New TSMC AI Chips Rules Bar Chinese Companies from Accessing Advanced AI Chips

Aadil Raval
By Aadil Raval
3 Min Read

Another clause added to TSMC AI Chips restrictions now prevents Chinese companies from sourcing chips required for AI training. It is the latest set of restrictions after Taiwan-based TSMC stopped shipping AI chips to companies in China following guidance from the U.S. government.

TSMC AI Chips Restrictions Will Probably Cause Ripple Effect in China

According to the latest piece from Taiwan Economic Daily, the Taiwan Semiconductor Manufacturing Company (TSMC) has added new restrictions. It now prevents Chinese companies and Chinese IC design makers from sourcing powerful silicon components such as GPUs, required for AI training. The new TSMC AI Chips restrictions reportedly kicked off in effect on January 31st and targeted fabless Chinese chip designers from ordering chips fabbed on 16nm and lower nodes. 

Apple and TSMC met in secret to reserve 2nm production for new AI chips, the A-series, and the M-series. 
Latest TSMC restrictions kicks in
Image Credits: India Times

Chinese customers can still get their hands on AI chips fabbed at 16nm or lower subject they obtain from any U.S.-whitelisted Outsourced Semiconductor Assembly and Test (OSATs). The current generation of chips may have tens of billions of transistors such as the Nvidia H800 GPUs used to train DeepSeek-R1 has 80 million transistors. However, the latest restrictions will prohibit Chinese customers from sourcing chips carrying 30 billion transistors or more (i.e. 16nm or lower node).

TSMC Is Being Extra Cautious To Be In Good Books Of the US

Apparently, a few incidents happened before these restrictions were imposed. In fact, TSMC is being cautious not to violate any orders coming from the US government. One of the recent incidences occurred when TSMC-fabbed chips landed on Huawei’s AI processors. Another one was with DeepSeek-R1 which was reportedly trained on over 2,048 units of Nvidia H800 GPUs. These GPUs are fabbed on 4nm nodes and have 80 billion transistors making it one of the most powerful chips out there.

It remains under investigation how DeepSeek was able to source these many GPUs even with the restrictions in effect. They reportedly used a Singapore-based intermediary to source the chips without a proper license. Both the FBI and the White House are currently investigating the incident and how it could have been pulled off.

When asked, US-based chip giant Nvidia said they don’t believe DeepSeek has broken the law, however, it believes the Chinese AI giant should have obtained the GPUs by acting within the law.

Chinese companies will now have to get their licenses from BIS-approved OSATs to get the shipments of chips fabbed at 16nm or lower nodes. These measures are in place to ensure transparency in the semiconductor industry given the fact that Taiwan holds superior chip volumes of mature nodes.

Share This Article
Follow:
A wordsmith, a kin tech observer, a sci-fi fanatic and a scientific documentary buff.
Leave a Comment

Another clause added to TSMC AI Chips restrictions now prevents Chinese companies from sourcing chips required for AI training. It is the latest set of restrictions after Taiwan-based TSMC stopped shipping AI chips to companies in China following guidance from the U.S. government.

TSMC AI Chips Restrictions Will Probably Cause Ripple Effect in China

According to the latest piece from Taiwan Economic Daily, the Taiwan Semiconductor Manufacturing Company (TSMC) has added new restrictions. It now prevents Chinese companies and Chinese IC design makers from sourcing powerful silicon components such as GPUs, required for AI training. The new TSMC AI Chips restrictions reportedly kicked off in effect on January 31st and targeted fabless Chinese chip designers from ordering chips fabbed on 16nm and lower nodes. 

Apple and TSMC met in secret to reserve 2nm production for new AI chips, the A-series, and the M-series. 
Latest TSMC restrictions kicks in
Image Credits: India Times

Chinese customers can still get their hands on AI chips fabbed at 16nm or lower subject they obtain from any U.S.-whitelisted Outsourced Semiconductor Assembly and Test (OSATs). The current generation of chips may have tens of billions of transistors such as the Nvidia H800 GPUs used to train DeepSeek-R1 has 80 million transistors. However, the latest restrictions will prohibit Chinese customers from sourcing chips carrying 30 billion transistors or more (i.e. 16nm or lower node).

TSMC Is Being Extra Cautious To Be In Good Books Of the US

Apparently, a few incidents happened before these restrictions were imposed. In fact, TSMC is being cautious not to violate any orders coming from the US government. One of the recent incidences occurred when TSMC-fabbed chips landed on Huawei’s AI processors. Another one was with DeepSeek-R1 which was reportedly trained on over 2,048 units of Nvidia H800 GPUs. These GPUs are fabbed on 4nm nodes and have 80 billion transistors making it one of the most powerful chips out there.

It remains under investigation how DeepSeek was able to source these many GPUs even with the restrictions in effect. They reportedly used a Singapore-based intermediary to source the chips without a proper license. Both the FBI and the White House are currently investigating the incident and how it could have been pulled off.

When asked, US-based chip giant Nvidia said they don’t believe DeepSeek has broken the law, however, it believes the Chinese AI giant should have obtained the GPUs by acting within the law.

Chinese companies will now have to get their licenses from BIS-approved OSATs to get the shipments of chips fabbed at 16nm or lower nodes. These measures are in place to ensure transparency in the semiconductor industry given the fact that Taiwan holds superior chip volumes of mature nodes.

Share This Article
Follow:
A wordsmith, a kin tech observer, a sci-fi fanatic and a scientific documentary buff.
Leave a Comment