Nvidia might still have a way to sell AI chips in China after H20 ban cost them billions
Table of Contents
At this point, it isn't news that tension is rising between the U.S. and China due to trade bans and tariffs that have been placed on each other. The idea of making products in America might have great long-term benefits, but in the short term, it means that businesses have to take a hit.
Recently, we saw TSMC's plate full with orders from big names like Nvidia, AMD, Apple, and more as the Arizona plant is firing on all pistons. However, there is a report that Nvidia has been given the green light to operate in China, but the product would be downgraded, much like H20 chips, which were eventually restricted and ended up costing Nvidia at least $4.5 billion.
Rumor: Nvidia might be selling downgraded chips to China, again
In a post by Jukanlosreve on X, they stated that Nvidia has been allowed to sell new downgraded AI chips, labeled B30, to China. Companies like Alibaba, Tencent, and ByteDance have allegedly placed orders, with Gigabyte reportedly ordering the most as it expects to raise its finances during an upcoming earnings call.
According to the rumor, the B30 chip isn't downgraded in terms of architecture, as it is based on the Blackwell architecture, but it actively avoids using high-bandwidth memory and TSMC's advanced packaging technology.
The advanced packaging technology is a process that combines multiple semiconductor chips on a single package. This leads to increased computing power, performance, and efficiency. So, taking it out of the equation, it wouldn't be surprising if the B30 couldn't keep up with chips that have this technology.
Especially in the AI race, where every company is racing to develop their models as quickly as possible, using a chip that doesn't deliver 100% is detrimental to progress. While U.S. restrictions remain in place, Chinese AI companies are only going to catch up. Nvidia CEO Jensen Huang has already been expressing this sentiment, calling Huawei “formidable,” something it has proven with its AI CloudMatrix 384 cluster.
We can only wait and see how this new development turns out, but this goes to show how challenging and competitive the international market is. And as always, be sure to take rumors like this with a pinch of salt until it is confirmed.
About the Author