CIOs unpack whether model development will actually slow, and what that could mean for the ongoing AI buildout
The prospect that AI model development could slow down rattled markets on Monday. A steady buildup of negative headlines and popular tension around AI safety concerns reached a head when Anthropic co-founder and CEO Dario Amodei released an essay calling for a slowdown in AI model development over the weekend. He was joined, in a rare show of unity, by OpenAI CEO Sam Altman as well as Elon Musk in his calls for model slowdown and governance.
While both the US President and the Chinese Foreign Ministry made statements rejecting the idea of slower or more regulated AI growth, markets still reacted strongly, with drops on major US equity markets. Semiconductor and other hardware stocks were the biggest losers as investors moved to price in the prospect of slower AI model development as a headwind to the wider AI infrastructure buildout that has been a major engine of market growth in the past year.
“While this issue just came out over the weekend calling for a slowdown and increased levels of governance around the development of foundational AI models, we've actually gone through a period of a couple of months here where the narrative has shifted a little bit around AI development,” says James Learmonth, co-CIO and Portfolio Manager at Harvest ETFs. “There’s been a political pushback by constituents against data center development that’s fed into negative sentiment, you also have rising treasury yields feeding expectations of higher costs for this investment… this is just another leg in the story that’s amplified those existing jitters.”
“We’re talking about a quick market reaction to short term news around sentiment. We’re not talking about any change in the fundamentals. We’re not talking about anybody changing announcing plans to change what they’re spending, what they’re building, what they’re doing,” adds Elliot Johnson, CIO at Evolve ETFs. “But you can tell the market’s on edge.”
Will a model slowdown actually happen?
Both Johnson and Learmonth expressed some scepticism as to whether the unity expressed by these AI leaders will actually result in slower model development. Perhaps the strongest argument against such a slowdown occurring has been articulated by President Trump and the Chinese Foreign Ministry. With both the US and Chinese governments stating opposition to AI slowdowns and the US explicitly framing AI model development as a “race” against China to develop artificial superintelligence, then competitive incentives to keep developing will remain in place.
Learmonth notes that Anthropic has talked about unilaterally implementing third-party evaluators, whether governments create a regulatory framework for AI or not, but that the company faces competitive pressure from OpenAI, XAI, and other providers which may only get more intense as it plans an IPO.
While AI safety is ostensibly the reason they’re calling for slower development, Johnson notes that there may be some additional incentives for companies to slow model development. They include a need to sort out the sometimes challenging unit economics in AI. He notes that the most cynical take would be to view the call for regulation as a way for these companies to build a competitive moat around their established positions.
While we can’t know whether this model slowdown will occur, both Johnson and Learmonth argue that it is not likely to change the nature of the AI buildout in the short to medium-term.
Why a slowdown in model development might not mean a slowdown in AI cap-ex
Johnson posits a thought experiment. If AI models stop progressing and stay exactly where they are, will we see AI usage increase, decrease, or stay the same in five years’ time? He firmly believes that even if AI models stay unchanged, there will be more users. For the hardware and semiconductor stocks currently being rattled by this discussion of a slowdown, that means their GPUs, DRAM chips, and other necessary widgets will continue to be ordered.
Learmonth says that the pullback in hardware stocks does not represent a reversal of investor consensus around this issue. Rather, he sees this as investors re-timing the growth prospects for AI data centers. The expansion of AI services will continue, he says, but investors are beginning to think it will slow down somewhat. A slight slowdown will also potentially resolve some of the supply bottlenecks giving certain hardware companies so much pricing power right now. Learmonth adds, however, that scepticism around the sustainability of the AI buildout was already being considered in the prices of some of these stocks. He believes that even if the pace of AI growth slows or regulation comes into effect, that the economic incentives to build out more AI infrastructure remain.
That regulation could also prove incredibly challenging to write and implement. Johnson says that AI governance is not an obvious area that experts can tackle. Many of the examples of AI behaving in unsafe ways come from unexpected and unplanned for behaviours. Learning how to regulate that and ensuring that AI agents actually follow and hold to those regulations could prove remarkably challenging.
Managing big questions in portfolios
While leaders like Amodei and Altman may be talking at a very high level about questions of safety, agency, governance, and geopolitics, investors have to deal with the immediate ripple effects in their portfolios. Both Johnson and Learmonth say that in this environment, diversification continues to prove itself to be valuable. Learmonth emphasizes the need for a diverse exposure to the tech sector as a whole, including both hardware stocks and the software names that got some much-needed reprieve in the wake of this slowdown discourse.
Hard as it can be for any observer to parse between apocalyptic predictions about AI safety, an immediate pullback in certain stocks, and the long-term fundamentals for the space, both Learmonth and Johnson argue that a focus on those fundamentals can be effective.
“You come back to first principles,” Johnson says. “I’ve not heard of anybody saying that they’re going to curtail CapEx spending. I’ve not heard of any people saying that they’re expecting people to stop consuming these products and services. This rally in the markets that we’ve seen in the past 18 months has been characterized by an expansion of margins rather than an expansion of multiples, meaning that it’s been driven by quarter after quarter of strong earnings from these large tech companies… which is very healthy.”
