The Orchard · Module 01 · Beginner

AI Infrastructure 101

Four short lessons on the hardware, power, and supply-chain constraints that actually decide which AI companies win — no finance degree required.

Lesson 1 of 4

The Compute Balance (Nvidia vs. AMD)

Why AI servers need both monster accelerators and heavy-duty host CPUs.


The Tech Foundation

While graphics chips (GPUs) process trillions of complex AI math matrix loops simultaneously, they cannot direct themselves. They require a powerful host CPU (Central Processing Unit) to fetch data from storage, feed it into the GPU memory, and direct networking traffic between server racks. If the CPU lags, the GPUs sit idle. This is why AMD's server market share has skyrocketed with their high-core EPYC processors, which run the backbone of Microsoft Azure and AWS AI infrastructure clusters alongside Nvidia hardware.

Investor Angle

Nvidia's moat isn't just their chips; it's CUDA software, which maps code natively to their hardware. AMD is directly attacking this via their open-source ROCm software ecosystem and their MI300 series accelerators, which feature huge on-board memory capacity specifically to process ultra-large AI models cheaper.

Mental Analogy

"Think of an AI server box as a massive, high-speed construction site. The GPUs are hundreds of laborers laying bricks at lightning speed, but the CPU is the foreman orchestrating the blueprints and logistics. Without the foreman CPU, the laborers can't work."

Progress: 1/4

More From The Orchard