MARKET SUMMARY: GPU price bands by segment, observed 2026-10-03 (new / used / rental)

lead_time_weeks is null because no lead time applies: this is a secondary/auction market priced per listing. What a buyer faces is listing availability at the moment of purchase - the record's availability field is a snapshot of what a seller posted, not a schedulable delivery date - plus whatever shipping the listing states. An auction close date is not a lead time.

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)

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.

AGGREGATE RECORD, price_usd DELIBERATELY NULL. One point estimate for 'the GPU market' would be a fabrication; this record exists to hold the segment separation that the per-segment records above cannot express individually. OBSERVATION DATE FOR EVERY FIGURE BELOW: 2026-10-03. All sources are the sup-price-* records cited in the sibling records. BAND 1 - NEW CONSUMER RETAIL (sealed, retail channel, amazon.com list only): RTX 5090 32GB $6,550-$6,750 (n=2). RTX 5080 $1,550-$1,804 (n=10). Smallest and cleanest channel; also the least representative (one retailer). BAND 2 - USED-CONSUMER ASKING (resale marketplace, condition UNKNOWN, not sold): RTX 4090 24GB $1,800-$5,000, quartiles $2,693/$3,299/$3,600 (n=35). RTX 3090 24GB $899-$2,500, quartiles $1,300/$1,514/$1,600 (n=72). Note both run AT OR ABOVE their new-retail MSRP, which is the crypto/LLM-demand effect and is only visible because MSRP is held separately as a non-market reference (sup-price-nanoreview-msrp). BAND 3 - USED-DATACENTER ASKING (resale marketplace, bare card, not sold): H100 80GB PCIe $32,000-$47,520, quartiles $40,880/$43,000/$45,000 (n=8). H100 80GB SXM5 module $24,118-$28,888, median $26,880 (n=5). A100 80GB $16,798-$34,999, quartiles $19,995/$21,995/$29,000 (n=17). H800 80GB PCIe $33,900-$41,221 (n_effective=2). BAND 4 - WHOLE-NODE / CONFIGURED SYSTEM (refurbished, integrator channel): Dell XE9680 8x H100 SXM5 $180,000/node (n=1, listing sold out). CORRECTION THIS RECORD EXISTS TO PREVENT: $180,000 is the node, not the card. Dividing by 8 gives $22,500 per GPU for the GPU content of a node - BELOW the SXM asking median of $26,880 and roughly HALF the PCIe asking median of $43,000. Reading either number as the other is an 8x or 2x error respectively. BAND 5 - PEER-TO-PEER RENTAL, per GPU-hour (verified hosts, asking, not realised): B200 $6.25-$9.38 (8-GPU median $7.97) > H200 $3.68-$6.05 (8-GPU median $5.00) > H100 SXM $1.69-$3.50 (median $2.47) > L40S $0.53-$0.82 (median $0.71) ~ RTX 6000 Ada $0.53-$0.80 (median $0.63) > RTX 5090 $0.40-$0.74 (median $0.49) > RTX 4090 $0.30-$0.94 (median $0.43) > RTX 3090 $0.12-$0.25 (median $0.15). BAND 6 - MANAGED CLOUD LIST (provider list rate, instance incl. CPU/RAM, not per-card): H100 SXM $3.49/hr, H100 NVL $3.19/hr, H100 PCIe $2.89/hr, A100 PCIe $1.59/hr, RTX PRO 6000 Blackwell $2.09/hr. Sits above the peer-to-peer H100 SXM band - the managed-service premium, not a hardware price. BAND 7 - INSTITUTIONAL INDEX (condition NOT stated by publisher): Stoa public index, observed 2026-10-03: A100 80GB $9,490, H100 SXM5 $25,860, H200 SXM $33,850, B200 $43,860, GB200 NVL $67,400. This is the ONLY source in the slice that presents itself as an index rather than a listing scrape, and it is also the only one whose condition basis is unstated - which is why it is quarantined here rather than used to price anything. Its A100 level ($9,490) sits below the entire used-A100 asking band; its H100 SXM5 level ($25,860) sits inside this slice's SXM asking band ($24,118-$28,888). That partial agreement is the reason to keep it and the reason not to trust it alone. WHAT THIS RECORD DELIBERATELY DOES NOT CLAIM: no sold/transaction price for any GPU is recorded anywhere in this slice, because every reachable source serves asking or list prices (see notes on sup-price-picclick-ebay-asking and sup-price-stoa-index). No datacenter-GPU depreciation curve by age is recorded: no such published study was reachable. No record conflates a consumer card with a datacenter card. Sample sizes are published per band because most are small, and the small ones (n<=8) are marked as such rather than being presented as stable prices. EXPORT CONTROL (2026-10-04, mechanism-level; sources spl-eccn-3a090-ccl, spl-bis-2026-01-advanced-computing-case-by-case, spl-bis-2026-05-d5-guidance): these parts classify under ECCN 3A090.a/.b, whose thresholds are 'total processing performance' (TPP) >= 4800, or TPP >= 1600 with performance density >= 5.92 (the .b band is TPP 2400-4800 with performance density 1.6-5.92, or TPP >= 1600 with performance density 3.2-5.92). The mechanism has THREE layers and all three must be cleared, not just one: (a) LICENSE FOR EXPORT FROM THE US - controlled by the January 2026 rule (91 FR 1684, effective 2026-01-15), which moved TPP < 21,000 AND total DRAM bandwidth < 6,500 GB/s parts from presumption of denial to CASE-BY-CASE REVIEW, 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', no diversion of foundry capacity away from US end users, aggregate China+Macau shipments capped at 50% of US end-use shipments, no end-user prohibition, consignee KYC procedures, and prior verification by a qualified third-party US testing lab of TPP, total DRAM bandwidth, interconnect bandwidth and copackaged DRAM capacity; (b) END-USER AND PARENT-COMPANY SCREENING under 742.6(a)(6)(iii)(A) - per BIS guidance of 2026-05-31 a licence is required where the consignee is HEADQUARTERED in Country Group D:5 or Macau OR ITS ULTIMATE PARENT is so headquartered, EVEN IF the consignee itself is outside D:5/Macau, which means the destination country alone never answers the question; (c) THE LIMITS THAT REMAIN - reexports and in-country transfers of the same chips to China/Macau remain PRESUMPTION OF DENIAL, and exports to D:5/Macau-headquartered or -parented entities remain presumption of denial wherever the buyer sits. The Jan-2025 AI Diffusion Rule was rescinded and not enforced from 2025-05-13, but that rescission did not remove (a) or (b). LEAD TIME: null, and deliberately so. A marketplace or big-box reseller publishes STOCK STATE, not a lead time - where this record shows in-stock that means the listing was available on the observation date, and where it shows backorder no week window is published at all. Converting one into the other would be an invention. The one exception in the whole chain is CDW, which does print windows (source spl-cdw-published-lead-times-gpu-2026-10); note that even there the scarcest SKUs - the RTX PRO 6000 96GB cards and DGX Spark - read 'Item Backordered' with NO estimate, so scarcity and lead time are inversely related in what distributors actually publish. Probe log in source spl-negative-no-oem-publishes-lead-time.