iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
China’s Ministry of Commerce has proposed adding 40 technologies to its export-control catalogue, including AI models, intelligent agents and computing clusters. Public comments are open until November 15, 2026, and interim pre-export registration applies to relevant technologies. This is a consultation and provisional-registration process—not evidence that a final blanket ban on exporting Chinese AI models or chips is already in force.
What China announced
The Ministry of Commerce said the catalogue revision would add 40 technologies across artificial intelligence, biomedicine and photovoltaics. In the AI portion, it identified three building blocks: AI models, intelligent agents and computing clusters.
The ministry said it would collect public opinions through November 15, 2026. While the consultation is open, relevant technologies are subject to interim pre-export registration. The announcement describes the move as routine and says it does not target a specific country, region or entity.
Free tools Windows power users keep installed
One-click scans. No signup required.
Those details establish a proposal with an interim administrative requirement. They do not establish the final wording, technical thresholds, licensing tests, penalties or a universal prohibition on exports.
#1 Best Overall
- NVIDIA Volta GV100 Architecture — 4,608 CUDA Cores, 640 1st-Gen Tensor Cores delivering 14 TFLOPS FP32 and 112 TFLOPS deep learning performance for AI training, inference, HPC, and scientific computing workloads
- 32GB HBM2 ECC Memory — 900 GB/s Bandwidth — High-bandwidth memory on a 4096-bit bus with ECC error correction provides the memory capacity and throughput required for the largest AI models, simulations, and datasets
- PCIe 3.0 x16 Interface — 250W TDP — Standard PCIe Gen3 connectivity with passive cooling designed for enterprise rack server deployment in HPE ProLiant, Dell PowerEdge, and Supermicro platforms with adequate chassis airflow
- NVLink — Scale to 96GB Unified Memory — Connect two V100 GPUs via NVLink at 300 GB/s bi-directional bandwidth to scale GPU memory from 32GB to 96GB for larger AI training and HPC workloads
- Multi-Precision Computing — Supports FP64 (7 TFLOPS), FP32 (14 TFLOPS), FP16 (112 TFLOPS) and INT8 precision modes for flexible deployment across training, inference, and scientific simulation workloads
Proposal versus final rule
| Question | What is established now | What remains undecided |
|---|---|---|
| Legal status | Proposed catalogue revision under consultation | Whether, when and in what form the additions become final |
| Public input | Comments accepted through November 15, 2026 | Whether comments change categories or implementation |
| Interim process | Pre-export registration applies to relevant technologies | Whether registration remains, becomes licensing, or is replaced |
| Technical scope | Models, intelligent agents and computing clusters are named | Model-size thresholds, hardware specifications, software criteria and covered entries |
| Enforcement | No final enforcement framework has been published in the available announcement | Penalties, review standards, responsible offices and enforcement staffing |
Which AI technologies are named?
AI models
“AI models” is a broad category that could encompass systems used for training, inference or commercial distribution, but the proposal does not state which architectures, parameter ranges, capabilities or release formats would qualify. It also does not say whether a model supplied as downloadable weights would be treated differently from one delivered through a hosted service.
Intelligent agents
Intelligent agents are listed separately from models. The wording signals that systems able to plan, call tools or execute multistep tasks are being treated as a distinct building block. No autonomy threshold, tool-use test, sector restriction or named agent product is specified.
Rank #2
- Delivers 25 TFLOPS of AI inference performance using four MemryX MX3 accelerators in a compact M.2 2280 M-key module.
- Connects through PCI Express Gen 3 and above for straightforward integration into embedded and edge systems with an M.2 M-key slot.
- Includes 42 MB of distributed on-chip weight storage to support low-latency, batch-1 pipelined inference.
- Designed for low-power deployment with a 1 W–12 W operating range and host-managed airflow requirements.
- Supports ONNX, TensorFlow, Keras, and TFLite models through the MemryX software stack on Linux and Windows hosts.
Computing clusters
Computing clusters cover the infrastructure used to train or run AI systems. The announcement does not publish a chip list, performance threshold, interconnect requirement or cloud-computing test. Consequently, it is too early to say that a particular processor, server, accelerator or data-center deployment is prohibited.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Model weights, data and proprietary technology
Earlier policy discussions reportedly considered model weights, training data and transfers of proprietary AI technology. Those subjects provide context for the proposal but are not confirmation that every one appears as a final catalogue entry.
Rank #3
- Professional AI & Creator Workstation: AMD Radeon AI PRO R9700 GPU with 32GB GDDR6 is engineered for AI development, professional content creation, and compute-intensive workloads.
- Massive 32GB Memory Capacity: 32GB of GDDR6 memory on a 256-bit bus provides ample bandwidth for large AI models, 8K video editing, and complex 3D rendering.
- Advanced RDNA 4 with AI Accelerators: 64 Compute Units with 3rd Gen Ray Tracing and dedicated 2nd Gen AI Accelerators for groundbreaking AI performance and visual computing.
- Professional Blower Cooling: Efficient single blower design exhausts heat directly out of the chassis, ideal for multi-GPU workstation and server configurations.
- Enterprise-Grade Thermal Solution: Vapor chamber heatsink with industrial Honeywell PTM7950 thermal interface material ensures reliable cooling under sustained professional loads.
What the proposal could mean for chips and cloud compute
The chip question currently turns on how “computing clusters” is defined. If final rules cover only complete clusters, registration might focus on integrated systems rather than individual components. If they later identify accelerators, networking equipment, software stacks or performance bands, the effect on chip exports and cloud capacity could be much broader. None of those thresholds has been published.
Cloud providers and overseas customers should therefore avoid treating the announcement as either a confirmed chip embargo or a guarantee that all Chinese AI compute remains exportable. The practical trigger will be the final list and the procedure attached to each entry.
How registration, licensing and prohibition differ
| Mechanism | What it normally requires | Status in this announcement |
|---|---|---|
| Pre-export registration | Exporter submits transaction or technology details before shipment or transfer | Interim requirement confirmed for relevant technologies |
| Licensing | Exporter seeks approval and may need to satisfy destination, end-user or security conditions | Not confirmed as the final mechanism |
| Prohibition | Specified exports or transfers are barred except under a stated exemption | No blanket prohibition established |
Until final rules are issued, companies should not describe registration as a ban. They should also not assume that a completed registration guarantees approval under a later licensing or prohibition regime.
Do these 3 things before closing this tab:
1Repair Windows errors before they cause bigger problems2Scan for outdated or missing drivers - takes under a minute3Clear out junk files and repair common Windows errorsWhich cross-border activities are most exposed?
| Activity | Why it matters | What is known |
|---|---|---|
| Exporting downloadable or open-weight models | Moving weights can give an overseas party direct control of the technology | Earlier policy discussions focused on overseas access; final treatment is unknown |
| Remote access to a Chinese AI service | Users receive outputs without necessarily receiving model weights | The announcement does not establish whether remote service access will be treated differently |
| Sending training data abroad | Data can enable reproduction, fine-tuning or improvement of a model | Training-data controls were discussed previously, but no final rule is stated |
| Transferring proprietary technology | Source code, designs and know-how may move even without a model export | Earlier reporting described possible controls; catalogue coverage is not confirmed |
| Exporting or operating computing clusters | Clusters combine chips, networking, storage and software | Clusters are named, but technical entries and thresholds are not published |
Why the policy appeared now
Reuters reported in July 2026 that Chinese authorities had met leading technology companies about potentially restricting overseas access to China’s most advanced AI models, including models not yet released. The discussions reportedly covered both closed-source and more open versions, stronger treatment of leaks or theft of proprietary technology, possible limits on who could fund domestic AI startups, and controls focused mainly on future models. Scope and timing were unresolved in that report.
Best Value
- Memory Size: 16 GB GDDR6 ECC.
- Memory Bus Width: 128-bit.
- Memory Bandwidth: 200 GB/s.
- CUDA Cores: 1280.
- Peak Single Precision floating point performance: 18 Tflops (GPU Boost Clocks).
The companies mentioned in that July context included Alibaba, ByteDance and Z.ai. Their Qwen, Doubao and GLM-5.2 systems were described as widely used or technically notable, but those references do not show that any named product is included in the October catalogue proposal.
The timing also matters geopolitically. China describes the catalogue revision as routine and non-targeted, while it arrives during intensifying technology competition between China and the United States. A non-targeted statement does not answer how rules will treat foreign destinations, counterparties or investors once technical criteria are published.
What companies should monitor before exporting
- Identify the transaction. Separate model-weight delivery, API access, agent deployment, data transfer, software transfer and physical or cloud cluster access.
- Check the final catalogue language. Look for named entries, capability thresholds, hardware specifications and definitions of “agent” and “computing cluster.”
- Track the procedure. Confirm whether registration remains the requirement or whether a licence, security review or prohibition applies to a particular category.
- Map counterparties and destinations. Record the exporter, importer, end user, cloud operator, funder and ultimate beneficiary rather than relying only on the immediate customer.
- Preserve technical records. Keep model versions, weight hashes, cluster specifications, training-data provenance and access logs so a filing can be matched to the exact technology.
- Reassess future-model plans. Earlier discussions contemplated controls focused on models not yet released, so a product that is outside the interim process today may not remain outside the final regime.
When will the new rules take effect?
There is no confirmed effective date for a final set of AI export controls in the available announcement. November 15, 2026 is the consultation deadline, not necessarily the date on which a final catalogue becomes enforceable. Interim pre-export registration is already being applied to relevant technologies, but the announcement does not specify every covered transaction or how long that interim process will last.
Recommended Free Tools
The next decisive publication should answer whether the catalogue is adopted, revised or withdrawn and which compliance mechanism applies to each category.
Quick Recap
What is still unknown
- Exact catalogue entries and technical thresholds.
- Whether individual chips, complete servers, cloud capacity or only defined computing clusters are covered.
- Whether downloadable weights, hosted inference, agent actions and training-data transfers receive different treatment.
- Licensing criteria, exemptions, review times and penalties.
- How rules apply by destination, end user, investor or technology partner.
- Which agencies will enforce the final regime and how compliance will be audited.
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
