Overview
- High‑bandwidth memory (HBM) shortages and price spikes are forcing system changes that reduce planned memory per rack and raise AI infrastructure costs, a constraint analysts say now affects Nvidia as well.
- Nvidia has introduced a revenue‑sharing, credit‑support financing model to let smaller cloud partners build capacity without full upfront payments, with Sharon AI and Firmus as early participants.
- Bank of America and other analysts continue to view Nvidia as the long‑run AI infrastructure leader able to capture roughly 65%–70% of hyperscaler capex even as its stock has pulled back from May highs.
- Meta plans to start manufacturing its in‑house 'Iris' AI chip this September, a project Reuters reported passed initial tests and is intended to supplement rather than replace GPU purchases.
- Memory winners such as Micron report strong results and long‑dated supply commitments that will likely keep tightness through 2027 while capacity additions in 2026–27 could eventually ease prices and alter rack economics.