Intel Foundry installed the first commercial High-NA EUV system this quarter, ASML Twinscan EXE 5200B, with 14A customer sampling scheduled for the second half. TSMC's senior vice president stated the company will not adopt High-NA before 2029. Samsung took two units. Three hundred eighty million per scanner, eight nanometer single-exposure resolution, the only path past the multi-pattern stacking the current node drowns in.

The architecture of the last fifteen years was clean. TSMC moved first on every EUV cliff, Apple and Nvidia rented the leading edge, Intel paid the gradualist tax, and the consensus marked the foundry order as fixed for the rest of the decade. The company that defined the curve told its customers it will sit out the next jump because the per-tool capex does not pencil. The same fiscal year, a struggling primary competitor took delivery, completed acceptance testing, and started shipping a 14A PDK no one else has.

The argument that ran through every TSMC earnings call, that incumbent advantage is cumulative because the process gap compounds, lands inverted. Compounding only works while the leader keeps spending into the cliff. The 14A risk production date sits in 2027, the customer PDK is already in hand, and the question of who manufactures the angstrom era reopens for the first time since Apple left Samsung in 2014.