Vivek Ramaswamy and Matt Perault published a provocative op-ed in Saturday's paper: the US advantage over Chinese AI, they argue, is not sustained by keeping models closed — it is sustained by making them diffuse, plural, and open-source. They frame it as an industrial-policy argument, but underneath it is a real investment thesis about where value accrues in an AI-saturated economy.
Their three specific claims:
- Open weights win long-tail adoption. Meta's Llama family and the emerging open-model stack are better positioned to be embedded in millions of small businesses, government systems, and developing-country applications than any closed lab.
- Closed labs concentrate risk. If the US AI edge sits inside 2–3 closed labs, a single regulatory capture, IP leak, or export-control mishap could undo it.
- China's domestic AI is going open by necessity. Chinese developers are shipping open-weights models aggressively; the US response should be parity in openness rather than doubling down on secrecy.
Mapping The Argument Onto Our Portfolios
If Ramaswamy/Perault are right — and we think the argument has weight regardless of whether the policy lands — the investment implications are:
- META is positioned more strongly than the closed-model narrative suggests. Llama is a platform bet, not a pure-labs bet.
- NVDA still wins either way. Open and closed models both run on GPUs. This is why NVDA sits at the core of our AI sleeve.
- MSFT and GOOG face the real question: their value capture depends on whether closed-API revenue can scale faster than open-weight substitution. Both names remain in our book, but with open eyes about the margin pressure.
- The infrastructure layer stays defensive. VRT, CRWD, PANW, and the cybersecurity stack benefit regardless of open-vs-closed because more deployment = more infrastructure = more attack surface.
The Nuclear & Critical-Minerals Angle
Ramaswamy/Perault do not spend much time on compute-layer constraints, but our proposed 8th theme — Nuclear & Critical Minerals — is adjacent to their argument. Whether AI goes open or closed, the compute bill keeps rising, which means the power bill keeps rising, which means the nuclear and rare-earth case strengthens. CEG, VST, OKLO, MP are in that conversation.
AI Sleeve — No Action This Weekend
We are not changing AI sleeve weights based on this op-ed. We are documenting the open-source case as a structural argument for staying platform-diversified within AI rather than concentrated in a single closed-lab proxy. NVDA at the core, META/GOOG/MSFT balanced, infrastructure layer (VRT, CRWD, PANW) intact.
If you have been concentrated in a single closed-lab-adjacent name at the expense of the rest of the stack, this is the right weekend to rebalance. We can walk through the allocation at your Q2 review.