An HBM capacity forecast that unlocked a two-year fab-utilisation commitment.

1. 01 The decision

   A hyperscaler had approached the client with a two-year, take-or-pay HBM3E capacity commitment. Signing would lock ~40% of the fab's output but insulate the client from a demand air-pocket if AI capex softened. The board needed independent evidence on where HBM demand was actually going.

2. 02 How we ran it

   A three-analyst team modelled HBM demand bottom-up from GPU shipment forecasts across the top 8 accelerator programs, cross-checked against foundry capacity announcements. We ran primary conversations with three hyperscaler procurement leads (under NDA) and two competing memory manufacturers. Weekly interim reads with the client's corporate development team.

3. 03 What we shipped

   A 92-page report, a live Excel demand-supply model with three demand scenarios, a supplier-competitor teardown, and two board-level presentation decks (one for the buy-side committee, one for the hyperscaler negotiation).

4. 04 What changed

   The client signed the commitment at a valuation ~9% higher than the hyperscaler's opening offer, backed by our demand model. Fab utilisation is now locked through 2027. The corporate development team retained us for a follow-on study on adjacent packaging capacity.

Client identity is anonymised in this write-up. On-record versions are available on request where the client has given explicit consent — reach the desk for named references.