Inside Atreides' AI infrastructure thesis: power shortages, wafer bottlenecks, orbital compute, GPU lifespan, the buildout-vs.-bubble debate, and potential conflicts of interest.
Jensen has conquered the demand side, but he can't move the two supply constraints — TSMC's wafers and the grid. Does the market eventually stop pricing these names off a 3–4 year GPU depreciation convention and start underwriting useful life instead?
Jensen has conquered the demand side, but he can't move the two supply constraints — TSMC's wafers and the grid. Does the market eventually stop pricing these names off a 3–4 year GPU depreciation convention and start underwriting useful life instead?
Useful-life repricing has started. Meta and Google extended assumptions while the supply bottleneck holds.