Nvidia, the world’s top chipmaker isn’t immune to the unpredictable fallout of global politics. Last week, as the company reported another strong earnings report, CEO Jensen Huang revealed the sobering news of a $4.5 billion write-off for chips that were supposed to be sent to China and now have nowhere to go.
Huang said during the earnings call that, “We are taking a multibillion dollar write off on inventory that cannot be sold or repurposed,” quoted Fortune.
The chips, which have led to the massive loss, known as the H20 chips, were designed by Nvidia specifically for Chinese clients to meet earlier US export restrictions, according to the report. These chips weren’t top-of-the-line, but they were still advanced enough for AI development and, also legal to ship under the Biden administration’s rules, as per Fortune.
Nvidia now faces a significant setback with a $4.5 billion write-off due to US export restrictions, rendering its China-specific H20 chips unusable. Designed to comply with previous regulations, these chips are now banned under new rules, leaving Nvidia unable to repurpose them for other markets.
But things changed after US president Donald Trump took office, as he went a step further in early April and banned exports of even these chips, according to the report.
However, Nvidia’s newest chips have made gains in training large artificial intelligence systems, new data released on Wednesday showed, with the number of chips required to train large language models dropping dramatically.
MLCommons, a nonprofit group that publishes benchmark performance results for AI systems, released new data about chips from Nvidia and Advanced Micro Devices, among others, for training, in which AI systems are fed large amounts of data to learn from.
While much of the stock market’s attention has shifted to a larger market for AI inference, in which AI systems handle questions from users, the number of chips needed to train the systems is still a key competitive concern. China’s DeepSeek claims to create a competitive chatbot using far fewer chips than U.S. rivals.
The results were the first that MLCommons has released about how chips fared at training AI systems such as Llama 3.1 405B, an open-source AI model released by Meta Platforms that has a large enough number of what are known as “parameters” to give an indication of how the chips would perform at some of the most complex training tasks in the world, which can involve trillions of parameters.
Nvidia and its partners were the only entrants that submitted data about training that large model, and the data showed that Nvidia’s new Blackwell chips are, on a per-chip basis, more than twice as fast as the previous generation of Hopper chips.
About the Author
