AI's buildout is now a financing and supply-chain problem, not a chip problem

Threads we are following

• Whether the next GPU-backed special purpose vehicle prices tighter or wider than the SpaceX deal, and whether its callable structure becomes the template.
• Whether Apple can secure enough LPDDR5X to meet initial iPhone 18 and foldable iPhone Ultra demand without extended wait times or retail stockouts.
• Whether Samsung Foundry will qualify its own 4nm or 2nm base die for NVHBM if TSMC's 3nm allocation hits a ceiling, and whether Nvidia would accept a second base-die source.
• Whether customer prepayments from Nvidia, Tesla and Big Tech will actually accelerate substrate capacity before 2028, or merely lock in allocation.
• Whether Samsung will disclose SRAM bitcell density figures for its GAA-based nodes, enabling direct comparison with TSMC's A14 and Intel's 18A.