Wirewalk Instituteof AI
inf-260 Intermediate Waitlist

On-Premises vs Cloud AI: Cost and Capacity Planning

For infrastructure, finance and engineering leads who must decide where AI compute should live and leave able to build a defensible total-cost model and a capacity plan.

The curriculum is being prepared and reviewed. Enrollment is not open yet.

About this course

Most on-premises versus cloud comparisons are wrong in the same few ways: they compare an owned server at 100% utilization with an on-demand cloud price, leave out power, cooling, networking and staff, assume a five-year life for hardware whose useful life for frontier work may be shorter, or ignore that cloud capacity at the scale needed may simply not be available on the date needed. This course teaches a total-cost and capacity method that survives a finance review.

You start by forecasting demand in GPU-hours from real workloads: training runs, fine-tuning, inference traffic and experimentation. You then build an owned-capacity model covering capital cost, depreciation, colocation or facility cost, power at a stated PUE, network and storage, support contracts, staff and refresh, and a cloud model covering on-demand, reserved and committed pricing, capacity blocks, storage, egress and the cost of idle reservations. The break-even module shows how sensitive the answer is to utilization, power price and hardware life. The final modules turn the decision into a staged capacity plan with procurement lead times, power and space constraints, hybrid bursting and review triggers, using hypothetical worked examples throughout. All numbers in examples are illustrative; learners bring their own quotes.

What you will be able to do

  • Forecast AI compute demand in GPU-hours and GPU memory from a portfolio of training, fine-tuning and inference workloads
  • Build a total-cost-of-ownership model for owned GPU capacity including facilities, power, network, storage, staff and refresh
  • Build an equivalent cloud cost model covering reservations, commitments, storage, egress and idle capacity
  • Calculate the utilization break-even between owned and rented capacity and test its sensitivity to key assumptions
  • Identify non-cost factors such as lead time, power availability, data residency and skills that can override the cost answer
  • Produce a staged hybrid capacity plan with procurement timelines and explicit review triggers

Syllabus

6 modules · 19 lessons · about 10 hours

Module 1: Forecasting Demand

Turning workloads and roadmaps into a defensible GPU-hour and memory forecast.
3 lessons
  1. From Workloads to GPU-Hours: Training, Fine-Tuning, Inference and Experimentation
  2. Peak, Baseline and Burst: Shaping a Demand Curve
  3. Measuring Today's Utilization Before Buying More

Module 2: The Real Cost of Owning GPUs

Every line item in an owned-capacity model and the ones most often forgotten.
4 lessons
  1. Capital Cost, Depreciation and Useful Life of AI Hardware
  2. Facilities: Colocation, Power at a Stated PUE and Cooling Upgrades
  3. Network, Storage, Support Contracts and Staff
  4. Refresh, Resale and the Cost of Falling Behind a GPU Generation

Module 3: The Real Cost of Renting GPUs

Modeling cloud and GPU-cloud cost the way the invoice will actually arrive.
3 lessons
  1. On-Demand, Reserved, Committed and Capacity-Block Pricing Models
  2. Storage, Egress and the Hidden Line Items on a GPU Cloud Bill
  3. Idle Reservations and Other Ways Cloud Cost Goes Wrong

Module 4: Finding the Break-Even

Comparing the two models fairly and stress-testing the conclusion.
3 lessons
  1. Normalizing the Unit: Cost per Useful GPU-Hour and per Million Tokens
  2. Utilization Break-Even and Sensitivity Analysis
  3. Worked Example: A Hypothetical 64-GPU Research Platform

Module 5: Beyond Cost: Constraints That Decide

Lead times, power, data rules and skills that can override the spreadsheet.
3 lessons
  1. Lead Times and Availability: Hardware, Power Connections and Cloud Quotas
  2. Data Residency, Sovereignty and Regulated Data
  3. Skills and Operating Model: Can You Run It?

Module 6: The Capacity Plan

Turning the decision into a phased plan finance and engineering both sign.
3 lessons
  1. Hybrid Patterns: Owned Baseline with Cloud Burst
  2. Writing a Staged Capacity Plan with Review Triggers
  3. Presenting the Decision to Finance and Leadership

Exam and certificate

The exam unlocks once you have completed every lesson. It draws 20 questions from a larger bank, spread across the modules, with the answer options shuffled for each attempt. You have 40 minutes and need 80% to pass. After a failed attempt you can try again after 24 hours, up to 3 attempts in total. Feedback is given per module, never as the answer key.

Passing earns a certificate of completion and assessment from the Wirewalk Institute of AI with a unique ID that anyone can check at institute.wirewalk.com/verify. It is not an accredited qualification or a vendor certification. How certification works.

The Wirewalk Institute of AI is independent. This course teaches publicly documented products and is not affiliated with or endorsed by NVIDIA, Amazon Web Services, Microsoft Azure, or Google Cloud. Product and company names are trademarks of their respective owners.

More in AI Infrastructure and HPC

All 14 courses in this track
inf-101 Basic Waitlist

AI Infrastructure Explained

A plain-language tour of the stack that runs AI: what a GPU server is, why clusters need special networks and storage, how jobs are scheduled, what an AI data center needs in power and cooling, and how cloud and on-premises options differ. No prior infrastructure background required.

6 h · 15 lessons Join the waitlist
inf-201 Intermediate Available

GPU Infrastructure for AI

The operator's view of data-center GPUs: what the current NVIDIA and AMD systems are, how they are packaged and connected, how the driver, CUDA/ROCm and container stack fit together, how to monitor and triage failing GPUs, how to share and size GPU memory, and what power, cooling and sourcing decisions follow.

12 h · 16 lessons See price at checkout
inf-210 Intermediate Waitlist

Linux Administration for AI Clusters

The Linux skills that differ on an AI cluster: choosing and imaging the OS, packaging GPU and RDMA drivers that survive kernel updates, NUMA and PCIe topology tuning, cgroup v2 isolation, shared identity and file systems, scientific software environments with Lmod, Spack and containers, and patching a fleet without breaking jobs.

14 h · 19 lessons Join the waitlist