A useful starting point for researching AI infrastructure is a map of the system: what does the work, what feeds it, and what connects it. This guide sets out the questions to ask before turning a technology story into a company thesis.

Start with the job

Define the workload before drawing the supply chain. Training a model and serving a model are different tasks. Ask which constraints matter in the particular deployment you are studying: computation, memory capacity, data movement, power, or a combination. A claim about “AI demand” is only a starting point.

Trace the layers

Work outward from the processor to memory, packaging, networking, power, and cooling. At each layer, record the component, its role, and the source that establishes that role. Keep a technology diagram separate from a verified commercial relationship.

Separate participation from economics

Being relevant to a technology does not establish revenue, pricing power, or investment value. Look for disclosed customers, qualification milestones, production status, capacity, and financial reporting. Where those details are missing, preserve the question instead of filling the gap with a confident conclusion.

Build a document trail

Keep the original link, publication date, and exact scope of each announcement. Revisit the map as disclosures change. The most useful research map makes it possible for another reader to trace every important connection back to its source.

Questions to take with you

  • Which part of the system is the actual constraint?
  • Is a supplier relationship explicitly disclosed?
  • What evidence would change the thesis?

Further reading