SpaceX, Alphabet, and SK Hynix Are Quietly Flashing a Bullish Signal Investors Should Not Ignore

The key mechanism is that AI capex is becoming self-reinforcing: each incremental dollar of infrastructure spend validates the next budget cycle, which shifts the debate from a near-term demand peak to a multi-year capacity race. That argues for continued multiple support in the semiconductor complex, but not evenly—investors are likely underweight the second-order beneficiaries where pricing power is still climbing, especially HBM/memory and networking, while NVDA is already the consensus expression of the theme.

Near term, the risk is not demand collapse but execution bottlenecks: wafer allocation, advanced packaging, power availability, and model monetization. If those constraints persist, the market will reward the infrastructure suppliers with the cleanest supply-side leverage more than the platform names funding the buildout; GOOG/GOOGL can keep spending, but custom silicon introduces a subtle long-run ceiling on NVDA share of wallet even as absolute GPU demand stays strong.

Over 1-3 months, the biggest catalyst is capex guidance commentary from hyperscalers and memory vendors, which should keep the AI basket bid unless one of them signals a rephasing. Over 6-18 months, the contrarian risk is that the market starts capitalizing AI spend like a utility buildout—lower ROIC, longer payback, and heavier depreciation—which would compress valuations for the most exposed buyers while preserving earnings power for the toll collectors. The thesis breaks if a major hyperscaler cuts 2026 capex or if GPU lead times collapse faster than revenue can absorb supply.