costs of potential exports from America and taking into account the views of experts across the US Government," a Commerce spokesperson said.
'Nobody can beat Nvidia'
Fedasiuk reviewed some of the records found by BI and said they show that the Chinese military, like Chinese AI companies, wants to use the best hardware possible, and domestic chips like Huawei's don't cut it.
"In terms of sheer processing power that a given chip is capable of bringing to bear, nobody can beat Nvidia. Huawei is not close," Fedasiuk said.
Nvidia CEO Jensen Huang has downplayed such concerns, saying China's military doesn't want to rely on US technology.
"They simply can't rely on it," Huang said in July on CNN.
"It could, of course, be limited at any time."
Try Our AI Features
Explore what Daily8 AI can do for you:
Comments
No comments yet...
Related Articles
The Hill
16 minutes ago
- The Hill
US deadlines in Ukraine are a gift to Putin and Xi
President Trump's announcement this week of a shortened window of '10 to 12 days' for Russian President Vladimir Putin to reach a ceasefire agreement in Ukraine reflects a continued evolution in his rhetoric. His growing frustration with Moscow and his willingness to speak plainly about Russia's escalation send a signal that many in the U.S. and Europe have been waiting to hear. But while the shift in tone signals growing frustration, it has not translated into action. Russia reads the action as a continued pause in pressure, which it has used to intensify its offensive against Ukrainian homes and hospitals. Russian forces are now making their fastest territorial gains in more than a year, and their attacks are becoming more sophisticated. Swarm tactics using Iranian-designed Shahed drones, now mass-produced and adapted inside Russia with Chinese parts, are overwhelming Ukraine's air defenses at an alarming rate. In just one day last month, Russia launched 728 drones, decoys and missiles in a single coordinated wave. Ukrainian interceptors and radar crews are doing heroic work, but they are stretched to the limit. The U.S. has tools at its disposal that remain unused. For months, a bipartisan sanctions bill, co-authored by Sens. Lindsey Graham (R-S.C.) and Richard Blumenthal (D-Conn.) and backed by 85 senators, a veto-proof majority, has been ready to move. The legislation would impose steep secondary tariffs on countries like China, India and Brazil that continue to buy Russian oil and gas, and would significantly raise the cost of doing business with Moscow. But in July, Senate leadership pulled the bill from consideration after President Trump suggested he would act if Russia failed to move toward peace within 50 days. Senate Majority Leader John Thune (R-S.D.) said he would 'hold off' on advancing the bill, signaling that Congress would defer to Trump's timeline. House leaders followed suit. That decision was a mistake. While it is encouraging to see President Trump express increasing resolve, deferring congressional action in the hope that Putin will suddenly negotiate has only given Moscow more time and space to escalate. Every week of delay is a missed opportunity to tighten the financial pressure on Putin's war machine. And the clock is not just ticking in Ukraine. The broader contest involves China, too. Beijing's role in this war has become increasingly visible. Chinese companies are supplying entire weapons systems, not just components. Chinese-made drones and decoys are helping Russia saturate Ukrainian airspace. Chinese officials have even welcomed delegations from occupied Ukrainian territories and continue to sell heavy machinery to companies operating there. European officials report that China's foreign minister recently told the EU that Beijing does not want Russia to lose the war and fears that a Russian defeat would allow the U.S. to focus more squarely on Asia. Ukraine has responded accordingly. In early July, Kyiv arrested two Chinese nationals on espionage charges after they allegedly attempted to steal information about Ukraine's Neptune missile program. Days earlier, President Volodymyr Zelensky imposed sanctions on five Chinese firms accused of supporting the Russian war effort. These are not symbolic gestures, they are signs that Ukraine is increasingly realistic about the stakes and about China's alignment with Moscow. Support for Ukraine is not a distraction from U.S. competition with China. It is a critical part of it. Weakening Putin's military capacity weakens a key pillar of China's global strategy. And allowing Russia to continue its aggression without consequence would embolden Beijing's worst instincts from the Taiwan Strait to the South China Sea. To its credit, the Trump administration has begun voicing stronger concerns about Beijing's role. In the recently concluded round of trade talks, senior U.S. officials reportedly raised objections to China's purchase of sanctioned Russian oil and its sale of more than $15 billion worth of dual-use technology to Moscow. These are important warnings — but without follow-through, they risk being absorbed into the pattern of delay that Moscow and Beijing are already exploiting. The Graham-Blumenthal sanctions bill should move forward. It represents the most serious effort yet to impose real costs not only on Russia, but on the network of countries (especially China) helping it survive sanctions. It complements, rather than competes with, the administration's efforts to pressure Moscow. And it sends a message that the U.S. is serious about backing up its warnings with action. Countdowns can be useful. They create urgency. But urgency without follow-through is no substitute for strategy. What matters now is not how many days remain on the clock, but whether we are using each one to act. Jane Harman is a former nine-term congresswoman from California and former ranking member of the House Intelligence Committee, who most recently served as chair of the Commission on the National Defense Strategy. She is the author of 'Insanity Defense: Why Our Failure to Confront Hard National Security Problems Makes Us Less Safe.'
Yahoo
3 hours ago
- Yahoo
If You'd Invested $3,000 in Nvidia (NVDA) Stock 20 Years Ago, Here's How Much You'd Have Today
Key Points The answer may make you want to kick yourself. Hindsight is 20-20, and few back then expected Nvidia to grow so quickly. You still may do well investing in the company now. 10 stocks we like better than Nvidia › Here's a question and answer that might make you kick yourself: If you'd invested $3,000 in shares of Nvidia (NASDAQ: NVDA) 20 years ago, what would it be worth today? The answer: $2.3 million. (It would be even better if you had reinvested your dividends in more shares of Nvidia along the way. Your stake would be worth around $2.5 million.) That's an average annual gain of 39.5%! The S&P 500 averaged a solid 9.22% in the same period. Don't be too hard on yourself if you missed the monster growth, though. Ask 100 people, and you may not find one who invested in Nvidia back in 2005 and held on. Holding on to great companies for many years, if not decades, is one of the best ways to build wealth, but it's easier said than done. For one thing, it's not always clear which companies will become long-term winners, and even some extremely promising companies fall on hard times occasionally, with their stock sinking. It can be hard psychologically to not sell shares `at those times. For a long time, Nvidia was a semiconductor company specializing in chips for gaming. It was very successful at that, but its explosive growth in recent years is largely due to its dominance in chips for data centers, which are in high demand due to artificial intelligence (AI) computing activities. Too late to buy? While it's too late to buy shares of Nvidia in 2005, it's not too late to buy shares in 2025, and they don't look terribly overvalued at recent levels, either -- despite the stock hitting an all-time high. Nvidia's recent forward price-to-earnings ratio (P/E) of 38 is roughly on par with its five-year average of 39. In its last quarter, Nvidia's revenue popped by 69%, with double-digit gains expected in the quarters to come. If you expect the use of AI to increase in the near future along with more demand for data centers and the chips on which they run, take a closer look at Nvidia. Do the experts think Nvidia is a buy right now? The Motley Fool's expert analyst team, drawing on years of investing experience and deep analysis of thousands of stocks, leverages our proprietary Moneyball AI investing database to uncover top opportunities. They've just revealed their to buy now — did Nvidia make the list? When our Stock Advisor analyst team has a stock recommendation, it can pay to listen. After all, Stock Advisor's total average return is up 1,036% vs. just 181% for the S&P — that is beating the market by 855.09%!* Imagine if you were a Stock Advisor member when Netflix made this list on December 17, 2004... if you invested $1,000 at the time of our recommendation, you'd have $625,254!* Or when Nvidia made this list on April 15, 2005... if you invested $1,000 at the time of our recommendation, you'd have $1,090,257!* The 10 stocks that made the cut could produce monster returns in the coming years. Don't miss out on the latest top 10 list, available when you join Stock Advisor. See the 10 stocks » *Stock Advisor returns as of July 29, 2025 Selena Maranjian has positions in Nvidia. The Motley Fool has positions in and recommends Nvidia. The Motley Fool has a disclosure policy. If You'd Invested $3,000 in Nvidia (NVDA) Stock 20 Years Ago, Here's How Much You'd Have Today was originally published by The Motley Fool
Time Magazine
4 hours ago
- Time Magazine
Trump's Decision to Fire BLS Chief Echoes Putin's Strategies
President Donald Trump's firing of the Commissioner of the Bureau of Labor Statistics (BLS) on Friday afternoon just after she delivered a negative jobs report echoes the impulse of many leaders to shoot the messenger. Trump declared, 'I've had issues with the numbers for a long time. We're doing so well. I believe the numbers were phony like they were before the election and there were other times. So I fired her, and I did the right thing.' While Trump may or may not be friends with Vladimir Putin, he is clearly following the Russian President's HR staffing guidelines to eliminate lieutenants who bring bad news. As we've documented before, the Federal State Statistics Service (Rosstat) has a long history of manipulating official economic statistics to please Putin, 'bending over backward to correct bad numbers and burying unflattering statistics' under the pressure the Kremlin has exerted to corrupt statistical integrity, especially since Putin's invasion of Ukraine in 2022. The reliability of official statistics from China has also been brought into question, leading analysts to rely on a wide range of unofficial or proxy indicators to gauge the true state of the Chinese economy. Even China's former Premier, the late Li Keqiang, reportedly confided that he didn't trust official GDP numbers. Read More: What to Know About the Jobs Report That Led Trump to Fire the Labor Statistics Chief Like other strongmen, Trump has repeatedly shown a pattern of manipulating data to suit his preferred narrative. Trump's surprise firing of BLS Commissioner Erika McEntarfer has quickly caught the attention of technical market analysts and economists on both sides of the political spectrum. One side cheers the push to disrupt a slow, bureaucratic federal agency. The other side shouts in dismay over concerns about yet another example of Trump politicizing an apolitical institution. Both responses are warranted. The accuracy of BLS data has long been questioned as major revisions only come in months later. To their credit, the BLS, in addition to other statistical agencies, has publicly recognized a need to modernize its methodology. Unfortunately, though, the severity of job revisions has worsened since the COVID-19 era, with no successful program to address the issue. The downward revision on Friday of more than 250,000 jobs marked the most significant adjustment since the depths of the pandemic. However, Trump's accusations against the BLS of rigging the job numbers to make him and the Republican base look bad, and his subsequent firing of McEntarfer based on a belief that BLS revisions were politically motivated, are yet another step closer to authoritarianism. Introducing his latest conspiracy theory, the President went even further by suggesting McEntarfer, whose career spans two decades across Republican and Democratic Administrations, rigged the numbers 'around the 2024 presidential election' in then-Vice President Kamala Harris' favor. Trump conveniently fails to mention that his definition of 'around' was back in August 2024. Recall, the 2024 presidential election was a full three months later in November. Revisions are not unusual behavior by the BLS. They are a critical part of the natural process for developing an accurate picture of the largest, most dynamic economy in the world. The average size of job revisions since 2003 is not insignificant at 51,000 jobs. And, despite what Trump may want Americans to believe, his tariff policies have created an unprecedented level of uncertainty in the U.S. economy, comparable only to that of 2020, with many economists expecting a recession to follow as a result. Bloomberg reporting has pointed to a possible connection between the severity of negative job revisions and recessionary economic environments. The BLS has also been subjected to DOGE-led hiring constraints and other resource rescissions. In addition, the Trump Administration's disbanding of the Federal Statistics Advisory Committee in March both eliminated one of the main engines for enhancing agency performance and, perhaps, in what should have been a concerning harbinger, abolished the canary in the data integrity coal mine. Complaints about BLS methods are legitimate, like the reliance on enumerators over scanner data, and deserve attention, but this is not how to fix it. Read More: What Trump's Win Means for the Economy This is far from the first time Trump has subordinated statistical integrity to political theater. From crowd sizes to weather forecasts, vote counts to tariff formulas, Trump has discarded facts for fictions that play to his political favor. Trump doesn't just bend the truth—he twists the numbers until they resemble propaganda and then silences those who disagree. As CBS News titan Edward R. Murrow warned 65 years ago: 'To be persuasive, we must be believable. To be believable, we must be credible. To be credible, we must be truthful.'
