Nvidia Shipped $2.4 Billion of Hopper Chips to China Last Quarter. Now It Says It Has Stopped, and Its $130 Billion Forecast Assumes None.
On Tuesday, November 17, Nvidia reported third-quarter revenue of $115.8 billion, up 103 percent from a year earlier. It was the company's first report since Jensen Huang resigned as CEO on Thursday, November 12, and the numbers were not the problem. Revenue beat the roughly $109 billion analysts expected, per CNBC and Reuters.
The line that changes things was on the call. Colette Kress, Nvidia's CFO and interim CEO, said “we are no longer shipping licensed Hopper data center products to China,” Gizmodo reports, and TechCrunch reports the same decision. Those are the H200s President Trump cleared for sale to China in December 2025, with a 25 percent tariff collected by the US government.
The CFO commentary puts a size on what just ended: “Shipments of Data Center Hopper products to China during the quarter were $2.4 billion.” The fourth-quarter outlook of $130 billion, plus or minus 3 percent, assumes none of it comes back.
The Quarter in Numbers
Here is what Nvidia reported for the quarter that ended Sunday, October 25, and what it guided for the next one.
| Metric | Q3 fiscal 2027 | Comparison |
|---|---|---|
| Revenue | $115.8 billion | Up 103% year over year, up 20% sequentially |
| Data Center | $107.6 billion | Up 110% year over year, up 21% sequentially |
| Hyperscale | $60.1 billion | Up 71% year over year |
| AI Clouds, Industrial & Enterprise | $47.5 billion | Up 195% year over year |
| Hopper shipments to China | $2.4 billion | About 2% of revenue; now halted |
| Non-GAAP gross margin | 74.9% | Q4 guide: 71.0% to 72.0% |
| Supply commitments | $390 billion | Up from $279 billion a quarter earlier |
| Q4 revenue outlook | $130.0 billion, plus or minus 3% | Includes an extra week; assumes no China data center compute |
Even the bottom of the range, about $126.1 billion, sits above the roughly $125 billion consensus that CNBC and Reuters cited from LSEG. Shares rose in extended trading on Tuesday, more than 3 percent by CNBC's count and 4 percent by Reuters', as carried by RTÉ. That is a small bounce for a stock CNBC says is down 24 percent this month.
Kress's own summary in the release: “Vera Rubin is off to an explosive start, our roadmap is on track and our ambition has never been greater.”
The Rule Issued the Day Before
The H200 halt did not happen in a vacuum. On Monday, November 16, the Commerce Department's Bureau of Industry and Security released an interim final rule on remote access to advanced chips. BIS says it was written “by and at the direction of the Department of Treasury and the National Security Council,” and that its aim is to make sure existing restrictions “cannot be circumvented through remote computing arrangements involving data centers in third countries.”
In plain terms, per Nixon Peabody's summary and TechCrunch:
- A license is now needed to send advanced chips and servers to a data center anywhere if it may provide compute to companies headquartered in China or other restricted countries. Indirect access through brokers counts.
- The threshold is more than 12 million in aggregate total processing performance at one site or across linked data centers. TechCrunch estimates that is roughly the peak compute of about 750 H100 or H200 chips.
- Shipments of advanced chips to Indonesia, Malaysia, Singapore, Thailand and Brazil now need a license too. Chips going to US-headquartered companies are excluded, with monthly reporting by the exporter, per Nixon Peabody.
- Data centers already serving covered customers get 90 days from the effective date to obtain a license. Nixon Peabody says BIS indicated the rule takes effect on Federal Register publication, expected Thursday, November 19.
The same day as earnings, Nvidia's board created a committee to oversee national security policy, export control compliance and global trade risk, CNBC and TechCrunch report. CNBC says it will meet “at least quarterly.”
Two Numbers That Do Not Match
Here is the gap I keep coming back to. The China business Nvidia just switched off was $2.4 billion in a quarter. At that pace it is close to $10 billion a year.
Jefferies, according to TechCrunch, estimates the new rules could trim $44 billion to $62 billion from Nvidia's data center revenue next year. That is roughly 4.6 to 6.5 times the annualized H200 run rate. The exposure is not the chips that went to China. It is the chips that went to data centers outside China that rent to Chinese customers.
Kress's answer, as TechCrunch paraphrases it, is that the backlog for chips like Vera Rubin is deep enough that the rules will not hurt top-line revenue. Nvidia's own disclosures give that some support: supply commitments rose by $111 billion in one quarter, mostly for multi-year memory, and inventory rose to $39.7 billion from $31.6 billion to feed the Rubin ramp. When supply is that tight, a buyer the rule blocks can usually be replaced by one it does not.
Nvidia had also been tightening on its own. CNBC, citing two people familiar with the matter, reports that Nvidia warned buyers they could face legal action over Chinese remote access. CNBC also reports that Nvidia would not, for now, sell chips to customers that rent more than half their GPUs to Chinese companies.
Our Take
The H200 halt is a gesture with a modest price, and I mean that as a compliment to the timing. Nvidia gives up about 2 percent of a record quarter, in a market where, TechCrunch reports, China had largely refused H200 deliveries anyway. In return it shows Washington, five days after Huang's exit and a day after a rule written at the direction of Treasury and the NSC, that the new leadership will not fight for the loophole. Board chair Harvey Jones opened the call by saying the board “does not intend to reinstate Jensen as CEO.” Huang argued for selling into China for years. The company is now signaling the opposite.
The harder question is the one the outlook does not answer. “No Data Center compute revenue from China” is a clean sentence about China. It says nothing about Singapore, Malaysia, Thailand or Indonesia, where a slice of the AI cloud demand Nvidia sells into now needs a license. Nvidia does not break out how much of the $47.5 billion AI Clouds, Industrial & Enterprise line, up 195 percent, serves Chinese end customers. Neither do we know. That line is where I would expect any damage to show first, if it shows at all.
I think Kress is right about the next quarter and unproven about next year. When demand outruns supply, an export rule moves chips between buyers rather than destroying sales. The Jefferies range is a bet that it will not stay that way. A supply ramp built on $390 billion of commitments is exactly the kind of bet that is fine until demand cools.
Three Signposts for the Next 60 Days
- The 90-day clock. If the rule takes effect on Thursday, November 19, existing operators serving covered customers have until about February 17 to get licenses. Watch for the first Southeast Asian cloud provider to say it has applied, or that it is cutting Chinese tenants.
- The AI Clouds line. Nvidia's fourth-quarter report will show whether the $47.5 billion AI Clouds, Industrial & Enterprise segment keeps growing faster than hyperscale under the new licensing.
- December 3. Huang has been asked to testify before the Senate Banking Committee that Thursday. Watch whether senators ask what Nvidia knew about chips reaching Chinese customers through overseas data centers.
Sources
Nvidia: Financial Results for Third Quarter Fiscal 2027, Nvidia CFO commentary (via MarketScreener), BIS press release on the remote access rule, Nixon Peabody, TechCrunch on the rule, TechCrunch on the China halt, CNBC on earnings, CNBC on customer warnings, Gizmodo, RTÉ (Reuters)
