Shares of Nvidia (NVDA 2.10%) tumbled after the company revealed that it will incur a $5.5 billion charge in the first quarter related to its H2O graphics processing units (GPUs) due to new export restrictions on the sale of its chips to countries like China. Nvidia’s stock has had a tough start to 2025, with its share price down about 25% as of this writing.
A dumbed-down version of its H100 and H200 GPUs, Nvidia’s H20 chip was designed specifically to sell into the Chinese market to meet prior export control restrictions. These chips have lower bandwidth and slower interconnection speeds to meet export guidelines. However, it will now require an export license to China to sell any chips.
China was Nvidia’s fourth-largest geography in fiscal year 2024, with $17 billion in sales last year. However, last quarter, the company reported that its revenue from China had dropped by half compared to the period before the original export restrictions were implemented. DeekSeek’s popular R1 model was supposedly trained on H20 chips.
Why is the dip a good buying opportunity?
The H20 export ban will impact Nvidia’s sales in China, which accounted for 13% of its $130.5 billion in total revenue for the last fiscal year. However, sales won’t stop entirely, as Nvidia sells other chips, such as the L20 and L2, that aren’t banned. Additionally, a sizable black market exists in China for Nvidia’s chips.
There is also no natural replacement for Nvidia’s chips in China. Huawei has developed artificial intelligence (AI) chips manufactured by SMIC, but the Chinese foundry has limited access to the extreme ultraviolet (EUV) lithography tools used to manufacture advanced chips, such as GPUs. This is because EUV tools from ASML, the only maker of these systems, are also prohibited from being sold to China. Consequently, the supply of Chinese-made AI chips is limited. Given this, I wouldn’t be surprised if sales of black market Nvidia chips to China increase.
Nvidia could also start directing some of the manufacturing capacity dedicated to its H20 chips to other chips, such as Hopper and Blackwell. These chips are estimated to cost between 2 and 4 times the cost of an H20 chip, so there could be a benefit. Now, this doesn’t just happen with a snap of the finger, but it is possible.
Shifting to the older Hopper architecture will be easier since the H20 chips share the same die design. Blackwell uses a dual-die design, and Nvidia would need to secure more of Taiwan Semiconductor‘s Chip-on-Wafer-on-Substrate (CoWoS) advanced packaging capacity, which would make it more difficult and time-consuming. Blackwell is already capacity-constrained. As such, shifting capacity to Hopper could help in the near term, while transitioning to Blackwell would likely take considerable time, although it could provide a longer-term benefit.
Overall demand for Nvidia’s chips remains robust, though. Cloud computing companies are spending big on AI infrastructure to keep up with the increasing demands of their customers who run AI workloads through their data centers. The big three — Amazon, Microsoft, and Alphabet — plan to spend a combined $250 billion-plus on AI data center capital expenditures (capex) this year.
Meanwhile, companies like OpenAI, Meta Platforms, and xAI have been spending heavily on building out their own AI infrastructure in a race to develop the best AI models. Meta’s Llama 4 AI model and xAI’s Grok 3 were both trained on 10 times as many GPUs as their prior versions. Enterprises have also begun increasing their AI infrastructure spending, as many look to deploy a hybrid cloud model.
Nvidia predicts that AI data center capex will surpass $1 trillion by 2028. Even without China in the mix, that’s a significant amount of potential growth for the company moving forward. It won’t capture all the spending directed toward AI chips, but it will secure more than its fair share. The company has established a wide moat in the GPU space with its CUDA software platform and the extensive collection of libraries and tools it has developed to facilitate AI model training and speed up inference.
Time to buy the stock
Nvidia’s stock is cheap enough at current levels that it can withstand any reduction in growth resulting from the loss of selling its H20 chips to China. The stock currently trades at a forward price-to-earnings ratio (P/E) of under 23 times this year’s analyst estimates and a 0.44 price/earnings-to-growth (PEG) ratio, with numbers below 1 being considered undervalued.
If you wiped away $15 billion in Chinese revenue, Nvidia’s estimated revenue growth this year would go from 54% to 43%. Meanwhile, it would reduce its 2025 earnings per share by about $0.35 (using a 75% gross margin, 25% tax rate, and 24.4 million shares outstanding) to $4.10. That would be a forward P/E of about 25 times.
So, even if most of its Chinese revenue were to disappear, Nvidia’s stock would still look attractively valued. As such, this is a good time to start accumulating the stock while looking to add to more shares on any further pullback.
John Mackey, former CEO of Whole Foods Market, an Amazon subsidiary, is a member of The Motley Fool’s board of directors. Suzanne Frey, an executive at Alphabet, is a member of The Motley Fool’s board of directors. Randi Zuckerberg, a former director of market development and spokeswoman for Facebook and sister to Meta Platforms CEO Mark Zuckerberg, is a member of The Motley Fool’s board of directors. Geoffrey Seiler has positions in Alphabet. The Motley Fool has positions in and recommends ASML, Alphabet, Amazon, Meta Platforms, Microsoft, Nvidia, and Taiwan Semiconductor Manufacturing. The Motley Fool recommends the following options: long January 2026 $395 calls on Microsoft and short January 2026 $405 calls on Microsoft. The Motley Fool has a disclosure policy.