The A100 and H100 are NVIDIA's data-center accelerators - the chips behind much of the current wave of large-scale AI compute. When export controls move from broad categories to naming individual products, the policy has reached the level of specific silicon, and NVIDIA's newly filed annual report records exactly that.
The disclosure appears in the FY2023 Form 10-K (sec.gov; accession 0001045810-23-000017). It describes US government rules affecting the ability to develop, produce, and manufacture certain chips for export to China - including Hong Kong and Macau - and Russia, and states the rules specifically impact the company's A100 and H100 integrated circuits.
Why does naming the products matter? A risk factor that gestures at 'geopolitical uncertainty' is boilerplate; a risk factor that names A100 and H100 is the company telling investors that identifiable, high-value parts fall inside the control perimeter. The specificity is the signal.
The mechanism is worth understanding. Export controls on advanced compute work by capping the capability that can be shipped to restricted destinations; when the most capable accelerators are named, the controls are aimed precisely at frontier AI training and inference performance, not at general-purpose chips.
For an explainer reader, the lesson is how to read a maturing controls regime through filings. The progression from general risk language to named products is the trail of policy tightening, and the 10-K is the contemporaneous, legally careful record of where that perimeter sat as of this filing.
The primary source is the sec.gov 10-K; it was surfaced through EdgarBeast, an SEC-filing evidence index. When a risk factor names specific chips, read it closely - it marks the moment export policy became product-specific for the AI-compute leader.