One of the sharpest reversals in AI-linked technology stocks in decades has rattled investors. But according to Sahil Mahtani, Director, Investment Institute at Ninety One, the recent sell-off says more about market positioning than the long-term outlook for artificial intelligence, which will depend on whether Chinese open-weight models can overtake US closed-weight models.
Equity factor constructions are notoriously nuanced, and different constructions can produce different numbers. Recently, however, the data all pointed in the same direction: a spectacular momentum unwind, particularly in AI-linked tech stocks.
Morgan Stanley’s long-short index of technology momentum recorded a 40 per cent fall in less than a month, the worst stretch in its 27-year history1, while Goldman’s high-beta momentum pair was roughly a third below its local high.2 Yet the broader market barely noticed. The S&P 500 slipped less than 1 per cent over the same period.
Sahil Mahtani, Director, Investment Institute: “We’ve witnessed one of the sharpest momentum reversals on record, on the reversal of many retail trades earlier this year. This has happened even as the broader index has been quiescent.”
Part of this is being driven by the unwind of leveraged structures, particularly among retail investors in East Asia. The proliferation of single-stock ETFs offering as much as five times daily leverage, with more than ten such vehicles trading SK Hynix alone, left the market vulnerable to this kind of reversal. Assets in those leveraged ETFs have shrunk substantially in a matter of weeks.
Retail investors have been at the centre of the unwind, having fuelled some of the strongest momentum trades of the past year. As leverage has unwound, the sell-off has accelerated.
Corrections of this magnitude rarely end until leverage has been cleared from the system. Retail investors were a key driver of this rally, with some of the biggest days of retail momentum occurring in early July.3 Yet many of the key retail favourites are substantially down. SpaceX is now trading below its IPO price, while rare earth and quantum computing names have round tripped. Even gold and silver failed to rally despite incredibly dovish inflation data.
Against this backdrop comes a new phase of the AI trade, with open-source models gaining ground against the frontier models. On Arena, one of the industry’s leading independent AI benchmarks, Chinese models have overtaken the best American ones for the first time.
Moonshot’s Kimi-K3 is now the leading open-weight model, allowing companies to run and adapt it on their own infrastructure. Closed-weight models such as ChatGPT and Claude can only be accessed through their developers. That distinction matters because it changes where value is likely to accrue across the AI ecosystem.
Mahtani: “The emergence of open-weight models transforms where value is likely to occur in the AI ecosystem, shuffling winners and losers.”
At the World AI Conference in Shanghai, Xi Jinping said China intends to make advanced models inexpensive and widely available. That raises questions about whether today’s pace of AI investment can be sustained over the medium term.
If Anthropic and OpenAI cannot build commercially viable businesses, future AI investment will inevitably slow. History suggests that when Beijing backs a strategic industry, from solar and shipbuilding to electric vehicles, competition intensifies and margins come under pressure.
Even so, this looks more like a rotation within the AI story than its end. Open-weight models still require enormous computing power. Moonshot gated new signups because demand had exceeded available computing capacity. Cheaper AI increases demand for processors, power and data centres, while shifting value creation across the broader AI ecosystem. A more competitive model layer could ultimately prove healthier for the broader AI ecosystem. Rather than concentrating value in a handful of model developers, it should create opportunities across semiconductors, infrastructure, power and the wider technology supply chain.
The first risk is economic growth failing to meet today’s elevated expectations. In BofA’s July survey, 54 per cent of managers expect no landing at all for the global economy over the next twelve months and just 2 per cent expect a hard one; three years ago, the no-landing camp was in single digits. Real GDP growth in the US for 2026 and 2027 is expected to be 2.3% and 2.2% respectively against a world economy growing 3% and 3.4%.4 Markets are priced for resilience. Even a small downside surprise to growth would affect markets.
The second candidate is a hawkish Fed. While Warsh and Waller spent the early part of the month signalling a more hawkish stance, recent inflation data has offered little support for those concerns. Both CPI and PPI came in meaningfully below expectations.
Moreover, while the reintroduction of tariffs, border enforcement and USMCA renegotiations could add to inflation over the coming months, the Trump administration has shown a willingness to adjust policy when economic conditions require it. That shifts the focus to the war. The biggest geopolitical risk is a renewed escalation in the Middle East. The US faces any further conflict with a diminished stockpile of key missile defence systems. Around half of its Patriot missile inventory is gone, along with more than half of its THAAD interceptors, and neither can be replaced on a timeline relevant to this conflict.5
For investors, the key risk is that the conflict spills over into energy markets. An oil shock through Hormuz that would hand the Fed a reason to act and could force it to tighten into a slowdown.
Mahtani: “The AI trade isn’t over, but it is evolving. Investors now need to understand where value is being created across the AI ecosystem and be open to the fact that some parts of the ecosystem are going to become increasingly commodified.”
1. MSZZTMTM index
2. GSPRHIMO Index
3. Citadel piece
4. IMF WEO July update
5. https://thehill.com/policy/defense/5842118-patriot-thaad-prsm-expenditure-iran/