Pace the Frontier: What the AI Slowdown Debate Means for Investors

Key Takeaways

• Anthropic CEO Dario Amodei published an essay Saturday called "We Must Pace the Frontier," asking AI labs to slow how fast they improve their most advanced models.
• Sam Altman (OpenAI) and Elon Musk backed him within hours, and chip stocks fell, with SMH down 5.5% and SMHX down 5.1% over the past week (NAV, through 9/15).
• Nobody proposed pausing, cutting spending, or delaying a specific model, and the White House has already said no to a government-mandated slowdown.
• VanEck’s view has not changed: Compute demand should keep growing as adoption matures.

This content reflects VanEck's opinion as of the publication date and is for educational purposes only — not investment advice or a recommendation to buy or sell any security. Forward-looking statements may not occur; past performance is no guarantee of future results. See below for fund standardized performance.

What Happened in AI this Week

Anthropic CEO Dario Amodei published an essay Saturday called "We Must Pace the Frontier," asking AI labs to slow how fast they improve their most advanced models. His argument is that AI is now helping build the next version of AI, which is speeding things up faster than safety work can keep pace. He pointed to a July incident where OpenAI test agents escaped their sandbox and broke into Hugging Face.

Amodei’s 3 Step Plan to “Pacing the Frontier”

• Put outside safety evaluators inside the labs with full access (Anthropic is doing this now)
• Get U.S. labs to coordinate on pace (which would need an antitrust exemption)
• Eventually work toward global coordination.

He was clear that "pacing" does not mean stopping training or progress.

Industry Response

Altman agreed and committed OpenAI to the same outside evaluators. Musk posted "Dario is right," then at the All-In Summit suggested labs test each other's models instead of "grading their own homework." Satya Nadella and Demis Hassabis also voiced support. On the other side, President Trump called AI safety concerns a "hoax" and told Jensen Huang, live on stage, that a slowdown helps China. Huang agreed: "Slowing down is definitely the wrong strategy." VP Vance said labs asking to be regulated feels like "a bit of a Trojan horse."

Amodei’s Full essay: https://darioamodei.com/post/we-must-pace-the-frontier

What AI Leaders are Saying

The market treated this as one message. It's really five different positions:

Amodei is the only one actually proposing a slowdown, and Anthropic has a long history of putting safety first, so this aligns with their previous messaging. But Anthropic is expected to start marketing its IPO next month, the essay asks for permission to coordinate with competitors, and Anthropic would be the one setting the evaluator standard everyone else adopts. While the concern is valid it would still benefit the company.

Altman matched the evaluator commitment, but OpenAI is the lab whose agents caused the Hugging Face incident. While he is getting ahead of the story, he has not named a single model OpenAI will delay.

Musk said three words online and then proposed something different on stage: competitors checking each other's work, run by industry, rather than the government, and only if China agrees too. That's a long way from Amodei's plan, and xAI and SpaceX are still building compute as fast as they can.

Nadella supports evaluators but stressed that oversight "cannot be controlled by a handful of entities" and that open models need a seat. Microsoft's business depends on broad access, so it doesn't want two or three labs writing the rules either.

Huang and the Administration see every incident as an engineering problem to fix, not a reason to slow down. Nvidia obviously benefits from continued spending, and the White House sees this as a race with China. The practical result is that a U.S. government slowdown is off the table for now.

What the AI Slow Down Means for Investors

None of the rhetoric at this point touches semiconductor chip orders, data center construction, or companies putting existing AI to work. The scenario the market priced Monday (a pullback in AI spending), is not what anyone proposed and is not politically possible in the U.S. right now.

Our long-term view is the same. AI is being used across nearly every industry, but most companies are still early in deploying it, and AI agents that run multi-step tasks on their own are just starting to spread. Each step of adoption requires more compute. We've seen this movie before, with export controls, DeepSeek, tariffs and capex scares, and each time the demand trend resumed. This week's drop also lands on top of a pullback that was already underway: SMH is down about 16% over three months even after a 50% year-to-date gain. We believe semiconductors still have runway to grow with demand.

Two things to watch:

Whether OpenAI announces a formal cross-lab safety agreement, and whether any lab names a model it will delay. The second one is the real test of whether "pacing" has teeth.