On Tencent’s third-quarter earnings call, president Martin Lau said a new US rule restricting Chinese companies’ access to high-end GPUs located outside China is setting the company back. The measure, as described in available reports, also covers remote access to computing capacity that Chinese firms already rent or have built abroad, with direct consequences for the training of future AI models.
๐ Key takeaways
- The rule restricts remote access to high-end GPUs hosted outside China, including capacity that is rented or was already built abroad.
- Lau describes a setback that hits everyone in terms of advanced computing power, without a quantified financial impact.
- Hy4, the next version of Tencent’s model, faces a longer development cycle in the short term.
- WorkBuddy, with 10 million users, should be less affected thanks to alternative models and third-party inference providers.
A restriction aimed at remote access
During the earnings call, Martin Lau commented on a US rule that limits Chinese companies’ access to high-end GPUs located outside mainland China. According to available descriptions, the measure bars them from reaching those chips remotely. The scope therefore goes beyond owning hardware: it also includes renting computing capacity abroad and infrastructure that was already built outside China. This widening of the perimeter is the central point of the analysis, because it hits companies that believed their access to computing power was secured.
- Scope: remote access to high-end GPUs located outside China.
- Capacity covered: GPU rentals and infrastructure built overseas.
- Not specified: the technical threshold for the chips, the effective date, the legal text and the responsible US authority.
Lau called the short-term effect on Tencent‘s operations “unfortunate”, notably on its ability to train artificial intelligence models. The sources offer no quantified estimate of that effect, and no share-price move attributable to his remarks.

A US regulatory framework that now reaches infrastructure
A report from the Congressional Research Service places the measure in a longer sequence. Executive Orders 13942 and 13943, adopted in 2020, targeted TikTok and WeChat, the service operated by Tencent. Their implementation was blocked by court injunctions. The orders were later revoked and replaced in 2021 by Executive Order 14034, which created a review process for risks tied to apps and software owned or controlled by foreign adversaries.
- On the Chinese side, data localisation requirements and export controls on data and algorithms have been tightened since 2014.
- Security review requirements have applied since 2021 to Chinese companies operating or listing abroad.
- In March 2026, the Trump administration was weighing whether to let Tencent keep its stakes in Epic Games, Riot Games and Supercell, among other gaming companies. Tencent had also invested in Snap.
The sources stress that this historical framework is distinct from the new rule on GPU access. It nonetheless reveals a broader trend: Washington now treats computing capacity and digital services as strategic assets, while Beijing mirrors this with tighter controls over data. For a group whose activities span gaming, social networks, cloud and AI, exposure to both logics is simultaneous.
Hy4: a longer development cycle in the short term
Lau said the next version of its Hy4 model will take longer to develop in the short term. Tencent had argued that Hy4’s training efficiency was high, but that reduced availability of cutting-edge hardware would lengthen the development cycle that follows. Teams that depend on rented foreign GPUs for training are therefore working with more limited access and slower iterations.
“It sets us back. It sets us back in China. So everyone actually has less advanced computing power, no matter when you actually started renting or building capacity outside China.”
Martin Lau, president of Tencent
This point deserves attention. According to the descriptions, the rule contains no grandfathering: a company that rented or built capacity abroad before the rule takes effect ends up in the same position as a new entrant. The sources do not allow the exact legal text or any exemptions to be verified.
What the sources cannot quantify
- The exact delay for Hy4, in months or as a percentage of the development cycle.
- The share of cloud growth that depends on the affected GPUs.
- The expected financial cost for Tencent, whether direct or linked to deferred investment.
WorkBuddy and cloud: uneven exposure depending on use
WorkBuddy, Tencent’s agentic assistant, illustrates the distinction the company draws between training and inference. It is described as a growing success, with a user base largely located outside China. Tencent can run the service on different models and use third-party providers for inference. Lau concluded that the new rule should not create a problem for this product, unlike the training of new models and certain cloud activities in the short term.
| Indicator | Reported value | Reading |
|---|---|---|
| WorkBuddy users | 10 million | Base reported by the company |
| Tokens consumed per day | more than 100 trillion | Aggregate inference demand |
| Cloud revenue growth | +35% year on year | Versus +8% for the group |
| Financial impact of the rule | Not specified | No figure in the sources |
The contrast between these figures is telling. Tencent’s cloud revenue rose +35% year on year, more than four times the +8% growth of the group as a whole. The sources, however, do not say how much of that growth rests on GPUs covered by the restriction. GPU rentals and other fast-growing cloud services may be affected, while cloud services unrelated to these accelerators can keep expanding outside China.
The nuance matters. Direct offers of high-end GPUs stop, but Lau indicated that Tencent could work with partners active in different markets. The two elements are not contradictory. What disappears is the company’s ability to supply these accelerators itself from abroad; what remains are activities that do not depend on them, along with international partnerships.
Domestic chips: a substitution announced without a timeline
Lau believes domestic chips can handle inference tasks. Over the long term, he expects Chinese companies to rely more on their local suppliers and domestic chips. The restriction could accelerate this shift by focusing development efforts on local providers. However, the forecast came with no timeline and no estimate of how long it would take to reduce Tencent’s dependence on US GPUs.
“But over time, the domestic supply chain would actually get set up.”
Martin Lau, president of Tencent
In the market for alternatives, Enflame Technology, an AI chip maker, jumped 179% on its Shanghai Stock Exchange debut, nearly tripling its valuation despite losses and heavy dependence on Tencent. The sources do not show that this rise was a reaction to the new US rule. It does, however, provide context on the Chinese chipmakers that Tencent plans to rely on more heavily in the long run.
Outlook and scenarios
In the short term, the available evidence points to three plausible trajectories, none of which can yet be weighted with figures. In the base case, access to foreign GPUs stays restricted, Hy4’s training cycles lengthen and inference gradually shifts to domestic chips. In an acceleration case, the constraint pushes Chinese players toward local suppliers faster than expected, which would benefit listed domestic chipmakers. In a workaround case, international partnerships and third-party models preserve consumer-facing activity, as Lau suggests for WorkBuddy, but leave open the question of training frontier models.
On sourcing, caution is warranted. The Information and The Register are the two first-hand sources. Mezha and Lavx rely on The Register’s reporting, and TokenHeart only points to it, so these are not additional independent confirmations. No contradiction was found between the sources, but the absence of the legal text, the technical threshold and the effective date limits how definitive any conclusion can be.
Sources
- The Register, November 19, 2026
- The Information, Juro Osawa, November 18, 2026
- Mezha
- Lavx
- TokenHeart, November 19, 2026
- Congressional Research Service, IF12640
This article is published for informational and educational purposes only. It does not constitute investment advice under any circumstances. Do your own research (DYOR) before making any decision.


