H100 IndexWhat it costs to rent one H100 GPU for an hour on demand — the market’s anchor price.
- Value
- $4.29 / H100-hr
- Last Price Change
- May 27, 2026
Morning Brief
Morning Brief: Saturday, June 20, 2026
Start-of-day AI compute market report.
Published each morning with the prior day’s source-backed signals across GPU pricing, cloud capacity, power constraints, data centers, and AI infrastructure. Track the prior day’s source-backed news and changes shaping GPU supply, cloud availability, power constraints, data centers, and compute market pressure.
Morning Brief
Morning Tape
- Report date
- Saturday, June 20, 2026
- Signal window
- Friday, June 19, 2026
- Announcing Amazon EC2 G7 instances accelerated by NVIDIA RTX PRO 4500 Blackwell Server Edition GPUs
- Nuclear physics research lab Jefferson Lab breaks ground on 30,000 sq ft data center
- Plans filed for three-building data center campus in Northumberland, UK
Today's Compute College lesson
What is Model Training Cost?
Model training cost is the compute expense required to teach or improve an AI model from data.
Why it matters today: Today's lesson connects the AI Compute 101 track to the compute market readers are watching.
Read the lessonLearn the Market
Term of the Day: High-bandwidth memory
High-bandwidth memory, often called HBM, is advanced memory placed close to AI accelerators so data can move quickly. AI chips need both compute power and fast memory access. If HBM supply is tight, chip production and AI cluster deployment can be constrained.
Market signals
Compute market signals
Spot Compute PriceThe cheaper, interruptible GPU price — a read on whether spare capacity is available.
- Value
- $2.46 / H100e-hourH100e-hour means an H100-equivalent GPU hour, a normalized unit used to compare different GPUs and providers.
- Last Price Change
- May 27, 2026
Model Serving CostRoughly what it costs to run a model and generate its output at scale, per million tokens.
- Value
- $1.74 / 1M tokens
- Last Price Change
- Jun 15, 2026
GPU AvailabilityHow easy or hard it is to actually get GPU capacity right now, scored 0–100.
- Value
- 78 / 100 score
- Last change
- May 27, 2026
Signals
Top market signals
These are the highest-impact source-backed items from the signal window used for this morning’s brief.
Jun 18, 2026
Announcing Amazon EC2 G7 instances accelerated by NVIDIA RTX PRO 4500 Blackwell Server Edition GPUs
Watch item for AI compute supply, demand, capacity, infrastructure, or market structure.
Jun 19, 2026
Nuclear physics research lab Jefferson Lab breaks ground on 30,000 sq ft data center
Watch item for AI compute supply, demand, capacity, infrastructure, or market structure.
Jun 19, 2026
Plans filed for three-building data center campus in Northumberland, UK
Watch item for AI compute supply, demand, capacity, infrastructure, or market structure.
Jun 19, 2026
California startup Orbital joins space data center craze with $5m pre-seed
Watch item for AI compute supply, demand, capacity, infrastructure, or market structure.
Jun 19, 2026
AWS inks recycled water supply agreement with Greater Western Water for planned data center in Melbourne, Australia
Watch item for AI compute supply, demand, capacity, infrastructure, or market structure.
Power & Data Center Watch
- Nuclear physics research lab Jefferson Lab breaks ground on 30,000 sq ft data center (Data Center Dynamics).
- Plans filed for three-building data center campus in Northumberland, UK (Data Center Dynamics).
- Aston Power raises $20m to scale private power platform for data centers (Data Center Dynamics).
Watch next
Signals for the next issue
- Applied Digital Reaches Significant Milestone, Surpassing 1 GW of Contracted Capacity with U.S. Based High Investment-Grade Hyperscaler Lease at Fourth Campus, Polaris Forge 3 (Applied Digital Press Releases).
- MLPerf Training v6.0: Lambda delivers fastest LLM training on NVIDIA GB300 NVL72 and fastest MoE training on NVIDIA HGX B200 (Lambda Blog).
- Nebius expands in UK with more NVIDIA-powered infrastructure, more customers, and more cloud capabilities for agentic and enterprise AI June 8, 2026 (Nebius Newsroom).
- How FERC’s Large-Load Interconnection Actions Help Address Grid Stress, Improve Affordability (NVIDIA Blog).
- DOJ intervenes on behalf of xAI in data center gas turbine lawsuit (Utility Dive).
Recent Briefs
Last 10 Briefs
Morning Brief: Saturday, June 20, 2026
Signal window: Friday, June 19, 2026. Announcing Amazon EC2 G7 instances accelerated by NVIDIA RTX PRO 4500 Blackwell Server Edition GPUs
Morning Brief: Friday, June 19, 2026
Signal window: Thursday, June 18, 2026. How FERC’s Large-Load Interconnection Actions Help Address Grid Stress, Improve Affordability
Morning Brief: Thursday, June 18, 2026
Signal window: Wednesday, June 17, 2026. New 339-mile transmission line brings Canadian hydropower to NYC
Morning Brief: Wednesday, June 17, 2026
Signal window: Tuesday, June 16, 2026. MLPerf Training v6.0: Lambda delivers fastest LLM training on NVIDIA GB300 NVL72 and fastest MoE training on NVIDIA HGX B200
Morning Brief: Tuesday, June 16, 2026
Signal window: Monday, June 15, 2026. Bitdeer breaks ground on natural gas plant and data center in Alberta, Canada
Morning Brief: Monday, June 15, 2026
Signal window: Sunday, June 14, 2026. Data centers are getting caught up in conflict. What does this mean for cloud strategy?
Morning Brief: Sunday, June 14, 2026
Signal window: Saturday, June 13, 2026. NVIDIA Blackwell Leads on First Agentic AI Infrastructure Benchmark
Morning Brief: Saturday, June 13, 2026
Signal window: Friday, June 12, 2026. What could save Arizona tens of millions in annual customer and infrastructure costs? Residential pool pumps.
Morning Brief: Friday, June 12, 2026
Signal window: Thursday, June 11, 2026. 1M+ customers have connected solar to PG&E’s grid
Morning Brief: Thursday, June 11, 2026
Signal window: Wednesday, June 10, 2026. NVIDIA Confidential Computing to Help Expand Apple’s Private Cloud Compute