NVIDIA chips underpin US$3.2B in Chinese AI projects Reuters identified

Chinese AI projects worth about US$3.2 billion span operating installations and plans that still depend on access to restricted American processors. NVIDIA Corporation NASDAQNVDA, the U.S. designer of AI processors, is the subject of Reuters' Nov. 13 investigation, which examined more than 150 Chinese procurement notices, investment plans, construction contracts and corporate disclosures.

The Nov. 13 estimate includes buildings, power systems and other infrastructure, alongside planned investment and awarded contracts. It measures project budgets and agreements across different stages of development, so it cannot be read as NVIDIA revenue or a backlog of chip orders.

Shares in NASDAQNVDA fell to US$162.85 at the Nov. 13 close, timestamped 4:00 p.m. EST by Google Finance. That closing price provides the market setting for the investigation without establishing how much of the decline reflected the report.

NVIDIA's Aug. 26 outlook put third-quarter revenue at US$108.0 billion, plus or minus 2 per cent, excluding China data centre compute revenue. Its next results call is scheduled for Nov. 17, when management can address the distinction between Chinese demand and sales it can recognize.

The Records Span Plans And Operating Clusters

Reuters' Nov. 13 project breakdown describes a proposed 6.2 billion yuan computing centre in Xianyou county, Fujian province, that was seeking commercial partners. Its plans specified H100 processors, placing it among developments whose proposed equipment requirements do not establish completed installations.

An April government prospectus described a 2.26 billion yuan project in Jilin combining H100 processors with Chinese-developed chips, Reuters reported. The developer expected bank loans to finance about 80 per cent of the investment, adding a financing requirement to the planned hardware purchases.

In Inner Mongolia, Reuters identified a 1.5 billion yuan development proposing 625 H200 servers. Industry reporting in June said construction had begun on two first-phase buildings, separating progress on construction from confirmation that the processors were installed.

The Nov. 13 breakdown also describes a 1.16 billion yuan construction contract awarded in July for a computing centre in Gansu province. Tender documents specified approximately 3,072 H100 and H200 processors, but did not establish whether those chips had been delivered.

An operating cluster provides a different category of evidence: the Wuxi municipal government said in March 2025 that a facility using H800 processors had entered operation. Reuters placed that installation alongside proposed developments, making the aggregate broader than either completed capacity or recent chip shipments.

Licensing Depends On Customers And End Use

The U.S. Bureau of Industry and Security announced on Jan. 13 that H200 export licence applications would be reviewed case by case, subject to security requirements. The policy describes a conditional route for authorized exports, so a project specifying that processor does not by itself establish a violation.

Applicants must demonstrate that exports would not reduce semiconductor production capacity available to U.S. customers. Chinese purchasers must adopt export compliance procedures, including customer screening, and the products must undergo independent testing in the United States.

Sunny Cheung, an analyst at the Washington-based Jamestown Foundation, told Reuters that enforcement should examine the ultimate users of computing power and their purposes, rather than focusing solely on the immediate chip buyer. His criticism identifies an end-use concern distinct from whether developers continue to want NVIDIA hardware.

Reuters said NVIDIA and China's commerce ministry did not immediately respond to its requests for comment on the Nov. 13 investigation. The agency's project tally includes both prospective purchases and operating facilities, which require different evidence to establish delivery dates and the applicable licensing circumstances.

The Forecast Excludes China Compute Revenue

"NVIDIA is not assuming any Data Center compute revenue from China in its outlook," the company said in its Aug. 26 results release.

That Aug. 26 report put data centre revenue at US$89.0 billion for the quarter ended July 26. Total company revenue reached US$96.2 billion, up 106 per cent from a year earlier, showing the scale of sales already recorded outside the project estimates Reuters assembled.

Advanced Micro Devices, Inc. NASDAQAMD, a competing designer of data centre and personal computer processors, reported third-quarter data centre revenue of US$8.5 billion in its Nov. 3 results. That was up 97 per cent from a year earlier. The same revenue figure featured in Mugglehead's Nov. 12 coverage of AMD's computing expansion.

Read more: AMD expands PsiQuantum collaboration and plans equity investment

The next scheduled test of NVIDIA's China revenue assumption is its third-quarter results call, set for 5:00 p.m. EST on Nov. 17.