NVIDIA HGX platform as sold inside OEM and system-builder servers

NEGATIVE, DELIBERATELY NOT STRUCTURAL: this is a quote-only channel - no price and no lead time is published on any page, checked 2026-10-04 per source spl-negative-no-oem-publishes-lead-time. The lead time exists inside a sales conversation; it is simply not printed. Contrast with the on-demand cloud records, where null means the concept does not apply.

EAR 742.6(a)(6)(iii)(A) + EAR 744.23 (Entity List) + EAR 744.6 (Unverified / MEU lists) + EAR Regional Stability (RS) controls for Country Groups D:1, D:4, D:5 (88 FR 73458, Oct 2023)

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. UPGRADE NOTE: this record's earlier text cited only the Oct 2023 interim final rule (88 FR 73458) and a presumption of denial, with no paragraph and no mechanism. That is retained here as history: what changed is that the Jan 2026 rule (91 FR 1684) carved the TPP < 21,000 / DRAM < 6,500 GB/s band - which includes the H200 and MI325X on this record - out of presumption of denial into case-by-case review for US exports to China/Macau end users, and that BIS's 2026-05-31 guidance fixed the end-user/parent limb at 742.6(a)(6)(iii)(A).

QUOTE-ONLY. The HGX platform page itself funnels buyers to a marketplace 'Explore Systems' link rather than to any price, so no figure is recorded. The genuinely useful fact here is the BASIS: HGX is a baseboard+GPU platform, so when you buy 'an HGX system' the accelerators come WITH the system and are not separable line items. NVIDIA notes HGX Rubin NVL8 integrates eight Rubin GPUs on sixth-generation NVLink, can be paired with NVIDIA Vera CPUs (HGX Vera Rubin NVL8) or with x86 CPU baseboards. Practical sourcing consequence: there is no such thing as ordering one H100 SXM or B200 module from an OEM; the minimum purchase unit is a 4- or 8-GPU node, which is exactly why per-GPU numbers quoted against this channel are 5-10x higher than bare-card prices elsewhere. Lead-time rather than in-stock is the realistic status for new-generation HGX; no week count is published so lead_time_weeks is null. EXPORT CONTROLLED (true). Regime: US BIS advanced-computing export controls (the Oct 2023 interim final rule, 88 FR 73458 - source sup-buy-bis-2023-advanced-computing-controls), which put datacenter accelerators under ECCN 3A090/4A090, added the advanced-computing Foreign Direct Product (FDP) rule, and extended Regional Stability (RS) controls to Country Groups D:1, D:4 and D:5. For items destined to Macau or Country Group D:5 the rule applies a PRESUMPTION OF DENIAL; other RS destinations get a presumption of approval. Practical effect: which accelerator you can legally be sold, and which SKU counts as compliant, is set by destination and end user - not by the channel. LEAD TIME: still null as a week count, and deliberately so - NVIDIA publishes no lead time in weeks anywhere, so any number here would be invented. The honest bound comes from NVIDIA's own SEC filing (source spl-nvda-10k-fy2026-supply-commitments): as of 2026-01-25 outstanding inventory purchase and long-term supply and capacity obligations were $95.2 BILLION, 'of which substantially all will be paid through fiscal year 2027', against $21.4 billion of consolidated inventories. That is a forward commitment covering roughly four quarters, which is the floor a buyer must plan around. The same filing states NVIDIA 'may place non-cancellable inventory orders for certain product components IN ADVANCE OF OUR HISTORICAL LEAD TIMES, pay premiums, or provide deposits to secure future supply and capacity' - i.e. NVIDIA itself buys further ahead than its own lead times, so a quoted lead time is a lower bound, never a promise. Supplier concentration is the reason no week count exists: wafers from TSMC and Samsung, memory from SK Hynix, Micron and Samsung, CoWoS packaging, and assembly/test at Hon Hai, Wistron and Fabrinet. Re-verify against a fresh 10-Q before trusting any planning figure. EXPORT CONTROL REGIME, UPDATED 2026-10-04 (sources spl-bis-2026-01-advanced-computing-case-by-case, spl-bis-2026-05-d5-guidance, spl-bis-2025-05-ai-diffusion-rescission, spl-eccn-3a090-ccl). The mechanism has moved three times and the OLD paragraph above describing a flat presumption of denial is now incomplete. Current state, in order of application: (1) CLASSIFICATION - HGX/Hopper/Blackwell datacenter parts are ECCN 3A090.a/.b, whose thresholds are 'total processing performance' (TPP) >= 4800, or TPP >= 1600 with performance density >= 5.92; 3A090.b is the lower band (TPP 2400-4800 with performance density 1.6-5.92, or TPP >= 1600 with performance density 3.2-5.92). (2) END-USER / PARENT-COMPANY CONTROL (742.6(a)(6)(iii)(A)) - per BIS guidance of 2026-05-31, a licence is required for advanced computing items destined to entities HEADQUARTERED in Country Group D:5 or Macau, OR WHOSE ULTIMATE PARENT is so headquartered, EVEN IF the entity itself sits outside D:5. This is the limb that makes the answer entity-dependent rather than geography-dependent, and it survives the AI Diffusion Rule's rescission. (3) THE JANUARY 2026 RELAXATION, AND ITS LIMITS - 91 FR 1684 (effective 2026-01-15) moved TPP < 21,000 AND 'total DRAM bandwidth' < 6,500 GB/s parts - naming H200 and AMD MI325X - from presumption of denial to CASE-BY-CASE, but only for exports FROM THE US to end users in China/Macau, and only against six exporter certifications: US supply sufficiency judged 'taking into account normal lead times' (the one place the regulator ties licence approval to lead time), no diversion of foundry capacity from US end users, aggregate China+Macau shipments capped at 50% of US end-use shipments, no end-user prohibition, consignee KYC, and prior third-party US-lab verification of TPP, DRAM bandwidth, interconnect bandwidth and copackaged DRAM capacity. (4) WHAT DID NOT RELAX - reexports and in-country transfers of the same chips to China/Macau REMAIN presumption of denial, and exports to D:5- or Macau-headquartered/parented entities remain presumption of denial regardless of where the buyer sits. AND THE OUTCOME, WHICH IS THE PART THAT DECIDES A PROCUREMENT (source spl-nvda-10q-fy2027q2-h200-licence-outcome): the licence regime exists on paper and NVIDIA has been granted H200 licences under it, and it still moved almost nothing. NVIDIA reports those sales were 'restricted by the PRC government', made 'a fraction of the allowed shipments', and that both those shipments and all Data Center Hopper shipments to China in the quarter were LESS THAN 1% of Data Center revenue - alongside a $0.4 billion H200 excess-inventory and purchase-obligation charge. The licences also require US inspection before shipment and therefore attract a 25% tariff on importation into the US, which NVIDIA says it cannot pass on. So: a licence is necessary but not sufficient, destination-country import approval was the binding constraint, and the inspection-plus-tariff mechanics add both cost and dwell time to any licensed channel. Export control here is not a yes/no on the SKU - it is a destination-and-end-user-and-consignee-screening question that also carries a tariff and an inspection step. LEAD TIME, recorded as an EXPLICIT NEGATIVE with a bound rather than a fabricated week count. NVIDIA publishes no lead time in weeks on any page, and this record is quote-only, so lead_time_weeks stays null: any number would be invented. What CAN be said from a primary source is the shape of the horizon. NVIDIA's FY2026 Form 10-K (filed 2026-02-25, source spl-nvda-10k-fy2026-supply-commitments) reports outstanding inventory purchase and long-term supply and capacity obligations of $95.2 BILLION as of 2026-01-25, 'of which substantially all will be paid through fiscal year 2027', against $21.4 billion of consolidated inventories. A forward book of that size against roughly one year of payments implies an order horizon on the order of several quarters, so plan in quarters rather than weeks. Two further reasons the true lead time is longer than any published figure: the 10-K states NVIDIA 'may place non-cancellable inventory orders for certain product components IN ADVANCE OF OUR HISTORICAL LEAD TIMES, pay premiums, or provide deposits to secure future supply and capacity' - the vendor itself buys beyond its own lead times - and supply is concentrated in TSMC/Samsung wafers, SK Hynix/Micron/Samsung memory, CoWoS packaging and Hon Hai/Wistron/Fabrinet assembly. Corroborating but NOT a substitute: CDW, one of the few distributors that prints a lead time, shows '4-6+ Weeks' for the NVIDIA DGX B200 8x180GB server (source spl-cdw-published-lead-times-gpu-2026-10) - that is CDW's inbound window on one SKU, an open-ended range, and it contradicts the multi-quarter horizon above, which is precisely why a distributor's number must not be read as the OEM's. Re-verify each quarter against a fresh 10-Q. EXPORT CONTROL REGIME: SCHEME RECLASSIFIED (2026-10-04): was a bare boolean plus 'EXPORT CONTROLLED (true). Regime: the Oct 2023 interim final rule, 88 FR 73458' with no paragraph, no mechanism and no expiry. Now recorded as a citation plus the three-layer mechanism.