What could end the tech rally? Jefferies flags investment risks as China’s cheap AI models threaten US players

The report also flagged cybersecurity risks, citing a Five Eyes Alliance warning that “advances in AI could dramatically accelerate cyberattacks in the near future with organisations having only months to prepare.”

The biggest risk to the AI-driven tech rally is not a sudden jump in semiconductor supply but a "sudden realisation by investors that the hyperscalers and the likes of OpenAI and Anthropic will not be able to make a return on their investment," according to latest research report by Jefferies.

The brokerage argued that "for now at least, there remains zero sign of AI capex slowing," yet warns that funding concerns could "trigger a sudden unwillingness to fund these investments which will then be aggravated by the circular arrangements between the main players, such as Nvidia financing OpenAI to buy its chips."

Jefferies' view comes as Hong Kong-listed Z.ai, formerly Zhipu AI, launched GLM-5.2 on 13 June. "GREED & fear is no expert but GREED & fear hears from more informed sources that this new model is almost equal to Anthropic as a competitor for the corporate market and is just one quarter of the cost in terms of cost per token," the report said.

The launch coincides with what Jefferies calls a "reaction against tokenmaxxing" that is likely to "lead to a slowdown in Anthropic's uptil now remarkable revenue growth ahead of its planned IPO." Anthropic's annualised run-rate revenue "surged from $9 bn at the end of 2025 to $47 bn in May," Jefferies noted.

The report highlights a shift in usage toward Chinese models. "Top Chinese AI models processed 21.37tn tokens on the global aggregator platform OpenRouter in the week ended 21 June, up from 4.37tn in late April, compared with 5.76tn tokens for the top US models," Jefferies said, citing weekly usage of the top nine models on OpenRouter.

"GLM-5.2 proves enterprises no longer have to sacrifice intelligence for privacy. We are seeing a massive acceleration in companies pulling their AI workloads out of the public cloud and back onto local corporate servers," the brokerage quoted AI feedback it received.

Despite token price pressure, Jefferies remains constructive on memory suppliers due to the Jevons Paradox. "Falling token prices should lead to rising DRAM prices," the report stated, adding that "the story that the DRAM industry has changed structurally... looks to GREED & fear an increasingly powerful argument."

Hynix, Samsung Electronics and Micron are now trading at 7.8x, 6.8x and 9.2x consensus 12-month forward earnings, respectively.

Reflecting this view, Jefferies is increasing tech hardware exposure across portfolios. "Hynix and Kioxia will be included in the global long-only portfolio with an initial 4 per cent weighting each... while the existing investment in Samsung Electronics will be increased by one percentage point," the report said.

On Taiwan, Jefferies noted "boom-like conditions" with real GDP up 14.55 per cent y-o-y in 1Q26, "the fastest quarterly growth rate in nearly 48 years." TSMC capex is forecast to rise to $56 bn in 2026 and $65-70 bn in 2027, with AI expected to account for 31 per cent of TSMC's revenues this year.

The report also flagged cybersecurity risks, citing a Five Eyes Alliance warning that "advances in AI could dramatically accelerate cyberattacks in the near future with organisations having only months to prepare."