One tactic I've used is decoupling access from volume by proposing a node-specific option structure rather than a straight take-or-pay LTA. Instead of committing to a large fixed wafer count, we locked in a smaller base volume with clearly priced upside tranches that only triggered if our internal milestones were met. The key was making those tranches operationally useful to the foundry, not just financially optional. What actually moved the foundry wasn't our top-line demand forecast. It was a single artifact: a time-phased tape-out and yield-ramp forecast tied to real product gates. We showed, quarter by quarter, when designs would hit MPW, when they would enter risk production, and when they would plausibly reach volume — with historical data from prior nodes showing how closely we'd hit those ramps. That credibility mattered more than raw TAM slides. We also translated that forecast into fab-relevant terms: expected wafer starts by month, sensitivity bands if yield lagged, and what we would release if things slipped. From their perspective, this reduced scheduling risk. They could see we weren't reserving capacity "just in case," and that unused wafers would roll back into the pool early enough to reallocate. The early result was incremental new-node capacity without locking us into worst-case volume. Internally, it preserved flexibility if market demand softened. Externally, it signaled we understood the foundry's constraint system, not just our own roadmap. That shift — from asking for capacity to showing how we'd responsibly consume it — is what unlocked the allocation.