Live AI GPU Rental & LLM Token Price Index
Daily empirical spot rate benchmark for AI compute infrastructure (verified on-demand $/GPU-hr medians across Blackwell, Hopper & Ampere clusters) paired with multi-provider $/1M LLM token deflation and downstream equity surplus rotation.
120-Day Cloud GPU Spot Rental Curve
Liquidity-screened market median $/GPU-hr across B200, H200, H100 SXM, A100 80GB & RTX 5090 cloud clusters.
120-Day Frontier vs. Open-Weight Token Deflation
Standardized enterprise workload $/1M tokens comparing US Frontier Labs against Open-Weight challengers & commodity floor.
Cloud GPU Cluster Spot Rates
Tradestie Cloud Compute Intelligence · Snapshot 2026-11-15
| GPU Tier / Form Factor | Median $/hr | Implied / Mo | 30D % | 90D % | Liquidity Tier |
|---|---|---|---|---|---|
| NVIDIA B200 (192GB SXM) Blackwell · Frontier Reasoning & Training | $10.63 | $7757/mo | +18.0% | +41.6% | Standard Liquidity |
| NVIDIA H200 (141GB SXM) Hopper · Frontier Inference & Training | $5.19 | $3791/mo | +9.6% | +28.1% | Standard Liquidity |
| NVIDIA H200 NVL (141GB PCIe) Hopper · Enterprise PCIe Inference | $4.36 | $3186/mo | +6.9% | +20.9% | Deep Liquidity |
| NVIDIA H100 SXM (80GB) Hopper · Cluster Workhorse Compute | $3.37 | $2463/mo | -5.3% | +1.7% | Thin Liquidity |
| NVIDIA H100 PCIe / NVL (80GB) Hopper · Enterprise Node Compute | — | — | — | — | Insufficient Depth |
| NVIDIA A100 80GB (SXM4) Ampere · Batch & Mid-Tier Inference | $1.07 | $785/mo | +49.1% | +15.0% | Deep Liquidity |
| NVIDIA RTX 5090 (32GB) Blackwell Consumer · Fine-Tuning & Edge Compute | $0.71 | $518/mo | -51.7% | +55.8% | Deep Liquidity |
| NVIDIA RTX 4090 (24GB) Ada Consumer · Quantized Edge Inference | $0.48 | $353/mo | -27.8% | +33.3% | Deep Liquidity |
LLM Inference Provider Pricing ($/1M Tokens)
Tradestie LLM Inference Economics Benchmark · 2026-11-15
| Lab / Public Proxy | Catalog Median | 30D Cut/Chg | Flagship Ask | Budget Floor | Models |
|---|---|---|---|---|---|
|
Anthropic
AMZN (proxy)
Frontier Proprietary
|
$5.50 | +10.0% | $20.00 claude-fable-5.1 | $0.100 | 32 |
|
OpenAI
MSFT (proxy)
Frontier Proprietary
|
$2.25 | 0.0% | $67.50 gpt-5.5-pro | $0.046 | 93 |
|
US Frontier Lab (X)
TSLA (proxy)
Frontier Proprietary
|
$1.98 | +26.7% | $3.00 frontier-x-4.7 | $1.250 | 10 |
|
Google DeepMind
GOOGL
Frontier Proprietary
|
$1.31 | +74.9% | $4.50 gemini-3.1-pro-preview | $0.062 | 36 |
|
Mistral AI
Open-Weight
|
$0.45 | +16.3% | $3.00 mistral-medium-3-5 | $0.029 | 21 |
|
Alibaba Qwen
BABA
Open-Weight
|
$0.41 | +2.7% | $6.00 qwen3.8-max-prime | $0.025 | 42 |
|
DeepSeek
Open-Weight
|
$0.31 | -28.2% | $1.19 deepseek-v4-pro-0813 | $0.072 | 17 |
|
Meta Llama
META
Open-Weight
|
$0.17 | +8.6% | $0.72 llama-3.1-70b-instruct | $0.025 | 8 |
Compute-to-Equity Regime, Neocloud Credit Stress & AI Layer Surplus Rotation
Compute-to-Equity & Neocloud Credit OAS Monitor
Cross-checking GPU rental economics against High-Yield, Single-B (levered data-center tier), and CCC corporate credit spreads (as of 2026-11-14).
AI Layer Surplus Rotation Funnel (Silicon Chokepoints vs. Enterprise Apps)
Evaluated across expanded institutional common-stock baskets ($5+ liquid equities, as of 2026-11-13) reporting both Median 20D Return and 20D Breadth (% Advancing) so single-stock earnings outliers never distort the regime call.
Physical AI Silicon Chokepoints (12)
Sorted by 20D RS| Ticker / Chokepoint | Price | 1D % | 20D % | AI Score | Tools |
|---|---|---|---|---|---|
| ANET Arista Networks AI Back-End Spine Networking | $197.29 | -4.52% | +11.48% | 38.3 | |
| AVGO Broadcom Inc. Custom XPUs & Ethernet Switching | $347.01 | -4.76% | +5.55% | 48.6 |
Index Methodology & Empirical FAQ
For journalists, Substack writers, infrastructure engineers, and equity research desks citing this dataset.
1. What is the current hourly cloud rental price for NVIDIA B200, H200, and H100 GPUs?
As of 2026-11-15, Tradestie Cloud Compute Intelligence liquidity-screened median cluster rental rates stand at $10.63/GPU-hr for NVIDIA B200 (192GB SXM), $5.19/GPU-hr for NVIDIA H200 (141GB SXM), and $3.37/GPU-hr for NVIDIA H100 SXM (80GB). Implied monthly equivalents reflect continuous 24/7 on-demand cluster utilization prior to reserved multi-quarter volume commitments.
2. Why does Tradestie report liquidity-screened market medians instead of minimum GPU spot prices?
Unscreened minimum floor quotes frequently reflect degraded interconnects, constrained networking, or short-duration interruptible capacity that production training and inference clusters cannot deploy. Tradestie applies depth-adjusted liquidity screening across production-grade cloud clusters—suppressing illiquid tiers automatically and labeling shallower books as Thin Liquidity so illiquid outlier quotes never distort the benchmark.
3. How are Frontier vs. Open-Weight LLM $/1M token prices calculated?
All LLM inference prices reflect a standardized enterprise workload token blend ($/1M tokens) calibrated to typical retrieval-augmented generation and agentic tool-calling traffic. The Frontier Proprietary tier tracks the catalog median across major US frontier model labs, while the Open-Weight tier tracks commercial serverless inference medians for leading open-weight model families (currently $0.36/1M vs. $2.12/1M).
4. Why report both Median 20D Return and Breadth (% Advancing) in the AI Layer Rotation Funnel?
In small cap-weighted or 4-stock thematic baskets, a single post-earnings outlier (+45%+) can flip the basket average positive even when 75% of peer equities are declining. Evaluating N=12 Physical Silicon Chokepoints against N=14 Institutional AI Software Platforms using both Median 20D Return and Breadth (% advancing) separates genuine broad sector rotation from isolated earnings dispersion.
Related Tradestie Research & Economics Hubs
Compare Wafer & CoWoS Die Manufacturing Cost vs. Cloud GPU Rental Rates
Inspect true-scale 26×33mm (858 mm²) EUV reticle floorplans, HBM stack BOM teardowns, KV-cache memory walls, and sole-source suppliers across 15 flagship AI chips.
Corporate R&D Intensity & Upstream Foundry Demand Hub
Track quarterly hyperscaler and fabless R&D capital intensity alongside upstream foundry wafer, advanced packaging, and server rack demand trends.
House Thesis Scoreboard & Invalidation Guard
Monitor live conviction scores, falsification guardrails, and empirical hit-rate tracking across active AI silicon, datacenter power, and software theses.