AWS EC2 on-demand GPU hourly rates read from the Price List API (per-GPU-hour, us-east-1)

USD per H100 SXM GPU-HOUR, ON-DEMAND list price. DERIVED, and the derivation is the point: the Price List API file publicationDate is 2026-09-25 and every rate below is an INSTANCE-hour divided by the GPU count of that instance shape. Node hourly / GPUs per node. p5.48xlarge = 8x H100 SXM, $55.04/instance-hr / 8 = $6.88. p5.4xlarge = 1x H100 SXM, $6.88/instance-hr / 1 = $6.88 - the single-GPU shape prices identically to the 8-GPU shape per GPU, which is the cleanest available proof that the division is correct rather than convenient. p5en.48xlarge = 8x H200, $63.296/instance-hr / 8 = $7.912. p4de.24xlarge = 8x A100 80GB, $27.44705/instance-hr / 8 = $3.4309. p4d.24xlarge = 8x A100 (40GB), $21.957642/instance-hr / 8 = $2.7447. g6e.48xlarge = 8x L40S, $30.13118/instance-hr / 8 = $3.7664. g6e.xlarge = 1x L40S, $1.861/instance-hr / 1 = $1.861 - note this is HALF the per-GPU rate of the 8-GPU shape, so L40S has a genuine shape-dependent price and per-GPU figures are not interchangeable across shapes. g6.48xlarge = 8x L4, $13.3504/instance-hr / 8 = $1.6688. g4dn.12xlarge = 4x T4, $3.912/instance-hr / 4 = $0.978. All figures per-GPU-hour, NOT per node, list price, observed 2026-10-04 from a price list published 2026-09-25 PERISHABLE-FIGURE STAMP (added 2026-10-04, uniform across data/supply/ so it is greppable). PRICE CLASS: PUBLISHED LIST PRICE - an asking rate a new customer is quoted, NOT a realised, discounted or negotiated transaction rate. A real bill is at or below this and typically well below it once commitments, credits and committed-use discounts apply. OBSERVED: 2026-10-04 RE-VERIFY BY: 2027-01-04 - the repo's quarterly re-verification bar. After that date this figure is presumed drifted and must be re-checked against its source before it is quoted again. A price without its observation date is noise; this stamp exists so a reader can tell a fresh figure from a stale one without reading prose. UNIT: USD per H100 SXM GPU-HOUR, ON-DEMAND list price READING RULE: No further perishable flag is published for this figure.

lead_time_weeks 8 is AWS's OWN published number, quoted from the EC2 User Guide (source spl-aws-capacity-block-booking-horizon): 'You can reserve a Capacity Block with a reservation start time UP TO EIGHT WEEKS IN THE FUTURE', up to 64 instances per block and 256 across blocks. WHAT IT IS: the MAXIMUM booking horizon for scarce GPU capacity - a FLOOR on how far ahead you must commit - not a typical wait and not a delivery time. It is a capacity-booking lead time; the on-demand capacity on this record still starts immediately, and only the Capacity Block tier requires the 8-week commitment.

EAR 742.6(a)(6)(iii)(A) + EAR 744.23/744.6 (Entity, Unverified, MEU list screening of the end user) + the advanced-computing IaaS obligations of 91 FR 1684 (consignee KYC, and management/limitation of IaaS access to the hardware)

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.

EXPLICIT UNIT DECLARATION, BECAUSE READING A NODE RATE AS A PER-GPU RATE MAKES EVERY NUMBER 8x WRONG: every rate in this record is PER SINGLE GPU PER HOUR. The node hourly figures are given in price_basis purely so the division is auditable. Do not compare $55.04 to a $3.49/GPU-hour Runpod figure. THE SHAPE EFFECT IS REAL AND NOT A DIVISION ARTIFACT: p5.48xlarge and p5.4xlarge both cost $6.88 per H100-hour (32x the CPU for the same GPU price), whereas g6e.48xlarge is $3.77 per L40S-hour against $1.86 for g6e.xlarge - a 2x per-GPU difference between an 8-GPU and a 1-GPU shape of the SAME GPU. So a per-GPU-hour figure is only comparable between providers if you also know the shape it came from; for large shapes AWS charges a premium per GPU and for small ones a discount. The single-GPU H100 shape is the exception that proves the rule. SELECTION FILTER used, so it can be reproduced: operatingSystem=Linux, tenancy=Shared, capacitystatus=Used, licenseModel='No License required', instanceType in the target set. Note g6.48xlarge returned four distinct rates in the raw file ($13.3504, $16.5952, $36.3904, $85.3504) because g6/g6f/g6e families and differing AZ or host-reservation variants share the shape name in the attributes - the $13.3504 figure is the plain Linux on-demand one and the others must not be mixed into a per-GPU figure without re-deriving them. HARDWARE CONTEXT for interpreting the p5 rate: p5.48xlarge is 192 vCPU, 2 TiB RAM, 8 x 3.84 GB NVMe SSD, 3200 Gbps EFA with 900 GB/s NVSwitch; p5.4xlarge is 16 vCPU, 256 GiB, 1 H100 80 GB HBM3, 100 Gbps EFA, no NVSwitch, local storage 3.84 GB NVMe. An 8xH100 node at $55.04/hour is $6.88 per GPU-hour BEFORE storage, egress or managed disks - see rent-hidden-costs-egress-storage-coldload-2026-10 for what those add. EXPORT CONTROL (2026-10-04, mechanism-level; sources spl-bis-2026-05-d5-guidance, spl-bis-2026-01-advanced-computing-case-by-case): renting capacity does NOT remove the export-control question, it relocates it. Providing remote access to controlled computing capability can constitute a transfer under the EAR, so the provider must screen the end user; and separately, under the 91 FR 1684 licence regime the consignee must run KYC procedures to prevent unauthorized remote access and must manage and limit IaaS access to the hardware - obligations pushed onto the customer, who must be told to implement them. For a customer headquartered or parented in Country Group D:5 or Macau the 742.6(a)(6)(iii)(A) end-user/parent test can bite even when the capacity is consumed in a permitted region. Recorded true because the underlying hardware is controlled and the provider carries the screening duty, not because access to a US region is itself restricted. LEAD TIME 2026-10-04, and this is the only cloud figure in the repo that rests on a PUBLISHED number rather than a sales conversation. AWS documents that 'you can reserve a Capacity Block with a reservation start time up to EIGHT WEEKS IN THE FUTURE', each block holding up to 64 instances and up to 256 instances across blocks. Read the direction of that number carefully: 8 weeks is the MAXIMUM booking horizon, so it is a FLOOR on how far ahead scarce GPU capacity must be committed, NOT a typical wait. It is the contractual expression of scarcity - AWS will not warrant GPU capacity further out than 8 weeks. It attaches to the Capacity Block tier specifically, which is also the discount tier already recorded here (p5.48xlarge $41.528 vs $55.04 on-demand, ~25% off), so the shape of the trade is explicit in this record: about 25% cheaper in exchange for committing 8 weeks ahead. Blocks are also only 'available for select instance types in some AWS Regions', so this is not a guarantee for every SKU priced here. On-demand and Spot carry no published lead time at all and remain governed by availability instead.