The moat fact first: when AMD acquired Xilinx, it bought a portfolio, and the parts of that portfolio that matter are the ones solving hard 3D-integration problems. US11270977B2, granted March 2022 to Xilinx, patents a power delivery network for active-on-active stacked integrated circuits (CPC H01L 25/0657).
Gloss it once. "Active-on-active" stacking means putting two functional dies directly on top of each other, not a die on a passive interposer. Powering the upper die cleanly through the lower one is genuinely hard — power has to traverse the stack without excessive loss or noise. A patent that addresses that delivery problem is foundational to making 3D stacks work.
Why a moat read cares: AMD's competitive case rests partly on system-level integration — stacking, packaging, and the surrounding fabric — not just core compute. Xilinx-era patents on stacked power delivery are exactly the kind of IP that strengthens that case, and they continue to issue under the acquired name as the acquisition thesis compounds.
The period framing matters. This grant issued shortly after the acquisition closed, capturing the technical rationale: AMD wanted adaptive-compute and advanced-integration IP, and stacked power delivery is a concrete instance of what that meant.
The caveat we attach: a patent is a defensive asset, not a benchmark or revenue line. This grant shows AMD-via-Xilinx holds IP at a hard integration problem; it does not establish that products ship using it or quantify any advantage.
For the period investor, the durable point is that big acquisitions are partly purchases of IP position. Reading the grants that keep issuing under the acquired name is one way to see whether the acquisition thesis was real — and stacked power delivery suggests it was.