Skip to main content

The Best AI Video Generators in 2026: Which One Is Worth Your Money?

Nvidia Stops Shipping H200 AI Chips to China: What Happened and Why It Matters

Nvidia H200 AI chips China

The world's biggest AI chipmaker just walked away from China — and Wall Street barely blinked. On Tuesday, November 17, interim CEO Colette Kress told investors Nvidia is no longer shipping its H200 AI chips to China. The surprise isn't that Nvidia stopped — Beijing had stopped accepting them.

Kress said it on her first earnings call as interim CEO, days after Huang stepped down. Her one sentence closed the book on the licensed Hopper chips Nvidia had been selling in China: the company's guidance, she said, "does not have any licensed Hopper data center products included, and we are no longer shipping licensed Hopper data center products to China."

At a glance — Chip: Nvidia H200, a Hopper-generation AI accelerator. Change: no more licensed Hopper shipments to China. Announced: November 17, 2026. China Hopper sales last quarter: $2.4 billion, about 2% of revenue. Total revenue: $115.8 billion, up 103%. Next quarter: about $130 billion, no China data-center revenue.

What did Nvidia actually announce?

The H200 was the most advanced AI chip Washington had licensed Nvidia to sell in China, the workhorse behind today's large AI models. But the business had shrunk to a footnote: $2.4 billion of licensed Hopper sales last quarter, roughly 2% of $115.8 billion in total revenue.

That is why Kress could deliver the news so calmly. Nvidia's forecast for the current quarter, about $130 billion, already assumes zero data-center revenue from China. Walking away cost Nvidia almost nothing — which raises the question of why it was selling the chips at all.

How did Nvidia end up selling H200 chips to China?

Under US export controls, advanced American AI chips are banned from sale to China. The H200 got through anyway, under a licensed exception brokered by Huang and President Trump: the chips had to be shipped through the United States, letting the government collect a 25% tariff. Critics called it a loophole.

But the arrangement was already failing. Reuters reported that roughly 600,000 licensed H200 chips were sitting in Hong Kong because Beijing had not approved large-scale imports. Georgetown researcher Jacob Feldgoise told TechCrunch that China had largely refused H200 deliveries even before Nvidia stopped shipping them, preferring its own chips. Nvidia wasn't losing a thriving market — it was closing a channel its biggest customer had already stopped using.

Why is Nvidia stopping now?

Kress gave no reason. The timing tells its own story. The day before earnings, Washington issued new rules barring advanced US chips from foreign data centers that serve Chinese customers remotely — closing the workaround that let Chinese firms rent American chip power from abroad. Jefferies put the possible cost at $44 billion to $62 billion of data-center revenue next year, though Kress said demand for chips like Vera Rubin means no top-line impact.

The same day, Nvidia announced a new board-level committee on national security and export compliance. Hours earlier, the Wall Street Journal reported an SEC civil investigation into whether Nvidia's disclosures misled investors. The SEC declined to comment; no charges have been announced. At the center is Megaspeed, a Singapore firm suspected of diverting Nvidia chips to China; Commerce has investigated it since late 2024, and the Justice Department now leads a criminal investigation.

What does this mean for Nvidia — and for the AI race?

For Nvidia, less than the headlines suggest. The $2.4 billion was already excluded from its forecasts, and demand for its newest chips outstrips supply. The AI money is flowing elsewhere anyway: Anthropic's $81 billion IPO, the biggest ever, and OpenAI's two new models, GPT-6.2 Astra and Sol, both landed the same week.

For China, the cutoff speeds up a shift already underway. Beijing's refusal to approve H200 imports was a bet its own chipmakers can close the gap — and the new remote-access rules make that bet more urgent. For American policymakers, the lesson is plain: export controls only bite when the workarounds close, too.

What happens next?

Watch the permanent CEO search, the board committee's findings, the reported SEC inquiry, and whether China's domestic chips can do what the H200 did. The real story isn't a company sacrificing billions for principle. It's a door closing that Beijing had already walked away from.

Comments

Popular posts from this blog

Google Gemini 4 Neon Explained: What the New Model Can Do and Who It's For

Google released a brand-new AI model this week without a flashy keynote — just a technical document, quietly published on November 18. If you write code, run a business, or do research for a living, Gemini 4 Neon might be the most useful Google launch of the year. So what is it, and is it worth your attention? The short answer: yes, if you do serious work with AI. What Is Gemini 4 Neon? Gemini 4 Neon is the middle child of Google's Gemini 4 family, and that is exactly the point. Google DeepMind's model card, published November 18, 2026, describes Neon as adapted from the flagship Gemini 4 Carbon, built for complex coding, analytical reasoning, and enterprise work. In plain English: it is built for real work, not demos. Model Built for Best at Hydrogen Everyday speed Quick answers, fast answers Neon Serious work Coding, analysis, long projects Carbon Hardest problems Deepest reasoning You get most of Carbon's muscle at a lower price and a quicker pace. What Can Gem...

What Is Agentic AI? The 2026 Trend Explained in Plain English

Diagram: Marxav, CC BY-SA 4.0, via Wikimedia Commons You've probably heard the phrase "agentic AI" by now. It is in every tech headline, every earnings call, every "top trends of 2026" list. And if you are like most people, you have nodded along while quietly wondering what it actually means. Here is the answer, up front, in one sentence: agentic AI is artificial intelligence that does not just answer your questions — it goes off and does the work for you. That is the whole idea. A chatbot waits for you to ask something, then replies. An agentic AI system takes a goal from you — "book me the cheapest flight to Chicago next month," say — figures out the steps, uses real software tools to carry them out, checks its own work, and keeps going until the job is done. And before you ask the obvious follow-up — no, it is not science fiction, and it is not only for big companies. It is already running in customer service departments and software teams, and...

How to Use AI to Write a Resume That Gets Past the ATS in 2026

Photo: Shixart1985 / Wikimedia Commons, CC BY 2.0 Your resume gets about six seconds. Not six seconds with a hiring manager — six seconds with software, before any human ever sees it. The good news: you can use AI to beat that software at its own game. This guide shows you exactly how — the prompts, the format, the keywords, and the one line most AI-written resumes get wrong. Short answer: yes — but only if you use AI to tailor your resume to each specific job, match the posting's exact language, and keep the formatting dead simple. A generic AI resume parses just fine, then ranks below everyone who tailored theirs. The trick is not the AI. It is what you ask it to do. There is one line most AI-written resumes get wrong, and it sits in the first three lines of the page. I will show you what it is near the end — the fix takes about ninety seconds, and it changes how the whole resume reads. What Does an ATS Actually Do to Your Resume? An ATS is the software a company uses to co...