AI Capex Arms Race Reframes Valuation Discipline From Meta's Open-Source Gambit to GE Vernova's Priced-In Perfection

The AI investment cycle has entered its discipline phase. Research published across the wire frames Meta's open-source model strategy less as altruism than as a bid for architectural control — commoditising the model layer to protect distribution and advertising economics while denying rivals a proprietary moat. That is a durable competitive logic, but it also implies sustained capital intensity with deferred payback, a profile equity markets are increasingly reluctant to fund without visibility. The same tension is acute in the electrification complex. Analysts argue GE Vernova now carries expectations so elevated that even flawless delivery on grid and gas turbine backlogs offers limited incremental upside, with the risk skewed toward multiple compression on any order-timing slippage. At the opposite end of the market-cap spectrum, Chardan initiated QumulusAI with a buy rating, underscoring continued appetite for early-stage compute and data-centre exposure even as large-cap AI beneficiaries look fully valued. The practical implication for portfolios is a barbell: pay for scarce, contracted infrastructure cash flows, but avoid paying twice for the same narrative. Where the AI premium has migrated from earnings into sentiment, the asymmetry has quietly inverted.