For managers, analysts and IT generalists who need to understand what sits underneath an AI service and leave able to follow, question and contribute to infrastructure conversations.
The curriculum is being prepared and reviewed. Enrollment is not open yet.
About this course
When an AI project stalls, the cause is often something nobody on the business side can see: a GPU quota that was never approved, a model too large for the memory of the server it was given, a storage system that cannot keep up, or a rack that cannot be powered. This course gives non-specialists a working mental model of AI infrastructure so they can follow those conversations and ask useful questions.
You follow a single request from a chat window down through the serving layer, the GPU server, the cluster network, the storage system and the data center building, and then back up through how training jobs are scheduled and how capacity is bought in the cloud or on premises. The course closes with a checklist of questions to ask your infrastructure team. Product names are given as of October 2026 as examples; the concepts are built to outlast them. Deeper treatment of every topic is available in inf-201 and the rest of the infrastructure track.
What you will be able to do
Describe the layers of an AI system from user request to data center, and name what each layer is responsible for
Explain in plain terms why AI work runs on GPUs and why GPU memory size so often decides which model fits where
Distinguish training from inference in terms of the hardware, network and storage each demands
Explain why AI clusters use dedicated high-speed networks and parallel storage instead of ordinary office IT equipment
Identify the role of schedulers such as Slurm and Kubernetes in sharing GPUs among users and teams
Compare cloud, neocloud, colocation and on-premises options for GPU capacity on cost, speed and control
Prepare informed questions for an infrastructure team about capacity, utilization, power and risk
Syllabus
5 modules · 15 lessons · about 6 hours
1
Module 1: Following a Prompt Down the Stack
A request's journey from chat window to GPU and back, and the vocabulary each layer uses.3 lessons
What Happens Between Pressing Enter and Seeing an Answer
Training Versus Inference: Two Very Different Infrastructure Jobs
Tokens, Parameters and Memory: The Three Numbers That Size Everything
2
Module 2: The GPU Server
What is physically inside an AI server and why its memory and connections matter more than its clock speed.3 lessons
Why GPUs and Not CPUs: Parallel Work in Plain Terms
Inside an 8-GPU Server and a Rack-Scale System
Will This Model Fit? Reading a GPU Memory Budget
3
Module 3: Clusters: Networks and Storage
Why many GPUs need their own fabric and file systems, and what goes wrong when they are undersized.3 lessons
Why GPUs Need Their Own Network: InfiniBand and Ethernet Fabrics
Storage That Keeps GPUs Busy: Parallel File Systems, Object Stores and Checkpoints
Who Gets the GPUs: Slurm, Kubernetes and Queues
4
Module 4: The Building Behind the Cluster
Power, cooling and space: the physical limits that now decide where and how fast AI capacity can grow.3 lessons
Kilowatts per Rack: Why AI Breaks Traditional Data Centers
Air, Direct-to-Chip Liquid and Immersion Cooling Explained
Grid Connections, Lead Times and Location Choices
5
Module 5: Getting Capacity: Cloud, Neocloud or Your Own
The main ways organizations obtain GPU capacity and the trade-offs a non-specialist should understand.3 lessons
Hyperscalers, GPU Clouds and Managed APIs: What You Are Actually Buying
Owning GPUs: Colocation, On-Premises and the Utilization Question
Questions to Ask Your Infrastructure Team
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 30 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, AMD, Amazon Web Services, Microsoft Azure, Google Cloud, Kubernetes, or Slurm. Product and company names are trademarks of their respective owners.
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.
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.
How to make Kubernetes a good home for GPU work: the GPU and Network Operators, device plugins versus Dynamic Resource Allocation, GPU sharing, RDMA with Multus, batch queueing and gang scheduling with Kueue, JobSet and KAI Scheduler, distributed training and multi-node inference patterns, and the observability and cost controls that keep a shared GPU platform fair.