Martin's BlogAI is no longer just hyping a single sector; it’s spilling over layer by layer along the hardware supply chain.
The first wave was the most direct: hyping GPUs and AI chips. Large model training and inference first require computing power—whoever provides the core computing power benefits first—so Nvidia exploded first, then driving AI servers, HBM, advanced packaging, optical modules, liquid cooling, and other directions.
The second wave started hyping storage. The bigger the model, the more data; parameters, caches, vector databases, and training datasets all need stronger read/write and storage capabilities. AI doesn’t just need to compute fast; it needs to read fast, store reliably, and retrieve efficiently. The GPU is like the engine; storage is more like the memory system.
The third wave is now turning to CPUs. No matter how powerful the GPU, someone has to schedule tasks, manage systems, coordinate networks, and handle general computing loads. The CPU is the scheduling hub for the entire system. So the AI narrative is starting to spread from single chips to full servers, motherboards, power management, high-speed interconnects, cooling, and cabinet systems.
Next, it’s likely to continue in a few areas: interconnects, cooling, networking, power, and finally capital returns.
How chips connect at high speed, how data flows, how heat is dissipated, whether the grid can handle it, and whether the computing power bought back achieves high enough utilization—these will all become new bottlenecks.
So this AI bull market, the further out you look, the more it resembles a massive industrial infrastructure rebuild. On the surface, it’s a model revolution; underneath, it’s all chips, storage, servers, data centers, power, cooling, and capital expenditures.