A memory supplier sells a wafer’s worth of chips. A cloud operator sells GPU hours. An application has to earn back what it spends on those hours. These five articles follow the money and equipment through the buildout, using public filings, price quotes, benchmarks and models you can rerun.

Read the series

  1. How the AI buildout reaches your DDR5 bill What a memory supplier sees in its order book and what a PC builder sees at checkout.
  2. GPU economics from break-even to cost per token How much a GPU has to earn per hour, and how three MLPerf runs change the estimated cost per token.
  3. What limits an AI cluster from delivery to peak load Why getting the GPUs early can leave the billing date untouched, and what happens to inference queues when power drops.
  4. AI infrastructure cash flow and who finances the buildout What three hyperscalers spent on equipment and how suppliers help finance the customers buying it.
  5. Six AI businesses and five tests of the boom Where each business earns its money, what customers are already paying, and which numbers to revisit in 2027.

Calculators and data