loading
loading
Illegal Nvidia Blackwell Ultra B300 GPUs found inside state-owned China Telecom's network — Nvidia says chip smuggling is 'a drop in the bucket,' insists it remains a non-factor A security firm was able to spot 32 of the high-powered chips within the state-owned telecom company's network, but it can't pinpoint where they are physically located. Lava, an Israeli security firm, says that it has been able to identify 32 Nvidia Blackwell Ultra B300 chips within the network of China Telecom, a state-owned company that provides telecommunications services within the country. These same chips have already been barred from export by U.S. export controls, a report by The Information notes, which begs the question of how they have turned up in a Chinese data center. Such chips can only be sold to whitelisted buyers who agree not to export them to sanctioned parties, including those in China. However, banning companies from buying Nvidia's high-performance GPUs and AI data center chips hasn't always worked. In 2025, it was reported that Chinese companies smuggled $1 billion worth of Nvidia AI chips into the country in three months. At the time, the B200 chips were the most sought-after, but some distributors were already gearing up to sell B300-series chips when they became available. While the latest finding shows that the proliferation of export-controlled chips into China continues to be an issue, Nvidia has downplayed the report but didn't deny the findings. "Governments, researchers, and media have all found insignificant diversion of controlled US products, less than one half of one percent in recent court cases — a drop in the bucket compared to the ocean of domestic compute China already has," an Nvidia spokesperson told Tom's Hardware. "With tens of millions of GPUs in service in approved countries, this report again demonstrates that smuggling is a nonfactor, blocked by governments on both sides of the globe." It's unknown how B300 chips found their way into China Telecom's server racks, nor can Lava confirm exactly where within the company's network they are located. But they are there, with Lava identifying them via Nvidia's own DCGM Exporter tool. The software can be installed by data center operators and is used to monitor server health and performance, and while designed for internal use, it can oftentimes be accessed beyond the data center's network. Oddly, it doesn't always require password authentication, either. Nvidia previously confirmed it was working on a tool that would allow its chips to be located, but it is yet to be released. It's easy to see why Chinese firms are keen to get their hands on the Blackwell Ultra B300. Announced in March of last year, the data center GPU is one of the most powerful on the planet, offering huge performance gains with 50% more memory and FP4 compute than the previous-generation B200. At the time, Nvidia touted that the chip was "built for the age of reasoning," while name-checking more sophisticated AI LLMs like DeepSeek R1. At the time of its announcement, the B300 was expected to ship by the end of 2025, and it didn't take long for it to become hot property. In August 2026, nine people were indicted in Taiwan over the illegal export of B300 GPUs to China. This came just days after Supermicro confirmed that it had fired several employees following an investigation into smuggling. The complicated smuggling process included the use of a Taiwanese server trading firm called Flying Tiget Tech — a firm with whitelisted status for GPU purchases. Nvidia's chips have proven so vital to AI data centers that demand for even its older silicon has caused skyrocketing prices. The company's A100 servers have tripled in price, despite being five years old, and can cost up to $82.,000. The ongoing crackdown on Nvidia chip smuggling has even seen some resort to retooling gaming GPUs to run AI workloads where possible. Follow Tom's Hardware on Google News, or add us as a preferred source, to get our latest news, analysis, & reviews in your feeds. Get Tom's Hardware's best news and in-depth reviews, straight to your inbox. Oliver Haslam has written about technology of all shapes and sizes for over 15 years, both online and in print. He's fascinated by how personal computing continues to change as tower PCs give way to foldable phones and beyond.
Illegal Nvidia Blackwell Ultra B300 GPUs found inside state-owned China Telecom's network — Nvidia says chip smuggling is 'a drop in the bucket,' insists it remains a non-factor A security firm was able to spot 32 of the high-powered chips within the state-owned telecom company's network, but it can't pinpoint where they are physically located. Lava, an Israeli security firm, says that it has been able to identify 32 Nvidia Blackwell Ultra B300 chips within the network of China Telecom, a state-owned company that provides telecommunications services within the country. These same chips have already been barred from export by U.S. export controls, a report by The Information notes, which begs the question of how they have turned up in a Chinese data center. Such chips can only be sold to whitelisted buyers who agree not to export them to sanctioned parties, including those in China. However, banning companies from buying Nvidia's high-performance GPUs and AI data center chips hasn't always worked. In 2025, it was reported that Chinese companies smuggled $1 billion worth of Nvidia AI chips into the country in three months. At the time, the B200 chips were the most sought-after, but some distributors were already gearing up to sell B300-series chips when they became available. While the latest finding shows that the proliferation of export-controlled chips into China continues to be an issue, Nvidia has downplayed the report but didn't deny the findings. "Governments, researchers, and media have all found insignificant diversion of controlled US products, less than one half of one percent in recent court cases — a drop in the bucket compared to the ocean of domestic compute China already has," an Nvidia spokesperson told Tom's Hardware. "With tens of millions of GPUs in service in approved countries, this report again demonstrates that smuggling is a nonfactor, blocked by governments on both sides of the globe." It's unknown how B300 chips found their way into China Telecom's server racks, nor can Lava confirm exactly where within the company's network they are located. But they are there, with Lava identifying them via Nvidia's own DCGM Exporter tool. The software can be installed by data center operators and is used to monitor server health and performance, and while designed for internal use, it can oftentimes be accessed beyond the data center's network. Oddly, it doesn't always require password authentication, either. Nvidia previously confirmed it was working on a tool that would allow its chips to be located, but it is yet to be released. It's easy to see why Chinese firms are keen to get their hands on the Blackwell Ultra B300. Announced in March of last year, the data center GPU is one of the most powerful on the planet, offering huge performance gains with 50% more memory and FP4 compute than the previous-generation B200. At the time, Nvidia touted that the chip was "built for the age of reasoning," while name-checking more sophisticated AI LLMs like DeepSeek R1. At the time of its announcement, the B300 was expected to ship by the end of 2025, and it didn't take long for it to become hot property. In August 2026, nine people were indicted in Taiwan over the illegal export of B300 GPUs to China. This came just days after Supermicro confirmed that it had fired several employees following an investigation into smuggling. The complicated smuggling process included the use of a Taiwanese server trading firm called Flying Tiget Tech — a firm with whitelisted status for GPU purchases. Nvidia's chips have proven so vital to AI data centers that demand for even its older silicon has caused skyrocketing prices. The company's A100 servers have tripled in price, despite being five years old, and can cost up to $82.,000. The ongoing crackdown on Nvidia chip smuggling has even seen some resort to retooling gaming GPUs to run AI workloads where possible. Follow Tom's Hardware on Google News, or add us as a preferred source, to get our latest news, analysis, & reviews in your feeds. Get Tom's Hardware's best news and in-depth reviews, straight to your inbox. Oliver Haslam has written about technology of all shapes and sizes for over 15 years, both online and in print. He's fascinated by how personal computing continues to change as tower PCs give way to foldable phones and beyond. - quorm Good. Glad someone is keeping "Open"AI and Anthropic's proprietary models honest. Every time someone in China releases an open source model with 99% of their performance, it reinforces that their $1 trillion valuations have at least 2 zeroes on the end that shouldn't be there. End this AI bubble already so I can maintain, maybe even upgrade my decrepit hardware.Reply - Vicecrow Fomdoo said:Remind me of which county makes the majority of our tech?Granada for sure. Reply - nrdwka well, US sanction are US problem. No-one outside have obligation to respect them. I'd say it is not smuggling if it entered China legally, it is just buying in free market :) Reply - bill001g To a point this is kinda surprising. You see smuggled GPU stuff all the time but it is mostly in privately owned china companies. The actual industries that are controlled by the government are very suspect of spying by western governments. You have to admit if it possible to use nvidia chips to spy the CIA will do it. China has import bans on many western products. Hard to say how much of this is for show and how much they really enforce. They have many photos of people they caught with cpu taped to the bodies....they seem to ignore the full pallets of gpu for some reason. Reply
document fetched 2026-10-24 12:14 · canonical https://tomshardware.com/tech-industry/data-centers/illegal-nvidia-blackwell-ultra-b300-gpus-found-inside-state-owned-china-telecoms-network-nvidia-says-chip-smuggling-a-drop-in-the-bucket-insists-smuggling-remains-a-nonfactor