Google is working on a new AI chip designed to make Gemini more efficient

The market is reacting to a credibility shift: Google is signaling that AI capex can be translated into lower inference cost rather than just higher spend. That matters for GOOGL because the biggest valuation debate is no longer model quality but whether Gemini can be monetized without cloud margins getting structurally diluted. The key second-order effect is that every step-function improvement in tokens-per-watt increases the economic moat of a vertically integrated stack, which should support cloud pricing power and advertising efficiency over time.

The near-term read-through to NVDA is mostly narrative, not fundamental. Google will still need external accelerators for training and peak demand well before any 2028 internal chip reaches scale, so this does not change unit demand in the next 4-8 quarters. The real risk for NVDA is multiple compression if hyperscalers keep proving that inference can be commoditized with custom ASICs; that is a 12-24 month sentiment risk, not a 2028 revenue cliff.

Contrarian view: the market may be overestimating how much of this becomes incremental margin. Custom silicon usually shifts cost from opex to fixed R&D and creates execution risk in software, packaging, and deployment. If Alphabet’s earnings show capex still rising toward the top end of guidance without a visible cloud-margin inflection, the stock can give back most of this move. What would falsify the bullish thesis is another quarter of elevated capex with no improvement in AI monetization metrics or cloud operating leverage.