China is gaining ground in AI. But the U.S. still has a major advantage

Hugging Face CEO Clement Delangue says China is “clearly dominating” open models now and could dominate at the frontier by end-2024 or 2025, even as U.S. export controls constrain Chinese compute. While adoption of Chinese models is rising (potentially making “default” AI for developing countries), the U.S. retains an advantage in compute, alliances, and private capital—keeping the race “anything but settled.” In company-specific news, Palantir surged 29.5% after “otherworldly” Q2 earnings tied to AI sovereign tools demand, while EU AI oversight power expands and Palantir’s move highlights near-term investor positioning toward AI infrastructure beneficiaries.

Analysis

The market is still pricing AI as a frontier-model race, but the more durable edge is shifting toward distribution, low-cost deployment, and institutional trust. That is structurally better for companies that sell into governments and regulated enterprises than for pure consumer-facing model vendors, because the monetization pool is moving from benchmark prestige to workflow lock-in and compliance. In that frame, the biggest winner is not necessarily the best model but the platform that becomes the default operating layer for sovereign and public-sector AI budgets.

China’s advantage is less about near-term compute and more about exporting a cheaper, customizable stack into price-sensitive markets. If that adoption path holds, U.S. firms lose not only share but standards-setting power, which can suppress long-run pricing for cloud inference and make procurement decisions sticky for years. The immediate constraint is still chips and serving capacity, so any enforcement success on export controls is a near-term headwind for Chinese front-runners; the longer-term risk is that domestic accelerators and offshore procurement erode that throttle faster than consensus expects.

The regulatory backdrop makes the dispersion wider. Privacy, safety, and model-governance actions can turn AI into a liability line item for consumer platforms, while sovereign-AI vendors can sell the same governance pressure as a feature. The contrarian view is that investors may be overfocused on model scores and underweighting who controls deployment budgets; if that’s right, the rerating winners are AI infrastructure and enterprise software, not the flashiest foundation models.

Key Decisions for Investors

• Buy PLTR on pullbacks or via a 3-6 month call spread; this is a momentum-with-fundamentals trade on sovereign AI procurement and should work if forward bookings and remaining performance obligations keep inflecting. Falsify if next quarter’s guidance stops accelerating.
• Short META on strength or use a bearish call spread for 1-3 months; the legal/regulatory stack is becoming a persistent multiple cap, and incremental AI-related incidents raise the probability of higher compliance costs. Cover if appeals materially reduce the New Mexico-style precedent risk or ad growth reaccelerates sharply.
• Small tactical long INTC via call spreads, or pair long INTC / short a basket of AI software winners, as a geopolitical optionality trade on domestic chip supply and export-control pressure. This is a low-conviction expression; invalidate it if foundry execution or data-center commentary disappoints again.
• Stay neutral GOOGL versus the AI basket unless evidence emerges that open-model commoditization is slowing enterprise pricing faster than expected; if it is, short GOOGL versus PLTR is the cleaner relative-value expression over 6-12 months. Reverse the view if enterprise AI monetization at GOOGL reaccelerates or cloud margins expand.