Beijing's reaction to Washington's new remote-access controls arrived quickly and in unusually familiar language. In a 17 November statement, the MOFCOM spokesperson said China "has noted" the 16 November measures, accused the US of "over-generalising the concept of national security" and "abusing export controls," and warned that "if the US persists in going its own way, China will take countermeasures." That is, almost word for word, the standard template deployed after every major US export-control action of the past several years.
The more useful reading is therefore not what Beijing said, but what it left out — and what the underlying US interim final rule (IFR) actually changes. The rule marks a conceptual shift that matters considerably to anyone exposed to advanced computing hardware: the control unit is no longer a chip or a server shipment, but the aggregate compute capacity a data centre makes available to specified end users, anywhere in the world.
The mechanism: controlling compute, not cargo
The BIS interim final rule, formally "Addressing Remote Access to Advanced Semiconductors" (RIN 0694-AK50), creates a new data-centre end-use control at EAR § 744.24. According to practitioner summaries of the text (Nixon Peabody; KPMG), a licence is required to export, reexport or transfer in-country advanced chips and servers — ECCNs 3A090 other than 3A090.c, 4A090 and related .z items, software and technology — where the exporter has knowledge that the destination data centre anywhere in the world supplies, or will supply, more than 12,000,000 aggregate total processing performance (TPP) to or for companies headquartered in, or whose ultimate parent is headquartered in, China, Country Group D:5 destinations, or Macau.
Three design features do the heavy lifting:
- Aggregation across facilities. The threshold is calculated per facility or across "linked data centres" — digitally interconnected infrastructure counts as one. A Chinese-headquartered tenant that spreads workloads across several third-country campuses cannot divide its way below the limit if those sites are networked.
- Knowledge-based liability. The obligation attaches to what the exporter knows or should know, backed by heightened know-your-customer diligence on data-centre customers and their ownership. A customer's refusal to supply ownership information is designated a "red flag" that presumptively creates knowledge.
- No de minimis escape where that knowledge exists, plus mandatory downstream written notice to recipients of unlicensed shipments on or after publication, and a 90-day deadline for existing arrangements to obtain a licence.
In parallel, the rule imposes a separate, country-level licence requirement for advanced semiconductor items to or within Indonesia, Malaysia, Singapore, Thailand and Brazil, with case-by-case review that explicitly weighs each country's enforcement posture. US-headquartered consignees and end users can escape that requirement by filing monthly reports under a new § 743.10 — a structure that effectively deputises US cloud and hyperscaler operations in those jurisdictions while raising the compliance bar for local operators and joint ventures.
The rule also amends the presumption-of-approval framework for front-end fabricators and advanced-node logic chip packagers, narrowing the routes by which designers in licence-requiring destinations can overcome the presumption. And it formally rescinds the Biden-era AI Diffusion Rule — including its country tiers, the Data Center Validated End User programme, licence exceptions AIA and LPP, and the 4E091 model-weights control — while retaining Licence Exception ACM and the worldwide scope of the advanced-computing foreign direct product rule at 734.9(h)(2)(i).
| Element | Effect |
|---|---|
| New § 744.24 data-centre end-use control | Licence required above 12 million aggregate TPP for China/D:5/Macau-headquartered users, worldwide |
| Linked-data-centre aggregation | Networked facilities counted together; defeats jurisdictional arbitrage |
| Country licences: Indonesia, Malaysia, Singapore, Thailand, Brazil | Case-by-case review factoring enforcement; US-headquartered recipients exempt via § 743.10 reporting |
| Foundry/packager presumptions | Tightened for advanced-node logic designers in restricted destinations |
| AI Diffusion Rule | Rescinded in full, including tiers, VEU and model-weights control |
| Compliance clock | 90 days for existing arrangements; 30-day savings clause for goods en route |
Sources: BIS press release; Nixon Peabody and KPMG client alerts. The Federal Register document (2026-23529) was not machine-readable at the time of writing, so details here rely on secondary professional summaries and should be checked against the published text.
Why Beijing's reply is milder than the rule
The MOFCOM statement is roughly 150 characters of substance. It names no countermeasure, no timeline, no product and no legal instrument. That reticence contrasts sharply with the rule's scope, which touches not only Chinese buyers but data-centre operators and cloud providers in Singapore, Malaysia and the Gulf, and which is framed by BIS as closing a circumvention channel.
Two readings are plausible, and they are not mutually exclusive.
First, the reference to "important consensus reached by the two heads of state" is doing specific work. It frames the US action as a breach of an existing bilateral understanding rather than as a standalone grievance — a rhetorical move that preserves the option of returning to negotiation while establishing a diplomatic record. The Foreign Ministry's parallel comment on the same day added the phrases "economic coercion" and "extraterritorial jurisdiction," language aimed at a wider audience than the bilateral channel.
Second, Beijing's most credible retaliatory levers sit outside semiconductors and have already been partly pre-committed. Analysts cited in press coverage flagged rare-earth export controls as the obvious response, noting that China's broader restrictions were reportedly suspended until 10 January 2027. That suspension date is analytically important: it means any rare-earth escalation would require Beijing to unwind a commitment it has already made, at a moment when it is simultaneously accusing Washington of breaking one. That raises the political cost of retaliation and may explain the delay. This is analysis, not confirmed policy — I have not found independent confirmation of the suspension terms, and no new rare-earth measure had been announced as of 17 November.
The industry split the rule creates
The Semiconductor Industry Association welcomed the rescission of the AI Diffusion Rule on the day of publication and said it was still evaluating the IFR — a split reaction that captures the industry's genuine dilemma. The Diffusion Rule was widely disliked by US allies and by chipmakers for its country-tier architecture, which imposed caps and licensing burdens on friendly jurisdictions. Its removal is a clear win for US firms selling into the Gulf, India and Southeast Asia.
The replacement is less objectionable in principle but harder in practice. Traditional export control operates on a shipping document; § 744.24 operates on a customer's compute architecture. Exporters, cloud providers and data-centre operators must now assess not just who is buying, but how much processing performance that buyer's facility delivers to which ultimate parents, and whether that facility is networked to another. Smaller operators in the five listed countries — many without the legal infrastructure to make and document such determinations — will find this materially more burdensome than the licence itself.
BIS has signalled active monitoring of jurisdictions that tolerate evasion, with possible import restrictions, additional licence requirements, downgrading of country status, Entity List designations and temporary denial orders. For operators in Singapore, Malaysia and Thailand, that is a direct commercial risk: enforcement reputation now has a price.
What to watch
- The Federal Register text. The 12-million TPP aggregation rule and the definition of "linked data centres" are where litigation and licensing disputes will concentrate. Confirm the calculation methodology against the published rule, not summaries.
- The 90-day compliance window. The deadline for existing arrangements to obtain licences falls in mid-February 2027. Any disruption to third-country cloud capacity serving Chinese-headquartered tenants will show up in contract renegotiations well before then.
- The 10 January 2027 rare-earth date. If Beijing intends to signal seriousness, this is the most likely trigger point. Absence of action by then would suggest a preference for negotiation over escalation.
- The 60-day comment period. The rule is an interim final rule, meaning it is already in force. Industry comments are a genuine opportunity to shape the final text — particularly the aggregation and knowledge provisions, which are the most operationally difficult elements.
For companies in steel and heavy industry, the direct exposure is limited but not zero: advanced computing underpins process modelling, logistics optimisation and increasingly autonomous operations, and some larger groups have begun contracting third-country cloud capacity. The indirect exposure — through the broader trajectory of US–China technology controls, and through any Chinese response that reaches critical-minerals supply chains — is considerably larger. That is the channel worth monitoring.
Source & methodology +
MetalMate separates original-source evidence from summary and analysis. Verify the original source before procurement, legal or engineering decisions.
- Article status
- Published · Version 1
- Source date
- Not supplied
- MetalMate publication
- 17 November 2026
- Verification
- Automated evidence checks; not human reviewed
商务部新闻发言人就美商务部发布限制远程获取先进半导体相关措施答记者问 — MOFCOM
MetalMate provides an English summary; consult the original wording.
商务部新闻发言人就美商务部发布限制远程获取先进半导体相关措施答记者问 — MOFCOM
MetalMate provides an English summary; consult the original wording.

