A licensing requirement, in export-control terms, means a chip cannot be shipped to a covered destination without specific government approval - effectively a case-by-case gate. NVIDIA's latest quarterly filing records that such a requirement was layered onto a subset of its A100 and H100 accelerators, and reading the language shows how controls evolve in steps rather than a single stroke.
The disclosure appears in the Q1 fiscal 2025 Form 10-Q (sec.gov; accession 0001045810-24-000124). It states that in July 2023 the US government informed NVIDIA of an additional licensing requirement for a subset of A100 and H100 integrated circuits, including DGX and other systems or boards that incorporate those chips.
The phrase that rewards attention is 'additional.' The A100 and H100 were already inside the control perimeter from earlier rules; this is a further requirement stacked on a subset of them. Controls here are not a one-time line but an accreting set of restrictions on the same flagship parts.
The reach to 'systems or boards which incorporate' those chips also matters. Controls that follow the silicon into DGX systems and boards close the path of shipping the restricted capability inside a larger product. The mechanism is designed to track the capability, not just the bare chip.
For an explainer reader, the value is in reading filings as a time series. The same product names, an additional requirement, the extension to incorporating systems - this is the texture of a tightening regime, recorded contemporaneously in the 10-Q rather than reconstructed after the fact.
The primary source is the sec.gov 10-Q; it was surfaced through EdgarBeast, an SEC-filing evidence index. To watch export policy evolve, read the successive risk disclosures - here, an 'additional' licensing requirement on chips already controlled.