US urges G20 countries to allow AI training on creators’ work

The U.S. urged G20 countries to adopt “fair use” frameworks for AI training on copyrighted material, seeking protections for creators but without adding restrictive requirements. Multiple major AI firms (including OpenAI, Google, Meta and Anthropic) face mounting U.S. lawsuits from authors and publishers, while the Justice Department backed OpenAI’s position, arguing AI training generally qualifies as fair use. Nvidia CEO Jensen Huang warned that broad, theoretically framed regulation could constrain profits by slowing new model releases or forcing security-driven product changes.

Analysis

The immediate market read is not about a change in next quarter’s earnings; it is about a lower probability of a catastrophic legal reset for frontier AI. That matters most for the firms with the largest model-training budgets and the cleanest balance sheets, because they can treat copyright disputes as a manageable operating cost rather than an existential capex overhang. In that frame, GOOGL and META look better than the smaller, standalone model vendors: they can absorb licensing or settlement costs, while also monetizing distribution faster if policy stays permissive.

The second-order winner is NVDA, but via duration rather than headline sentiment. If policy risk fades, hyperscalers are less likely to slow training cycles or delay model releases, which supports GPU demand through the next 2-4 quarters; the bigger issue is whether this also accelerates supply expansion and weakens pricing power into 2026. The losers are content-rights holders like NYT, not because the litigation is gone, but because their leverage shifts from injunction threat to negotiated pricing, which is usually a worse bargaining position once precedent turns in favor of fair use.

The contrarian view is that the consensus may be overestimating how much this helps the AI complex near term. This is mostly a de-risking event, not a monetization catalyst; if courts still force data-licensing or revenue-sharing, the cost burden falls hardest on open-ended, cash-burning AI firms while the megacaps simply pass it through. The real falsifier is a court ruling or injunction that materially raises training costs or slows model release cadence over the next 1-3 months; absent that, the policy path points to lower multiple compression on GOOGL/META and less downside skew for NVDA.

Key Decisions for Investors

• Long GOOGL / META vs. broad tech for 1-3 months: policy de-risking should reduce the AI legal discount on platform names faster than it improves the economics of content owners; target ~5-8% relative outperformance, stop if litigation headlines shift back toward injunction risk.
• Tactical long NVDA on any post-news weakness for a 2-4 quarter horizon: the catalyst is reduced probability of training slowdowns, with the trade thesis invalidated if hyperscaler capex guidance rolls over or GPU lead times compress materially.
• Small short NYT or use NYT as a hedge against AI-law positivity for 1-6 months: downside is that fair-use-friendly policy weakens publisher leverage in settlement talks; cover if courts or regulators force a licensing framework with explicit per-use payments.
• Avoid chasing smaller AI application names until there is clarity on model-training liability: the best risk/reward is in balance-sheet winners that can self-fund legal uncertainty, not in companies whose valuation assumes frictionless data access.
• Watch for a 10-K/earnings season tell: if GOOGL/META reaffirm AI capex while legal reserves stay immaterial, add to longs; if management starts quantifying copyright settlement costs, reduce exposure quickly.