Explained — Second Order by TKN

AI training vs inference — live GPUs are the stuck step: a blue-lit AI data hall with one short row of racks glowing amber

Explained

AI Training vs Inference: Same GPUs, Different Cheques

AI training builds the model; AI inference runs it for every request. Training is usually paid for with long reserved contracts, inference by the hour or the token. Both run on the same GPUs, and live, powered GPUs are the stuck step.

How long does it take to build an AI data center: a data hall under construction at night beside an unfinished substation lit amber, with the words 2 years to build, power is the wait

Explained

How Long Does It Take to Build an AI Data Center? The Building Is the Quick Part

The building takes about one to three years: Applied Digital's first 100 MW AI hall took two, from groundbreaking to live. Getting power to it often takes longer, with grid connection waits of five to ten years in many places. Power is the wait.

EUV vs DUV lithography — EUV boxes are the stuck step: a lithography cleanroom bench in night blue with one tool lit amber

Explained

EUV vs DUV: Two Kinds of Chip Lithography and Who Gets Paid

EUV and DUV are the two kinds of light used to print chips. EUV prints the finest layers in one pass and costs about €231 million a machine; DUV prints most layers for about €83 million. The stuck step is the EUV box, not the DUV one.

What is AI inference — PAID PER ANSWER, spare GPUs are the stuck step: a blue-lit GPU aisle with one amber rack held ready at the end

Explained

What Is AI Inference? How Neoclouds Get Paid for It

AI inference is running a trained AI model to answer a new request. It is the part of AI people use, so it is billed again and again, by the token or by the GPU hour. The stuck step is the spare GPU kept ready that earns nothing.

What is an AI data center — POWER TO THE HALL is the stuck step: a dark AI data hall with rows of racks and one amber-lit power feed at the end of the aisle

Explained

What Is an AI Data Center? Who Gets Paid, Who Waits

An AI data center is built for dense racks of AI chips, which need far more power, cooling and networking than a normal data center. Utilities, landlords and chip makers get paid at different steps. The stuck step is getting power to the hall.

Explained

What Are AI Tokens? The Unit Hyperscalers Bill AI By

AI tokens are the chunks of text, image or sound an AI model reads and writes. Clouds bill by the million. Google’s model APIs now handle about 22 billion a minute. The stuck step is capacity to serve them.

Who makes wafer fab equipment — FIVE TOOLMAKERS, paid before the first wafer: a dark semiconductor clean room with one tool bay lit amber

Explained

Who Makes Wafer Fab Equipment? The Five Toolmakers Behind AI Chips

Wafer fab equipment is the set of machines that turn blank silicon into chips. Five toolmakers sell most of it. They are paid before a fab makes its first wafer. The EUV scanner is the tightest tool.

GPU utilization — IDLE HOURS: a dark GPU supercomputer aisle where most racks glow teal and one run of racks glows amber, idle

Explained

What Is GPU Utilization? Why Idle GPU Hours Earn Nothing

GPU utilization is the share of time a GPU does useful work. A GPU earns only when it is busy, so idle hours are the stuck step. The contract decides who pays for them.

TPU vs GPU — TPU OR GPU: two rack aisles in a dark hall lead to one chip package whose HBM memory stacks glow amber

Explained

TPU vs GPU: Google’s Chip, Nvidia’s Chip, and Who Gets Paid

A TPU is a chip Google designs for AI matrix math only; a GPU is a general-purpose parallel chip. Different designers and builders get paid, but both chips wait for the same memory.

What is a TPU — GOOGLE’S OWN CHIP: one custom accelerator glowing amber in an open rack while crated hardware waits in the aisle

Explained

What Is a TPU? Google’s Tensor Processing Unit, Explained

A TPU, or tensor processing unit, is a chip Google designs for one job: the large matrix math that trains and runs AI models. Broadcom builds and ships it. Google earns only once the chip runs in a powered, finished hall.

Capex vs opex — BUY OR RENT: a dark GPU rack aisle with one amber pipe at its end; the cash still leaves first

Explained

Capex vs Opex: Buying GPUs vs Renting GPU Hours

Capex vs opex is the split between buying an asset that lasts years and paying a running cost as you go. Renting GPU hours turns a buyer’s capex into opex, but the capex does not vanish. The neocloud pays it first.

ONE MAKER — one open EUV scanner in a dark cleanroom, its gold mirror lit amber; the mirrors set the pace

Explained

Who Makes EUV Machines? ASML, Its Suppliers, and Its Rivals

Only one company makes EUV lithography machines: ASML. The roles that matter sit inside the machine: Zeiss makes the mirrors, Trumpf makes the laser, and the optics set how many boxes ship.