Meta Makes Big Moves Around AI Models and Chip Production

Meta’s more important signal is bargaining power, not product novelty. Aggressive pricing on inference pushes the low-end AI layer toward commodity status while preserving Meta’s advantage where it actually monetizes: distribution and ad data. That is constructive for META and for suppliers with design-win economics like AVGO/TSM, but it is a slow-burn negative for the merchant-GPU narrative at NVDA/AMD over 6-18 months; near term, this is still capex-heavy and only works if ad ROI improves faster than infrastructure spend.

GE Vernova looks expensive only if you value it on current margins instead of the cash-flow stack it is building for 2027-2035. The market is already paying for backlog, but may still be underpricing the embedded service annuity that comes from a larger installed base; that’s the real margin expansion lever. The bear case is not “valuation”; it is a slowdown in AI/power demand or execution slippage that prevents the backlog from converting into higher-margin aftermarket revenue within the next 2-3 quarters.

American Tower’s issue is less satellite displacement than leverage combined with slower dividend flexibility. Direct-to-cell is a threat to marginal rural/international growth, not the dense urban tower network, so the first-order risk is multiple compression if rates stay sticky and growth decelerates, not an immediate earnings collapse. The contrarian view is that the market may be overestimating substitution speed; the thesis breaks only if satellite coverage improves indoors and at scale over 12-24 months, or if leverage fails to trend down.