Forget GDP. Meet GDI: The new economic scorecard for AI power

Morgan Stanley argues investors may increasingly assess national 'Gross Domestic Intelligence' resources as an investment overlay, highlighting AI compute capacity as a new measure of competitiveness. Epoch AI's Q4 2025 data show the US dominating global AI compute at roughly 75%, with China at about 10%, while Google stands out as the largest player and every top company in the dataset is American. The article is largely explanatory, but it underscores the strategic advantage of US AI infrastructure and compute concentration.

Analysis

The key implication is not that AI is broadly important; it is that AI advantage is becoming a capital-intensive balance-sheet moat, and the market is likely underpricing how concentrated that moat is among a handful of US incumbents. If compute is the new strategic input, then the winners are the firms that can self-fund the fastest capacity expansion without relying on external financing cycles, which structurally favors hyperscalers over most software and semiconductor-adjacent peers. That creates a second-order squeeze on smaller cloud vendors, regional data-center operators, and model builders that depend on rented capacity rather than owned infrastructure.

For NVDA, this is supportive but not as simple as "more compute equals more upside." The more important effect is pricing power durability: if the largest buyers are still racing to add capacity, near-term demand remains elastic, but over a 12-24 month horizon the risk shifts to mix and bargaining power as hyperscalers internalize more of the stack with custom silicon. That is why GOOGL is the cleaner expression of the thesis than NVDA — it owns both demand and supply for AI compute, so incremental AI intensity can compound into margin expansion rather than only revenue growth.

The contrarian miss is that the market may be extrapolating US dominance too linearly. A national compute lead does not automatically translate into equity outperformance if capex intensity rises faster than monetization, especially for firms subsidizing AI infrastructure to defend platform share. ORCL is an interesting secondary beneficiary because its relative scarcity of compute becomes more valuable in a world where every large enterprise wants optionality, but it also faces the risk that its AI narrative is benchmarked against hyperscaler scale and therefore vulnerable to any slowdown in cloud spend.

The main reversal catalyst is a capex pause, not a breakthrough at the model level. If power constraints, GPU supply normalization, or investor pushback on free-cash-flow conversion forces hyperscalers to slow spending over the next 2-3 quarters, the "compute race" trade will unwind quickly in the hardware and networking names first, then in the broad AI complex. Conversely, if enterprise AI workloads start generating visible revenue per watt, the winners will be the vertically integrated platforms with the cheapest marginal compute, not the names with the largest headline installed base.

Key Decisions for Investors

• Long GOOGL vs. NVDA on a 3-6 month horizon: express via long GOOGL / short NVDA pair to capture the higher-quality AI monetization with lower dependency on external capex cycles; target 10-15% relative outperformance if hyperscaler spend remains elevated.
• Add to NVDA on pullbacks only, using a staggered entry over 2-4 weeks: keep position sizing modest because upside is still tied to capex duration, but near-term demand remains strong; use a 12-month horizon with a 15-20% downside stop if hyperscaler commentary turns cautious.
• Buy ORCL calls or call spreads 6-9 months out: the stock benefits if enterprise buyers seek alternate compute access, but the payoff is asymmetric only if Oracle is re-rated as a strategic capacity provider; risk is that the trade fails if AI spend consolidates further into the hyperscalers.