Explore the five-layer AI infrastructure value chain—from chips and manufacturing through power, networking, and software—to understand how industry news affects different companies across the ecosystem. Each layer has distinct competitors, economics, and dependencies, so a single headline's impact depends on where a company sits in the stack.
When you read a headline like "Microsoft signs multi-billion-dollar deal for AI data center capacity" or "ASML reports strong EUV lithography bookings," you are reading about two very different parts of the same industrial ecosystem. This course gives you a map of that ecosystem — the AI infrastructure value chain — so you can quickly place any piece of news into its correct location and understand who benefits, who is threatened, and why.
AI infrastructure is not a single product. It is a stack of physical and software layers, each built by different companies with different business models, cost structures, and competitive dynamics. Understanding this stack matters because a single AI-related headline can be bullish for one company and irrelevant — or even bearish — for another company that sits one layer away in the same chain.
Consider a concrete example: a report that a hyperscaler (a large cloud provider like Microsoft, Google, or Amazon) is designing more of its own custom AI chips. To a generalist, this might read as "bad for AI chip companies." But with a layered view, you know to ask: bad for which chip company? Nvidia, which sells general-purpose AI chips, faces a genuine competitive threat. But Broadcom, which often designs those very custom chips as a partner to the hyperscaler, and TSMC, which manufactures them regardless of who designed them, may see this as good news. The same headline moves through the value chain differently depending on where a company sits.
The course organizes AI infrastructure into five interconnected layers:
Figure: The five layers of the AI infrastructure value chain, from foundational manufacturing at the bottom to user-facing software at the top. News about any one layer requires asking which companies in the other layers are affected, and how.
These layers are not independent — they are tightly interlocking. A chip design (layer 1) cannot exist without a foundry and equipment to manufacture it (layer 2). A finished chip is useless without power and cooling to run it (layer 3) and networking to connect it to thousands of other chips (layer 4). And none of it is usable by an actual business without software (layer 5) to program and orchestrate it. This is why a shift in one layer — say, a move toward liquid cooling because chips run hotter, or a shift in networking standards because clusters are growing larger — tends to ripple through the others. A change in chip power consumption directly affects cooling equipment demand; a change in cluster size directly affects networking architecture; a shift in chip design philosophy (GPU versus ASIC) directly affects which foundry packaging techniques are needed.
This course will not teach you how a transistor works, how to write code that runs on a GPU, or how to build a discounted cash flow model for these companies. It is deliberately conceptual: enough working vocabulary and structural understanding that when a company's earnings call mentions "CoWoS capacity," "HBM bit growth," or "scale-up versus scale-out networking," you understand what business dynamic is actually being discussed — and which companies in the value chain are the ones to watch as a result.
Each subsequent concept in this course drills into one node of this map in more depth: what the technology is, why it exists, which companies compete there, and what tends to drive demand or margin for that specific business. As you go, keep coming back to this five-layer map — it is the scaffolding that ties every individual concept back to the broader competitive and financial picture.