Track 1
AI Compute 101
Start here. GPUs, GPU-hours, model costs, and why compute became a market.
0 of 7 complete
Learn AI compute in plain English.
Free lessons, calculators, and explainers for understanding GPUs, model costs, cloud capacity, data centers, power constraints, and emerging AI compute infrastructure. Explore 60+ short lessons, follow a learning path, or use a calculator to estimate real compute costs.
A different lesson is highlighted in the Morning Brief each day.
Sponsor slot available
Sponsor Compute College
Reserved placement for infrastructure, data-center, energy, and AI compute sponsors.
View sponsor optionsThe curriculum
Eight tracks, plain English
Pick a track and work through it in order, or jump to the topic you need. Track your progress in this browser — no account required. Every lesson is free.
Track 2
GPU Pricing & Capacity
What H100, H200, and B200 hours cost, and how on-demand, reserved, and spot pricing differ.
0 of 8 complete
Track 3
Model Costs
What it costs to train a model and to serve one, and why utilization drives the bill.
0 of 7 complete
Track 4
Model Benchmarks & AI Compute Economics
How benchmark scores connect to token pricing, latency, throughput, and real inference spend.
0 of 23 complete
Track 5
Power & Data Centers
Power, cooling, networking, memory, and the physical sites where AI compute actually runs.
0 of 17 complete
Track 6
Compute Market Structure
Neoclouds, capacity markets, and reservations: how compute supply gets priced.
0 of 8 complete
Track 7
Compute Futures
Forward pricing for compute: futures, forward curves, and forward contracts.
0 of 5 complete
Track 8
Buyers & Operators
How labs and teams buy, compare, and budget GPU capacity in practice.
0 of 8 complete
Compute College tracks
Explore the eight tracks
Once you know the basics, work through pricing, model costs, infrastructure and power, market structure, compute futures, and the buyer-and-operator playbook. Emerging topics and the calculators sit alongside them.
AI Compute 101
What is AI compute?
The basic resource behind training and running AI models.
Why compute matters
Why chips, power, and capacity are becoming economic constraints.
What is a GPU-hour?
The basic unit behind compute pricing.
H100 vs H200 vs B200
How accelerator generations affect performance, supply, and cost.
GPU Pricing & Capacity
H100 price per hour explained
What an H100 hourly quote includes and why offers vary.
H200 price per hour explained
When memory and workload fit can justify a premium.
B200 price per hour explained
How next-generation rates reveal scarcity and demand.
GPU vs TPU vs custom ASIC
How accelerator types trade flexibility for efficiency.
On-demand vs reserved vs spot GPU pricing
Compare access terms, savings, and interruption risk.
What are GPU rentals?
How buyers rent accelerator capacity and what rental signals can reveal.
What are spot prices?
How short-term, interruptible capacity can become a pricing signal.
What is GPU cloud capacity?
Why buyer-accessible GPU supply differs from headline inventory.
Model Costs
What is model training cost?
The cost of teaching a model before deployment.
What is frontier model serving cost?
How high-end AI usage becomes recurring compute demand.
What is cost per million tokens?
How hosted APIs price inference, and why output drives the bill.
What is GPU utilization?
Why paid capacity can cost more when it sits idle.
How fast does an H100 depreciate?
Why useful life and obsolescence drive the real cost per GPU-hour.
What is Model FLOPs Utilization (MFU)?
How much of a paid GPU does useful work, and why it sets cost per result.
How to estimate monthly AI compute burn
Turn training and serving usage into a recurring budget.
Model Benchmarks & AI Compute Economics
Why reasoning models cost more to serve
How "thinking" tokens turn one query into far more compute.
Benchmark score vs production cost
Why the highest score is not always the cheapest to run.
What is SWE-bench?
The coding benchmark that mirrors real agent debugging work.
What is Terminal-Bench?
Measuring how well agents operate in a real terminal.
Claude Opus 4.8 benchmark explained
Read a current model release through compute economics.
What is Claude Mythos Preview?
Why restricted frontier-model access can still move compute demand.
What is GPQA Diamond?
A graduate-level reasoning benchmark and what it signals.
What is MMLU-Pro?
A harder knowledge benchmark for comparing model quality.
What is Humanity’s Last Exam?
A frontier-difficulty test of model reasoning limits.
What is an agent benchmark?
How multi-step agent tasks drive repeated inference demand.
How model releases affect AI compute demand
Read launches and benchmark jumps as demand signals.
Power & Data Centers
Why power matters
Why electricity and site capacity shape AI compute markets.
What is a data center?
The physical site where chips, power, cooling, networking, and operations come together.
Why cooling matters
Why heat limits how densely AI chips can be deployed and operated.
Why networking matters
Why fast interconnects turn individual chips into useful AI clusters.
Why memory matters
Why high-bandwidth memory can constrain accelerator supply and model performance.
What is an AI cluster?
How many GPUs become one useful compute system.
What is NVLink?
Why fast GPU-to-GPU links improve connected systems.
What is InfiniBand?
How large AI clusters move data between servers.
What is NVL72? Scale-up vs scale-out
How rack-scale systems link many GPUs as one.
What is liquid cooling?
Why cooling becomes a constraint as AI racks grow denser.
What is data center interconnection?
How network access connects compute with data and buyers.
What is Power Usage Effectiveness?
How facility overhead power affects AI economics.
Power sourcing for AI: PPAs and behind-the-meter
How PPAs, on-site power, and nuclear/SMR deals gate AI capacity.
What is a megawatt of AI compute?
Translate power announcements into potential capacity.
What is a data center interconnection queue?
Why waiting for grid connection delays compute supply.
What is AI data center cooling density?
How concentrated rack heat limits deployments.
What is High-Bandwidth Memory (HBM)?
Why memory supply and fit matter for accelerator economics.
Compute Market Structure
What is a neocloud?
Compute-first cloud operators and why they matter.
What is a compute capacity market?
How compute access becomes a supply and pricing signal.
What is GPU cloud capacity?
Why buyer-accessible GPU supply differs from headline inventory.
What is a compute reservation?
Why buyers commit to GPU access ahead of need.
What are spot prices?
How short-term, interruptible capacity can become a pricing signal.
What is GPU-backed financing?
Borrowing against GPUs and rental contracts to fund AI buildouts.
What is sovereign AI compute?
AI capacity a nation controls within its own borders.
Compute Futures
What are compute futures?
Forward-looking pricing for compute capacity.
How to read a forward curve
How curve shape becomes a market signal.
What is a compute forward contract?
How future capacity could be priced before delivery.
What is a compute capacity market?
How compute access becomes a supply and pricing signal.
What is a compute reservation?
Why buyers commit to GPU access ahead of need.
Buyers & Operators
How AI labs buy compute
How major buyers secure flexible and committed capacity.
How to compare GPU cloud quotes
Compare rate, reliability, network, and completed-workload cost.
When to use spot GPUs vs reserved capacity
Match access terms to interruption risk and deadlines.
What is a neocloud SLA?
Understand the reliability terms behind GPU access.
What is AI compute procurement?
Build a buyer scorecard for capacity and risk.
Tools & Calculators
GPU-Hour Cost Calculator
Estimate AI compute cost from GPU price, runtime, utilization, and overhead.
AI Training Cost Calculator
Estimate a training-run budget using GPU-hours and operating assumptions.
Model Serving Cost Calculator
Estimate recurring inference cost from usage and capacity needs.
Reserved vs On-Demand Calculator
Compare committing to reserved GPU capacity against paying on-demand for a workload.
API vs Self-Hosted Calculator
Compare paying per token for a hosted API against running the model on your own GPUs.
Emerging Topics
What is Terafab?
A proposed chip and compute-capacity project.
What is Stargate?
A mega-scale AI infrastructure buildout and what it says about future compute supply.
What is Colossus?
xAI’s large-scale compute buildout and why power can become the bottleneck after GPUs arrive.
What is Project Rainier?
AWS’s custom-silicon AI cluster and why proprietary chips matter to compute supply.
What are Prometheus and Hyperion?
Meta’s multi-gigawatt AI campuses and what industrial-scale compute really means.
Put it to work
Turn the lesson into a number
Use the GPU-Hour Cost Calculator, AI Training Cost Calculator, or Model Serving Cost Calculator to estimate real compute costs from your own inputs.
Keep up with the market
Follow the market after the lesson
Get the ComputeTape Morning Brief for daily AI compute pricing, power, capacity, and infrastructure signals — plus a different Compute College lesson highlighted each day.
Contact Compute College
Help improve Compute College
Send lesson ideas, corrections, source material, or questions about AI compute education to the editorial team.