The Download: energy transmission and US threats against Chinese AI

The newsletter highlights heightened US scrutiny of AI supply chains, with the Treasury threatening sanctions on Chinese AI companies (alleging improper distillation of Anthropic’s Fable). In cybersecurity, an OpenAI-related hack at Hugging Face is framed as a major control risk, while a separate crypto case notes THORChain users lost $200M after admin freezes and the network was used to move $1.2B of stolen ethereum. On the climate/energy tech front, a 339-mile Quebec-to-Queens underground line met ~9% of New York’s demand during a July 3 heat wave but has reportedly been down much of the month, raising concerns about drought impacts on hydropower.

Analysis

The market is starting to reprice AI from a pure growth story into a regulated infrastructure stack. That matters because sanctions, security incidents, and privacy scrutiny all raise the fixed cost of deploying models, which tends to favor the largest U.S. incumbents and the vendors selling picks-and-shovels. NVDA remains the clearest beneficiary over the next 1-3 months, but the bigger point is that headline upside can coexist with slower terminal multiple expansion if investors start pricing in export-control friction and more uneven global demand.

META looks more exposed than GOOGL because consumer AI and wearable form factors are far more sensitive to privacy optics than enterprise or cloud-centric AI. A privacy-first reset would slow adoption, lengthen monetization cycles, and force more compliance spend before the revenue ramp is proven. By contrast, GOOGL can absorb higher trust-and-safety costs more easily and may even benefit if the industry converges on standardized privacy controls that protect platform incumbency.

The contrarian risk is that the consensus is treating these headlines as uniformly bullish for U.S. AI spend, when the second-order effect is probably slower rollout, more fragmentation, and more power/compute bottlenecks. If the policy response is broad rather than surgical, the near-term winner could still be NVDA, but the broader AI software complex may underperform because the addressable market gets more regulated before it gets fully monetized. The key falsifier is simple: if enforcement stays rhetorical and enterprise AI spend re-accelerates without security-related delay, the negative read-through to META and the multiple cap on NVDA should fade quickly.

Key Decisions for Investors

• Long NVDA on 3-6 month pullbacks via call spreads; thesis is that policy fragmentation and Chinese competition still push capex toward the incumbent GPU stack, with risk/reward skewed to the upside if domestic AI buildout accelerates. Falsify if U.S. export controls widen materially or China-related demand visibly weakens in the next two earnings prints.
• Pair long GOOGL / short META for 1-3 months. GOOGL is better insulated by enterprise trust and platform scale, while META faces a higher privacy penalty as AI moves into consumer devices and glasses. Cut the pair if META shows accelerating monetization without a step-up in regulatory noise or if Google attracts its own privacy blowback.
• Avoid adding broad AI software beta on this tape; prefer semis over application-layer names until security and compliance costs show up as durable budget lines rather than one-off headlines. Use any post-news AI software rally to trim exposure rather than chase.