Free Intensive Nov 21–22Free 2-day hands-on Vibecoding Intensive on Nov 21–22 at 16:00 MSK (Sat–Sun) · Live Cohort 3 starts Nov 23Join Free →Cohort 3 (Nov 23)
ReleaseInsight · On RadarTechCrunch→ original

Nvidia quits selling H200 in China, and new US rules threaten $44–62 billion in lost revenue

Nvidia has stopped selling H200 accelerators in China, interim CEO Colette Kress announced. The US 25% tariff scheme is closed, and new rules also keep advanced chips out of overseas data centers that serve Chinese customers. Jefferies expects a $44–62 billion revenue hit, but Kress is confident that orders for Vera Rubin will be enough. If you rent GPUs abroad, ask your provider how it will prove it has no Chinese customers.

AI-processed from TechCrunch; edited by Hamidun News
Nvidia quits selling H200 in China, and new US rules threaten $44–62 billion in lost revenue
Source: TechCrunch. Collage: Hamidun News.
Listen to article
Article Fact Sheet & Key Figures
ParameterDetails
Primary sourceTechCrunch ↗
Editor-in-Chief
Huang and Trump struck the H200 dealchips were routed through the US with a 25% tariff
Criticsup to 20% of Nvidia's revenue is tied to demand from China
China itself barely accepted the H200authorities back domestic chips

Nvidia is ending sales of H200 accelerators in China. These are the chips used to train and run AI models. Interim CEO Colette Kress announced this on the first earnings call since Jensen Huang's departure. At the same time, the US has banned, effective Monday, shipments of advanced chips to overseas data centers that remotely serve Chinese customers. Put simply, two doors closed at once: direct sales to China, and the workaround through third-party servers abroad. If you rent GPUs from cloud providers or plan to buy compute, this affects prices, queues, and legal risk.

What exactly has closed

H200 sales to China rested on a deal brokered by Huang and President Donald Trump. The chips were supposed to pass through US territory, and the government collected a 25% tariff on them. Formally, this was a loophole in the export ban on advanced accelerators. China hawks were unhappy: they wanted to deprive Chinese labs of computing power and preserve US leadership in AI. Huang believed that competing in China mattered more. In April, he told podcaster Dwarkesh Patel that giving up that market would be a "disservice" to the country.

On the same day, Nvidia announced a board-level committee. It will oversee national security policy, export control compliance, and trade risks. Huang was removed in part over a romantic relationship with the head of a company that is under investigation on suspicion of smuggling chips into China.

How much it could cost

The new rules matter more than the H200 itself. Investment bank Jefferies forecasts that they could cut Nvidia's data center revenue by $44–62 billion next year. That is the bank's forecast, not a confirmed result.

Kress disagrees with the estimate. By her account, accumulated demand for advanced chips such as Vera Rubin is so large that the rules will not affect total revenue. In the long term, though, the company may lose potential customers in Singapore, Malaysia, Thailand, and Indonesia.

Critics of Nvidia's dependence on China claim that up to 20% of its revenue is tied to demand from the country. According to their figures, the annual revenue of its Singapore subsidiary reached $104 billion in 2025, after which Nvidia stopped disclosing such numbers. TechCrunch's assessment is that this will clearly change.

Why China isn't upset, and how the workaround worked

Even before Nvidia's decision, China had largely refused to accept H200 shipments. According to Jacob Feldgoise of the Georgetown Center for Security and Emerging Technology, authorities want to create demand for domestic chips, although major compute users would prefer stronger Nvidia hardware.

Chinese labs, meanwhile, looked for other routes. All leading Chinese AI models show signs of having been trained on Grace Blackwell chips, which are banned from export to China. The training likely took place in data centers outside the country. Sales to such centers are now closed.

Imagine a startup that rents a GPU cluster in Malaysia. If the data center has customers from China, it will have to take that into account. According to export control lawyer Dan Fisher-Owens, large data centers suitable for model training will have to certify that they do not serve Chinese customers. Those that can't will look for other chips, and Nvidia's share of this market could shrink further.

What to do in practice

Teams that depend on external GPUs should ask their provider in advance how it will pass these checks and whether it will switch to other accelerators. It remains unclear how the new board committee will affect strategy without Huang, the main champion of Nvidia's presence in China, and whether Kress's assurances about revenue will hold up.

Story

Key Questions Answered

What exactly has closed?

H200 sales to China rested on a deal brokered by Huang and President Donald Trump. The chips were supposed to pass through US territory, and the government collected a 25% tariff on them. Formally, this was a loophole in the export ban on advanced accelerators. China hawks were unhappy: they wanted to deprive Chinese labs of computing power and preserve US leadership in AI. Huang believed that competing in China mattered more.

How much it could cost?

The new rules matter more than the H200 itself. Investment bank Jefferies forecasts that they could cut Nvidia's data center revenue by $44–62 billion next year. That is the bank's forecast, not a confirmed result. Kress disagrees with the estimate. By her account, accumulated demand for advanced chips such as Vera Rubin is so large that the rules will not affect total revenue. In the long term, though, the company may lose potential customers in Singapore, Malaysia, Thailand, and Indonesia.

Why China isn't upset, and how the workaround worked?

Even before Nvidia's decision, China had largely refused to accept H200 shipments. According to Jacob Feldgoise of the Georgetown Center for Security and Emerging Technology, authorities want to create demand for domestic chips, although major compute users would prefer stronger Nvidia hardware. Chinese labs, meanwhile, looked for other routes. All leading Chinese AI models show signs of having been trained on Grace Blackwell chips, which are banned from export to China.

ZK
Zhemal Khamidun — CPO at Alpina Digital, product manager of AlpinaGPT. Building AI products since 2019. Hamidun News is a pet project that grew into a production aggregator with 6,000+ readers.
✦ Private AI Builders Community · Vibe Club

Ship AI Faster with Founders, Engineers & Weekly Architecture Reviews

Inside Vibe Club (hamidun.com/club) — weekly live masterminds with Zhemal Khamidun, Tech Radar, full LMS access, and 370+ production Claude Code skills.

Join Vibe Club →Explore Hamidun Academy
What do you think?
Loading comments…