Livewire Special Report

The Memory Supercycle: who really wins the HBM build-out

DRAM economics, the high-bandwidth-memory oligopoly, and the ASX-adjacent ways to own the squeeze. A research-grade map of the trade of the decade.

By Livewire Research · 12 June 2026 · 25 min read · Data appendix included
Stacked memory dies on a fabrication line

Photo illustration: Livewire · Source: Getty

Every supercycle begins as a shortage nobody believes. In 1999 it was fibre. In 2007 it was iron ore. In 2026 it is high-bandwidth memory — the stacked DRAM that feeds every accelerator in every data centre now being poured, raised and energised across the world's grids. The AI build-out gets the headlines; the memory squeeze pays the dividends.

The arithmetic is brutal and simple. An AI accelerator carries six to eight HBM stacks. Each stack consumes roughly three times the wafer area of the commodity DRAM it displaced — and yields worse. Every point of share the AI complex takes from conventional memory removes supply the rest of the economy still needs. Three companies control effectively all of it.

That oligopoly — SK Hynix, Samsung and Micron — has spent two decades destroying its own pricing power with ill-timed capacity races. The discipline this cycle is different, and the reason is structural: HBM is sold on multi-year contracts, qualified part-by-part into each accelerator generation. Switching costs have arrived in a commodity industry. The result is the most interesting margin story in global hardware — and almost nobody on the ASX is positioned for it.

This is a Livewire Special Report

The remaining 22 minutes cover what the headlines can't:

  • — The HBM cost curves: who makes money at every point of the cycle, modelled
  • — Six listed ways to own the squeeze — including the two ASX-adjacent plays
  • — The data appendix: contract pricing, capex and wafer-start tables, sourced

Special reports are for members. Everything else on Livewire — every wire, video and scan — stays free.

1 · The shortage nobody believes

Memory has been a value trap so reliably and for so long that the muscle memory of an entire generation of analysts says sell the first up-quarter. That reflex is the opportunity. The qualification moat around HBM4 means 2027 supply is already spoken for — sold out before the fabs that will make it are finished.

2 · Oligopoly economics

Three suppliers, two real process roadmaps, one customer that matters. We model the contract structures and show why the marginal wafer now earns triple what it did in the last cycle peak — and why the discipline holds even if AI capex growth halves.

3 · Six ways to own the squeeze

From the obvious (Micron) to the overlooked (the test-equipment chain, the substrate makers, and the two ASX-adjacent exposures most portfolios already touch without knowing it). Position sizes, entry logic and the exit signals, laid out.

— The full data appendix follows in the member edition. This demo report ends here.