Proposed US Export Controls for Advanced AI Accelerators
The U.S. Department of Commerce is preparing a sweeping, tiered export-control regime for advanced AI accelerators from U.S. vendors such as Nvidia and AMD, expanding beyond country-specific restrictions into a broader licensing framework that could require U.S. approval for a wide range of global shipments. Reporting describes a multi-level structure tied to computing scale: smaller shipments (e.g., up to 1,000 Nvidia GB300 GPUs) would face an expedited review, while mid-scale deployments would require pre-authorization before an export license application, along with compliance measures such as operational transparency, disclosure of business activities, and potential on-site inspections by U.S. authorities.
For very large AI clusters (described as deployments on the order of 200,000 GB300 GPUs operated by a single entity in one country), the proposed approach would elevate requirements to government-to-government engagement and could condition approvals on commitments to invest in U.S. AI infrastructure as part of national-security assurances. Separate reporting in the set covers adjacent U.S. technology-security policy issues—NIST leadership testimony on AI standards and manufacturing priorities, China’s semiconductor industry calls to consolidate efforts to build an ASML alternative under export-control pressure, and CFIUS deliberations over Tencent’s stakes in major game companies—but those items are not the same event as the AI-accelerator export-rule proposal and should be treated as distinct policy stories.
Get ahead of threats like this
Mallory correlates global threat intelligence with your attack surface — know if you’re exposed before adversaries strike.
How this story unfolded
3 events from the most recent confirmed update back to the earliest known activity.
Reports detail U.S. investment requirement for very large foreign AI chip deals
Further reporting said the proposed framework would require very large foreign purchases of advanced U.S. AI chips to include commitments to build AI infrastructure in the United States, alongside possible inspections and national security assurances. The approach was reported to mirror conditions used in a recent UAE-related licensing arrangement involving Nvidia and Cerebras.
Commerce Department confirms new AI export-control approach under consideration
The U.S. Department of Commerce confirmed it is considering changes to AI hardware export rules and said the effort is not a return to the Biden-era AI Diffusion Rule. The department described a multi-level licensing structure tied to shipment size and computing capacity, with stricter conditions for larger deployments.
U.S. drafts worldwide licensing plan for AI chip exports
The U.S. government drafted proposed export regulations that would require licensing approval for nearly all global shipments of advanced AI accelerators from U.S. companies such as Nvidia and AMD. The draft would replace country-based restrictions with a worldwide, compute-scale-based licensing regime and add compliance requirements such as transparency disclosures and possible inspections.
Related entities
Vulnerabilities, threat actors, malware, products, organizations, and breaches Mallory has linked to this story.
Sources
3 references tracked. Mallory keeps watching after this page renders.
New Commerce Department AI export rules could be seismic change for CSPs and data center operators - buying American GPUs at scale means committing to building American infrastructure | Tom's Hardware
tomshardware.com
Open sourceUS Commerce Department confirms harsh new AI export rules, shoots down reports over the return of Biden-era AI Diffusion rule - DoC to formalize a new approach to strategic AI accelerator export controls | Tom's Hardware
tomshardware.com
Open sourceUS gov't preps sweeping export controls for Nvidia, AMD AI hardware - worldwide licensing system would give Trump admin broad authority to block global sales | Tom's Hardware
tomshardware.com
Open sourceSee the full picture, correlated to your attack surface.
Map indicators from this story to your assets and identify affected systems in minutes.
Every observed campaign, victim, and pivot linked to actors named in this story.
Malware, exploits, and IOCs connected to the activity described here.
YARA, Sigma, and Snort rules deployed to your SIEM as soon as they’re published.
Get matching new stories delivered to your team as they break — not the next morning.
Ask questions about this story and take action on the answers.
Related stories
Other stories Mallory is tracking that overlap on entities, sources, or topic labels.
U.S. Revises Export Controls to Allow Limited Sales of Nvidia H200 and AMD MI325X AI GPUs to China
The U.S. Department of Commerce’s Bureau of Industry and Security (BIS) announced revised export rules that can allow **limited, licensed exports** of certain advanced U.S.-designed AI/HPC accelerators—explicitly including **Nvidia’s H200** and **AMD’s Instinct MI325X**—to China and Macau. The policy shifts from purely performance-threshold gating toward a licensing framework that emphasizes **U.S. supply priority** and compliance assurances, including requirements that exporters demonstrate U.S. demand is met, U.S. orders are not delayed, and that PRC-bound shipments are capped relative to U.S. volumes (e.g., aggregate shipments to China not exceeding **50%** of the same product shipped into the U.S.). BIS also cited conditions such as Chinese purchasers adopting export-compliance procedures (including customer screening) and **independent third-party testing in the U.S.** to verify product performance and security. The change triggered political scrutiny in Congress, with House lawmakers warning that expanded Chinese access to advanced U.S. chips could accelerate Beijing’s AI and military ambitions, even as the administration framed the move as controlled via licensing and review. Separately, reporting indicated Beijing may be **constraining domestic purchases** of H200s to “special circumstances,” potentially limiting access largely to **university R&D labs**, reflecting China’s tension between acquiring leading-edge foreign accelerators and supporting domestic semiconductor development. Commentary from Nvidia CEO Jensen Huang about “god AI” and AI “doomer” narratives was unrelated to the export-control policy and did not add security-relevant detail to the export-rule story.
Mar 21, 2026
US Policy Tightens Controls and Adds Costs on Exports of Advanced AI Accelerators
The U.S. government issued new, compliance-heavy rules governing exports of advanced **AI/HPC accelerators** to China and Macau, keeping a general presumption of denial while allowing limited, case-by-case licensing for narrowly defined parts such as **Nvidia H200** and **AMD Instinct MI325X**. Eligibility is tied to technical thresholds (e.g., total processing performance and memory bandwidth caps) and to supply-side constraints intended to prevent disruption of U.S. availability, including requirements that exports not exceed shipments to U.S. entities and that products be readily available domestically—effectively constraining China-only SKUs and tightening scrutiny of re-exports and certain restricted destinations. Separately, the Trump Administration announced a **25% tariff** on exports of certain “advanced computer chips,” explicitly naming the H200 and MI325X, while carving out domestic usage (U.S.-based datacenters, R&D, startups, public sector, and non-datacenter applications) so U.S. customers are not expected to bear the fee. The policy also adds an exporter burden to demonstrate domestic demand is met before shipping abroad and signals potential further adjustments within 90 days, increasing uncertainty for vendors and customers planning non-U.S. deployments of advanced AI compute.
Mar 21, 2026
U.S. AI Chip Export Controls Shift as ByteDance Expands Access to Nvidia GPUs in Malaysia
The U.S. Commerce Department withdrew an early draft **AI hardware export rule** that would have tied access to advanced American accelerators to foreign investment in U.S. AI infrastructure, while officials said a broader replacement framework is still being developed. The abandoned proposal reportedly envisioned a tiered licensing model based on computing scale, with expedited approval for smaller shipments and tighter conditions for large AI cluster deployments, underscoring that Washington is still recalibrating how it will control the global distribution of high-end AI compute. At the same time, **ByteDance** is moving ahead with access to a **36,000-GPU Nvidia Blackwell cluster** in Malaysia through cloud operator *Aolani Cloud*, in a deal Nvidia said complies with current U.S. export controls. The planned deployment, reportedly worth about **$2.5 billion** and built from **500 NVL72 GB200 systems**, highlights how Chinese firms may still obtain substantial overseas AI compute capacity through third-country infrastructure even as direct chip access remains restricted. A separate analysis of **Huawei’s HarmonyOS** and China’s broader AI ecosystem is not about this specific export-control development and should be excluded.
Apr 2, 2026