�鶹��

Podcast: How frontier AI regulations could impact the tech trade, on and (5 min)
Podcast: Across the Pond | Europe's overlooked innovators, on and (17 mins)
Video:Investors Club | Your questions answered on gold, the Fed, and semis (9 mins)

Anthropic’s latest frontier AI models remained suspended globally as of Tuesday, following a new White House policy directive restricting access for foreign-born people, which led the company to pull both its “Fable 5” and “Mythos 5” services globally. Fable is the safeguarded public-facing version that was more widely released, while Mythos had a limited release to select partners for cybersecurity purposes. US Commerce Secretary Howard Lutnick in a letter on Monday said he had invoked the Commerce Department’s export controls over concerns the leading-edge models could be used by military intelligence users in countries of concern.

Talks between Secretary Lutnick, other US officials, and Anthropic technical staff are expected to continue on Tuesday in search of a solution to address the security concerns and restore access. Separately, more than 80 cybersecurity executives and tech leaders later signed an open letter urging the administration to reverse the suspension on cybersecurity grounds. Ahead of the model’s launch last week, Anthropic had detailed stronger model safeguards against both criminal misuse and “distillation” by competing foreign labs. Distillation refers to techniques that allow rival labs to replicate much of a frontier model’s performance by learning from its outputs, rather than training and rebuilding those capabilities from scratch.

This still unfolding episode has raised the possibility of AI regulation moving beyond chips and tools, and toward more direct restrictions on advanced models. For investors, this matters because AI, and especially AI-related semiconductors, have been a central driver of equity gains this year, with semiconductor stocks accounting for more than half of the S&P 500’s year-to-date advance.

While our conviction is low that tighter restrictions on US frontier models will be sustained, the episode has put a spotlight on how AI regulatory decisions could affect competition, market leadership, and capital spending. Below, we detail several potential implications:

Controls could reduce financial incentives at the leading edge. A lasting set of restrictions along these lines could reduce incentives for frontier labs to invest in training ever more capable models, in our view, especially if new capabilities risk triggering regulatory curbs. In a bear-case scenario, this could weigh on AI compute demand, since about half of this goes to training new models, while the rest comes from inference to run existing models for end-users. It could also result in a slower training cycle and less rapid innovation, which could weaken investor confidence in semiconductor demand and pressure near-term token pricing (the cost of processing text, images, or other data through an AI model). This is not our base case, however, and we think any hit is likely be marginal for the sector, unless restrictions broaden materially or last longer than markets currently expect.

The impact on global talent could become costly. As currently worded, the restriction may be more stringent than intended, with for example US-based British- and Canadian-born researchers reportedly locked out of frontier model work. Sustained nationality-based limits on model access or contribution could make some ex-US labs more attractive to top technical talent. This matters because the US still benefits from a strong concentration of talent, capital, infrastructure, and leading AI companies. A more restrictive regime could erode that edge by making access, mobility, and long-term career paths less predictable.

Model leadership could tilt to competitors. In a more stringent outcome, restrictions on frontier-model access could give an economic advantage to companies in countries that make advanced models available without similar constraints. Chinese-developed models already account for roughly 61% of token usage among the top 10 models on OpenRouter, which unified service developers can use to access AI models from multiple providers. That’s up from less than 1.2% in late 2024, showing how quickly demand can shift when access is easier and costs are lower. The risk of sustained restrictions could also encourage global users, enterprises, and governments to employ domestic or ex-US models. It is not yet clear if the latest techniques to block model distillation will prove effective. At present, this remains a key risk for frontier labs, whose model leadership can prove temporary if offshore competitors can simply replicate their capabilities and bring them to market independently.

So, without taking any views on individual companies or models, we believe this is a policy risk to monitor with potential implications that could extend well beyond one disrupted model release. We note that unilateral AI restrictions may be difficult to sustain if they weaken relative national competitiveness or slow domestic innovation—outcomes we think would be hard for both the US private sector and policymakers to accept. Still, if progress at the frontier were to slow meaningfully, there may be knock-on effects across tech sub-sectors, with traditional software facing less immediate displacement pressure from rapidly improving AI tools and parts of the semiconductor market exposed to slower expected growth in AI computing demand.

Stepping back, we remain constructive on the broader AI investment thesis, while maintaining balanced exposure across the enabling, intelligence, and platform layers. As the industry moves into the agentic era, we think the next phase of spending is likely to be distributed more widely across semiconductors, memory, optics, power-related components, and semiconductor capital equipment, all of which can offer more demand visibility, pricing power, and earnings support. More broadly, we continue to advocate exposure across AI-related equities and across our transformational innovation themes, which also include Power and resources and Longevity.