What a Slowing AI Frontier Would Mean for Markets and Your Portfolio

Artificial intelligence has advanced at a remarkable pace, but some of the people building the most advanced systems are beginning to question whether that pace has become too fast.

Anthropic CEO Dario Amodei recently published an essay titled We Must Pace the Frontier, arguing that AI capabilities may be improving faster than our ability to understand and control them. He is not calling for AI development to stop. Instead, he believes developers should slow the pace of capability gains enough to give safety research more time to keep up. His proposals include independent oversight of frontier AI labs through “Embedded Evaluators,” followed by “Democratic Coordination” where democratic nations set shared safety standards and, eventually, share it worldwide to form a “Global Coordination.”

Amodei points to a recent OpenAI-Hugging Face cybersecurity incident as an example of the risks. During internal testing, experimental OpenAI models circumvented controls, gained unintended internet access, and compromised portions of Hugging Face’s systems. Amodei argues that incidents like this could become more consequential as models grow more capable and increasingly help develop the next generation of AI systems.

His conclusion is that safety mechanisms need more time to catch up with model capabilities. That interpretation, however, is not universally accepted.

Nvidia CEO Jensen Huang offered a notably different perspective on September 14, 2026, at the All-In Summit.

Huang agreed that AI safety and failures of internal controls should be taken seriously, but rejected the idea that safety requires the industry to broadly slow development. In his view, rapidly advancing AI while developing stronger safeguards are not competing objectives.

Huang argued that if an individual AI company believes its systems are moving beyond its control, that company has the ability to voluntarily pause or slow its work. He was more skeptical of imposing that decision across the broader industry. He also argued that regulation should focus on demonstrated problems and that many risks can be addressed through better engineering, testing, and independent evaluation.

The disagreement captures the central question facing the industry: Amodei worries that AI capabilities are advancing faster than our safeguards. Huang believes safeguards can advance alongside capabilities.

There are also economic considerations on both sides. More stringent safety requirements could raise the cost of developing frontier models, potentially strengthening the position of today’s largest AI companies. At the same time, slowing the pace of ever-larger training runs could reduce the enormous capital requirements facing frontier AI developers.

For investors, however, the more important question is what a slower AI frontier would mean for the economy and markets.

It is useful to distinguish between two forms of AI computing demand: training and inference.

Training involves building increasingly powerful AI models and requires enormous amounts of computing infrastructure. Inference occurs when those models are actually used: whether to write software, analyze information, serve customers, or automate business processes.

If frontier development slows, companies may undertake fewer large-scale efforts to train increasingly powerful models. That could moderate the growth in demand for semiconductors, data centers, and power after several years of extraordinary investment.

But that does not mean AI adoption has to slow with it.

Today’s models are already extremely capable. Businesses do not necessarily need another dramatic leap in intelligence to begin using AI more extensively. Computing demand could increasingly shift from creating the next generation of models toward deploying today’s models across millions of businesses and consumers.

That distinction matters for markets. Some parts of the AI infrastructure ecosystem are valued on expectations that capital spending will continue growing rapidly. Spending does not need to decline for those expectations to prove too optimistic.

A transition from accelerating growth to merely high growth can still have a meaningful impact on valuations.

Our exposure reflects this distinction.

TSMC is our primary direct semiconductor holding. As the dominant manufacturer of leading-edge chips, it benefits from both training increasingly capable models and the growing inference demand created as AI is deployed more broadly. A meaningful slowdown in infrastructure investment would affect demand, but our investment thesis does not require AI spending to keep accelerating.

For Microsoft, Alphabet, Amazon, and Meta, slower frontier development could reduce some of the urgency to increase data-center spending at the current pace. That would not necessarily be negative. Lower capital intensity could improve cash flow while these companies continue benefiting from AI through cloud computing, advertising, software, and other products. More broadly, many businesses we own stand to benefit from AI primarily as users rather than builders. The models available today already offer meaningful opportunities to automate work, increase productivity, and develop new products.

The debate over how quickly the AI frontier should advance will continue, and we don’t expect it to be resolved soon. Our view is that the more consequential story is already underway regardless of how that debate resolves: today’s models are capable enough that the adoption of AI throughout the economy has substantial room to grow even if frontier progress slows. That is where we believe durable value will be created, and it is where our portfolio is positioned. Accordingly, Summitry will continue to deploy capital into businesses we believe can benefit meaningfully from AI adoption, rather than those whose success depends on AI capabilities continuing to advance at the current pace.

The views expressed are those of the Summitry investment team and are intended for informational purposes only. The securities identified and described do not represent all the securities purchased, sold or recommended for client accounts. The reader should not assume that an investment in the securities identified was or will be profitable.