Air-cooled vs liquid-cooled is a per-SKU form-factor fact, not an architecture fact: H100 PCIe is dual-slot air-cooled, H100 SXM is a bare module

STRUCTURALLY NOT APPLICABLE, and this is the distinction that matters: the lead time for this record's subject is set by construction, electrical service and cooling/CDU plumbing, not by GPU allocation. The accelerator is not the scarce item, so an accelerator lead time is the wrong question - the record's own notes make this argument. Null here means 'wrong question', not 'no data'.

EAR 3A090.a / 3A090.b (advanced-computing ICs) + EAR 742.6(a)(6)(iii)(A) (D:5/Macau end-user and ultimate-parent) + EAR 744.23 (Entity List screening) + EAR 744.6 (Unverified/MEU lists) (governs the most-restricted part on this record; the workstation-class parts on the same record fall under EAR 3A001 or EAR99 (below the 3A090.a TPP / performance-density thresholds) + EAR 744.23 (Entity List) + EAR 744.6 (Unverified / MEU lists))

LAYER 1, EXPORT LICENCE: under the Jan 2026 BIS rule 91 FR 1684 (effective 2026-01-15) the license review policy for advanced-computing commodities with TPP < 21,000 AND total DRAM bandwidth < 6,500 GB/s - the NVIDIA H200 and AMD MI325X are the rule's named examples - moved from PRESUMPTION OF DENIAL to CASE-BY-CASE REVIEW, but only for exports FROM THE UNITED STATES to end users located in China or Macau, and only against six exporter certifications filed via SNAP-R before shipment: (1) items below the thresholds with the count of units shipped in the US; (2) sufficient US supply, judged 'taking into account normal lead times'; (3) no diversion of global foundry capacity away from US end users; (4) aggregate China+Macau shipments <= 50% of total US end-use shipments of that product; (5) no end-user or use prohibition; (6) the consignee runs rigorous KYC. Every shipment must then pass a qualified third-party testing lab - US-headquartered, not controlled by a D:5/Macau entity, no ownership stake in any party, qualification revocable by BIS at any time - verifying TPP, total DRAM bandwidth, interconnect bandwidth and copackaged DRAM capacity. Reexports and in-country transfers of the same chips to China/Macau REMAIN PRESUMPTION OF DENIAL. LAYER 2, END-USER AND PARENT SCREENING: per BIS guidance of 2026-05-31 a licence is required where the consignee is HEADQUARTERED in Country Group D:5 (15 CFR part 740 supplement no. 1) or Macau OR whose ULTIMATE PARENT COMPANY is so headquartered - even when the consignee itself sits outside D:5/Macau. The requirement sits at EAR 742.6(a)(6)(iii)(A) for destinations outside the US, with an exception available only under the license exception at 740.2(a)(9)(ii). The destination country alone therefore never answers the question; screen the parent. LAYER 3, REMAINING LIMITS: exports to D:5/Macau-headquartered OR -parented entities stay PRESUMPTION OF DENIAL wherever the buyer sits, and the AI Diffusion Rule that would have set a country-tier structure on who could lawfully buy what was RESCINDED on 2025-05-13 and is not being enforced, so any recorded status rests on the instruments named here rather than on tiering. PER-RECORD QUALIFIER: Both parts on this record are 3A090-class: the PCIe H100 is the same die as the SXM module, and export status attaches to the CONTROLLABLE ITEM rather than the die, so the two form factors ship under different licensing analyses even though the silicon matches. The record's own point is that a buyer cannot assume a PCIe card inherits the SXM module's clearance or the reverse.

THIS RECORD EXISTS TO KILL A COMMON ERROR IN POWER-DENSITY PLANNING: THE MENTAL MODEL THAT 'HOPPER IS AIR-COOLED AND BLACKWELL IS LIQUID-COOLED' IS WRONG AT THE PART LEVEL. NVIDIA's H100 datasheet form-factor row reads, verbatim and side by side: 'SXM' for H100 SXM and 'PCIe dual-slot air-cooled' for H100 PCIe. The word 'air-cooled' is attached to a specific SKU's physical form factor, in the same row as its Max TDP of 350 W, not to the Hopper architecture as a whole. So: H100 PCIe at 350 W IS air-cooled and goes in a conventional server in a conventional colo with no liquid anywhere. H100 SXM at 700 W is a bare module with no cooler, normally deployed in an HGX baseboard or DGX system, which is itself air-cooled at the node level - see supply/dgx-h100-h200-10kw-air-cooled-node, which documents a 10.2 kW max air-cooled 8-GPU node with 1105 CFM of front-to-back airflow. The H100 NVL sits between the two, typically shipped as a PCIe card with a factory cooler. THE CONSEQUENCE FOR ANY POWER-DENSITY DECISION: you cannot ask 'is this chip air-cooled or liquid-cooled'. You have to ask what the specific SKU's form factor and TDP are, then what the host chassis is, then what the facility is. A 350 W air-cooled H100 PCIe and a 700 W SXM H100 are both 'Hopper', differ by 2x in power, and can have completely different facility requirements. ALSO WORTH FLAGGING FOR POWER SIZING: NVIDIA's HGX H100/H200 datasheet publishes NO TDP and NO rack power figure at all - only the Hopper SXM datasheet gives 700 W - so anyone power-sizing an HGX baseboard must derive it from the SXM datasheet and mark it as derived rather than vendor-stated. That asymmetry between generations is itself a signal: every Blackwell platform datasheet publishes Max TDP per GPU, while the Hopper HGX datasheet publishes none, because Blackwell customers had to spec for facility integration from the start. EXPORT CONTROL (2026-10-04, mechanism-level; sources spl-eccn-3a090-ccl, spl-bis-2026-01-advanced-computing-case-by-case): this record compares the AIR-COOLED PCIe form factor against the liquid-cooled SXM module, and the two are NOT equivalent under export control - which is the procurement point of the comparison. The PCIe H100 is the same die as the SXM part, but export status attaches to the CONTROLLABLE item, so the two ship under different ECCN treatment and different licensing analysis; a buyer cannot assume a PCIe card inherits the SXM module's clearance or vice versa. Both are 3A090-class, so the three-layer mechanism applies: US export licence (case-by-case for TPP < 21,000 and DRAM bandwidth < 6,500 GB/s parts per 91 FR 1684), the 742.6(a)(6)(iii)(A) end-user/ultimate-parent test that reaches entities outside D:5, and the surviving presumption of denial on reexports and on D:5/Macau-headquartered or -parented buyers. Recorded true for the controlled datacenter parts; the NVL variant is likewise controlled as an integrated system. LEAD TIME: null - and for this record that is the WRONG QUESTION, which is the point worth recording. This record measures a build constraint that is not the accelerator supply chain: the lead time for a 120 kW liquid-cooled rack is set by construction, electrical service, and CDU/manifold plumbing, not by GPU allocation. Its own notes already make this argument for the colo case, so this entry only fixes the boundary: do not read a GPU lead time into a facility lead time, and do not read a facility lead time into a GPU lead time. No vendor publishes a weeks figure for either, so both stay null - probe log in source spl-negative-no-oem-publishes-lead-time. Where a real published horizon exists for the supply side it is on the accelerator record (NVIDIA's $95.2bn purchase and supply/capacity obligations substantially payable through FY2027, source spl-nvda-10k-fy2026-supply-commitments).