The AI Capex Cycle Broadens: Nuclear, Space and Cloud Partnerships Absorb Capital

The AI trade is no longer a single-factor bet on accelerator demand. Analysis of the global layers of the investment cycle frames the build-out as a stack: silicon at the top, but with increasingly binding constraints in electricity generation, data-center siting, launch and connectivity infrastructure. That framing explains the capital rotating toward next-generation nuclear developers such as Oklo — whose shares recovered dramatically from a disastrous IPO debut — and toward Rocket Lab, where the Neutron vehicle is being reassessed as a genuine revenue driver rather than a development-stage option. At the application layer, Elon Musk's public concession that he underestimated Anthropic's models is being read by investors as validating Amazon's strategic stake and cloud partnership, a differentiator in the hyperscaler contest. Yet dispersion is the operative risk. Oracle delivered its best single session since 1992 during the past twelve months and still finished the period with a negative total return — a reminder that narrative velocity in AI infrastructure does not reliably convert into shareholder returns. Positioning discipline matters more than thematic conviction: the constraint layers offer the cleanest exposure to secular capex, but valuations in speculative nuclear and space names already discount execution that has not yet occurred.