# AI Compute Tracker — full dataset > The price of compute, tracked. Independent tracker of AI compute economics: GPU rental prices, accelerator specs and cost-efficiency, token prices, and the datacenter buildout. > Every current number inlined below for answer engines — one URL deep. License: Datasets and open data feeds are licensed CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/); reuse with the attribution “AI Compute Tracker (aicomputetracker.com), CC BY 4.0”. Index: https://aicomputetracker.com/llms.txt — curated site map and feed links. Contract: https://aicomputetracker.com/data/ — field schemas, freshness clocks, stability policy. Series IDs: citable class aggregates are named ACT..@; every such ID means posted list/ask prices, not transactions. Three freshness clocks stamp the sections below and are never merged into one "last updated": market snapshot 2026-08-22T02:03:10.544Z (forward board) · live quotes 2026-07-23T05:00:07.996Z (automated quote rows) · core data 2026-07-23 (manual survey, specs, token prices). ## GPU rental quotes Clock: core data 2026-07-23 · automated rows: live quotes 2026-07-23T05:00:07.996Z. All rows are posted list/ask prices, not transactions. Prices are USD per GPU-hour (per chip-hour for TPUs/Trainium — see UNIT). Billing models stay explicit per row and unlike models are never comparable; est = secondary-source snapshot. GPU PROVIDER CLASS BILLING USD/HR UNIT EST B300 RunPod Specialist cloud On-demand 7.39 GPU-hr B300 Nebius Specialist cloud On-demand 7.85 GPU-hr B300 Vast.ai Marketplace Marketplace / variable availability 8.34 GPU-hr B300 AWS (p6-b300) Hyperscaler Capacity Blocks 11.70 GPU-hr B200 RunPod Specialist cloud On-demand 5.89 GPU-hr B200 Hyperstack Specialist cloud On-demand 6.00 GPU-hr B200 Modal Specialist cloud Serverless / per-second 6.25 GPU-hr B200 Vast.ai Marketplace Marketplace / variable availability 6.36 GPU-hr B200 Lambda Specialist cloud On-demand 6.69 GPU-hr B200 Nebius Specialist cloud On-demand 7.15 GPU-hr B200 Together AI Specialist cloud On-demand 8.19 GPU-hr B200 CoreWeave Specialist cloud On-demand 8.60 GPU-hr B200 AWS (p6) Hyperscaler On-demand 14.24 GPU-hr B200 Google Cloud (A4) Hyperscaler Secondary market snapshot 16.11 GPU-hr est GB200 NVL72 CoreWeave Specialist cloud On-demand 10.50 GPU-hr GB200 NVL72 AWS (u-p6e) Hyperscaler Capacity Blocks 10.58 GPU-hr H200 FluidStack Specialist cloud Secondary market snapshot 2.30 GPU-hr est H200 Vast.ai Marketplace Marketplace / variable availability 3.93 GPU-hr H200 Hyperstack Specialist cloud On-demand 3.99 GPU-hr H200 Crusoe Specialist cloud On-demand 4.29 GPU-hr H200 RunPod Specialist cloud On-demand 4.39 GPU-hr H200 Nebius Specialist cloud On-demand 4.50 GPU-hr H200 CoreWeave Specialist cloud On-demand 6.31 GPU-hr H200 AWS (p5en) Hyperscaler On-demand 7.91 GPU-hr H200 Azure (ND v5) Hyperscaler On-demand 10.60 GPU-hr H100 Vast.ai Marketplace Marketplace / variable availability 1.73 GPU-hr H100 Voltage Park Specialist cloud On-demand 1.99 GPU-hr H100 SF Compute Marketplace Marketplace / variable availability 2.04 GPU-hr H100 Hyperstack Specialist cloud On-demand 3.20 GPU-hr H100 RunPod Specialist cloud On-demand 3.29 GPU-hr H100 Nebius Specialist cloud On-demand 3.85 GPU-hr H100 Crusoe Specialist cloud On-demand 3.90 GPU-hr H100 Lambda Specialist cloud On-demand 3.99 GPU-hr H100 Together AI Specialist cloud On-demand 3.99 GPU-hr H100 CoreWeave Specialist cloud On-demand 6.16 GPU-hr H100 AWS (p5) Hyperscaler On-demand 6.88 GPU-hr H100 Oracle OCI Hyperscaler Secondary market snapshot 10.00 GPU-hr est H100 Google Cloud (A3) Hyperscaler Secondary market snapshot 10.98 GPU-hr est H100 Azure (ND v5) Hyperscaler On-demand 12.29 GPU-hr A100 80GB Hyperstack Specialist cloud On-demand 1.35 GPU-hr A100 80GB RunPod Specialist cloud On-demand 1.49 GPU-hr A100 80GB Crusoe Specialist cloud On-demand 2.30 GPU-hr A100 80GB CoreWeave Specialist cloud On-demand 2.70 GPU-hr A100 80GB Lambda Specialist cloud On-demand 2.79 GPU-hr MI355X Vultr Specialist cloud Secondary market snapshot 2.59 GPU-hr est MI355X TensorWave Specialist cloud Secondary market snapshot 2.95 GPU-hr est MI355X Oracle OCI Hyperscaler Secondary market snapshot 8.60 GPU-hr est MI300X Vultr Specialist cloud Secondary market snapshot 1.85 GPU-hr est MI300X RunPod Specialist cloud Secondary market snapshot 1.99 GPU-hr est MI300X Crusoe Specialist cloud On-demand 3.45 GPU-hr MI300X Oracle OCI Hyperscaler Secondary market snapshot 6.00 GPU-hr est L40S RunPod Specialist cloud On-demand 0.99 GPU-hr L40S Crusoe Specialist cloud On-demand 1.50 GPU-hr L40S Nebius Specialist cloud On-demand 1.55 GPU-hr RTX 5090 Salad Marketplace Secondary market snapshot 0.27 GPU-hr est RTX 5090 Vast.ai Marketplace Marketplace / variable availability 0.31 GPU-hr RTX 5090 RunPod Specialist cloud On-demand 0.99 GPU-hr RTX 4090 Salad Marketplace Secondary market snapshot 0.18 GPU-hr est RTX 4090 Vast.ai Marketplace Marketplace / variable availability 0.26 GPU-hr RTX 4090 RunPod Specialist cloud On-demand 0.69 GPU-hr V100 Market average Marketplace Secondary market snapshot 0.99 GPU-hr est TPU v6e Google Cloud Hyperscaler On-demand 2.70 chip-hr TPU v5e Google Cloud Hyperscaler On-demand 1.20 chip-hr TPU v5p Google Cloud Hyperscaler Secondary market snapshot 4.20 chip-hr est Trainium2 AWS (trn2) Hyperscaler Capacity Blocks 2.24 chip-hr ## Forward board Clock: market snapshot 2026-08-22T02:03:10.544Z. Market-implied rental levels reconstructed from Kalshi contract ladders — indicative, not current spot and not executable quotes. Values are USD per GPU-hour. (est) marks a point whose ladder failed a quality gate (thin or interpolated); tenors from Aug '26 onward may include estimates, and the per-point (est) markers are authoritative. Anchors: B200 = Aug 28 weekly; H200 = Aug 28 weekly; H100 = Aug 28 weekly; A100 = Aug 28 weekly; RTX 5090 = Aug '26 monthly avg. TENOR B200 H200 H100 A100 RTX 5090 Anchor 6.91 4.89 2.73 1.12 0.58 (est) Aug '26 6.27 (est) 4.82 2.74 (est) 1.13 (est) 0.58 (est) Sep 6.44 5.00 2.88 1.09 0.54 Oct 6.62 5.07 2.88 1.08 0.54 Nov 6.70 5.10 2.86 1.07 0.54 Dec 6.69 5.07 2.80 1.07 0.54 Jan '27 6.63 5.03 2.75 1.07 0.54 Feb 6.55 5.00 2.76 1.07 0.55 Mar 6.54 4.98 2.80 1.07 0.55 Apr 6.56 4.95 2.83 1.07 0.55 May 6.58 4.93 2.83 1.06 0.54 Jun 6.61 4.96 2.82 1.06 0.52 Jul 6.64 5.03 2.82 1.06 0.50 Aug '27 6.66 5.07 2.83 1.06 0.49 ## Token prices Clock: core data · token-price survey 2026-07-06. USD per million tokens, standard list prices (batch and cached-input discounts run lower). BLENDED = (3 × input + output) / 4 — a typical 3:1 chat/agent mix. est = corroborated from secondary sources. MODEL ORG TIER IN OUT BLENDED EST GPT-5.5 Pro OpenAI frontier 30.00 180.00 67.50 Claude Fable 5 Anthropic frontier 10.00 50.00 20.00 GPT-5.5 OpenAI frontier 5.00 30.00 11.25 Claude Opus 4.8 Anthropic frontier 5.00 25.00 10.00 Gemini 3.1 Pro Google frontier 2.00 12.00 4.50 GPT-5.4 OpenAI workhorse 2.50 15.00 5.63 Claude Sonnet 5 Anthropic workhorse 2.00 10.00 4.00 Gemini 3.5 Flash Google workhorse 1.50 9.00 3.38 Grok 4.3 xAI workhorse 1.25 2.50 1.56 Claude Haiku 4.5 Anthropic budget 1.00 5.00 2.00 GPT-5.4 mini OpenAI budget 0.75 4.50 1.69 DeepSeek V4 Pro DeepSeek budget 0.44 0.87 0.55 Gemini 3.1 Flash-Lite Google budget 0.25 1.50 0.56 GPT-5.4 nano OpenAI budget 0.20 1.25 0.46 DeepSeek V4 Flash DeepSeek budget 0.14 0.28 0.18 ## Accelerator specs Clock: core data 2026-07-23. FLOPS are DENSE tensor TFLOPS by precision (FP16/BF16, FP8, FP4) — never vendor sparsity numbers, and precisions are never mixed in one comparison. Memory in GB, bandwidth in TB/s, TDP in watts. — = unsupported or undisclosed; est = row contains estimated figures. VENDOR CHIP LAUNCH BF16 FP8 FP4 MEM GB BW TB/S TDP W EST NVIDIA V100 SXM2 2017-06 125 — — 32 0.9 300 NVIDIA A100 SXM 80GB 2020-11 312 — — 80 2.04 400 NVIDIA H100 SXM 2022-10 989 1979 — 80 3.35 700 NVIDIA H200 SXM 2024-06 989 1979 — 141 4.8 700 est NVIDIA B200 (HGX) 2024-12 2250 4500 9000 192 8 1000 est NVIDIA B300 (Blackwell Ultra) 2025-07 2250 4500 15000 288 8 1400 est NVIDIA Rubin (VR200) 2026-09 — — 25000 288 20 1800 est AMD Instinct MI300X 2023-12 1307 2614 — 192 5.3 750 est AMD Instinct MI325X 2024-12 1307 2614 — 256 6 1000 est AMD Instinct MI355X 2025-10 5000 10100 20100 288 8 1400 est AMD Instinct MI400 2026-12 — 10000 20000 432 19.6 — est Google TPU v5e 2023-08 197 — — 16 0.82 — Google TPU v5p 2023-12 459 — — 95 2.77 — Google TPU v6e (Trillium) 2024-10 918 — — 32 1.64 — Google TPU v7 (Ironwood) 2025-11 — 4614 — 192 7.37 — est Intel Gaudi 3 2024-09 1835 1835 — 128 3.7 900 AWS Trainium2 2024-12 650 1300 — 96 2.9 500 est AWS Trainium3 2025-12 1260 2520 — 144 4.9 — est Microsoft Maia 200 2026-01 — 5000 10000 216 7 750 est NVIDIA GeForce RTX 4090 2022-10 330 661 — 24 1.01 450 NVIDIA GeForce RTX 5090 2025-01 419 838 1676 32 1.79 575 ## Citable class-aggregate series Clock: core data 2026-07-23 (aggregates recomputed from the current quote table). Every ID means posted list/ask prices, not transactions. A comparison-policy version bump mints a new ID; superseded IDs stay valid names for their published data. 21 of 23 registered series have a current aggregate; the rest are archived history only and are suppressed here rather than fabricated. SERIES ID GPU CLASS BILLING AGG USD/HR UNIT ACT.B300.SPEC-MED@primary-on-demand-v1 B300 specialist on-demand median 7.62 GPU-hr ACT.B300.MKT-FLOOR@marketplace-access-v1 B300 marketplace marketplace-variable floor 8.34 GPU-hr ACT.B200.HYP-FLOOR@primary-on-demand-v1 B200 hyperscaler on-demand floor 14.24 GPU-hr ACT.B200.SPEC-MED@primary-on-demand-v1 B200 specialist on-demand median 6.92 GPU-hr ACT.B200.MKT-FLOOR@marketplace-access-v1 B200 marketplace marketplace-variable floor 6.36 GPU-hr ACT.GB200-NVL72.SPEC-MED@primary-on-demand-v1 GB200 NVL72 specialist on-demand median 10.50 GPU-hr ACT.H200.HYP-FLOOR@primary-on-demand-v1 H200 hyperscaler on-demand floor 7.91 GPU-hr ACT.H200.SPEC-MED@primary-on-demand-v1 H200 specialist on-demand median 4.39 GPU-hr ACT.H200.MKT-FLOOR@marketplace-access-v1 H200 marketplace marketplace-variable floor 3.93 GPU-hr ACT.H100.HYP-FLOOR@primary-on-demand-v1 H100 hyperscaler on-demand floor 6.88 GPU-hr ACT.H100.SPEC-MED@primary-on-demand-v1 H100 specialist on-demand median 3.88 GPU-hr ACT.H100.MKT-FLOOR@marketplace-access-v1 H100 marketplace marketplace-variable floor 1.73 GPU-hr ACT.A100-80GB.SPEC-MED@primary-on-demand-v1 A100 80GB specialist on-demand median 2.30 GPU-hr ACT.MI300X.SPEC-MED@primary-on-demand-v1 MI300X specialist on-demand median 3.45 GPU-hr ACT.L40S.SPEC-MED@primary-on-demand-v1 L40S specialist on-demand median 1.50 GPU-hr ACT.RTX-5090.SPEC-MED@primary-on-demand-v1 RTX 5090 specialist on-demand median 0.99 GPU-hr ACT.RTX-5090.MKT-FLOOR@marketplace-access-v1 RTX 5090 marketplace marketplace-variable floor 0.31 GPU-hr ACT.RTX-4090.SPEC-MED@primary-on-demand-v1 RTX 4090 specialist on-demand median 0.69 GPU-hr ACT.RTX-4090.MKT-FLOOR@marketplace-access-v1 RTX 4090 marketplace marketplace-variable floor 0.26 GPU-hr ACT.TPU-V6E.HYP-FLOOR@primary-on-demand-v1 TPU v6e hyperscaler on-demand floor 2.70 chip-hr ACT.TPU-V5E.HYP-FLOOR@primary-on-demand-v1 TPU v5e hyperscaler on-demand floor 1.20 chip-hr