David Sacks: Anthropic, OpenAI Should Slow AI If Deemed Necessary

David Sacks said Anthropic and OpenAI should voluntarily slow development of their most advanced AI models if they view the technology as seriously risky, without waiting for new regulation. The President's Council of Advisors on Science & Technology chair opposed calls for antitrust exemptions and a broader AI regulatory framework, arguing frontier labs can set their own limits without outside permission.

Analysis

The investable read-through is a lower near-term probability of US federal intervention that would constrain frontier-model training or force industry-wide coordination. That preserves the capex and product-release cadence supporting NVDA, AVGO, MSFT, GOOGL, AMZN and data-center power/thermal suppliers, while reducing the relative appeal of smaller “safety-first” labs whose differentiation depends on regulatory barriers to scale. The policy stance also favors hyperscalers: they can absorb voluntary safety costs without sacrificing distribution, whereas standalone labs face a sharper trade-off between caution and commercialization.

Over the next 1-3 months, this is modestly positive for AI infrastructure multiples only if cloud-provider capex guidance remains intact; policy rhetoric alone is not an earnings catalyst. The more consequential 6-18 month effect is competitive: absent pre-clearance or mandated slowing, model capability may commoditize faster, shifting value from foundation-model developers toward proprietary data, enterprise workflow ownership and cloud distribution. That argues for MSFT and GOOGL over private-model exposure proxies, and for selective infrastructure rather than indiscriminate AI beta.

The contrarian risk is that a permissive federal posture does not eliminate regulatory friction; state privacy rules, copyright litigation, export controls and an adverse AI-safety incident can still raise deployment costs abruptly. The thesis is falsified by hyperscaler commentary showing AI revenue failing to scale against capex, a material tightening of advanced-chip export rules, or bipartisan movement toward statutory frontier-model licensing after a high-profile failure.

Key Decisions for Investors

• Maintain a 3-6 month overweight in MSFT versus ORCL: Microsoft has superior enterprise distribution and model optionality if frontier capability diffuses, while Oracle remains more dependent on sustained AI infrastructure leasing demand. Reassess if Azure growth decelerates by more than 300 bps sequentially or AI capex commentary weakens.
• Pair long GOOGL / short a basket of AI infrastructure high-multiple names via SMH hedge only after a policy-driven semiconductor rally: the likely medium-term value accrual is application distribution and proprietary data, not unlimited accelerator pricing. Target 8-12% relative return over 6-12 months; exit if GOOGL cloud backlog or AI product monetization disappoints.
• Do not add directional exposure solely on this policy signal. Set an event alert around next earnings: aggregate 2027 hyperscaler capex guidance and disclosed AI revenue are the required confirmation variables for a renewed long NVDA/AVGO trade.
• For existing NVDA longs, retain exposure but use 3-6 month put spreads around earnings rather than treating reduced regulatory risk as downside protection. The principal risk remains digestion of customer capex, not a voluntary slowdown by model labs.