An investment banking expert has warned the CalPERS’ board of the risks inherent in AI, emphasising the importance of investors understanding how their exposure to AI is at risk because of Chinese competitors.
Investors should see the AI boom in the context of the US/China tech race because it distinguishes today’s digital revolution from previous leaps forward in industrialisation, according to Anikent Shah, managing director, sustainability and transition strategies at investment banking and capital markets firm Jefferies.
In a sweeping conversation on how AI will shape the reallocation of capital, Shah, who was speaking alongside CalPERS chief investment officer Stephen Gilmore at a CalPERS board education day in January, flagged key trends and risks in the emerging technology that long-term allocators like CalPERS must navigate for success.
He argued that AI adoption is part of a bipartisan industrial strategy in the US – yet flagged potential bumps in its adoption like more state-level regulation as both Republican and Democratic lawmakers respond to voters’ concerns around labour dislocation and high power prices.
He also urged the CalPERS board to be mindful that China is edging ahead in the AI race. US allocators should know what China’s equivalent AI-focused champions are doing in their home market, and assess “holding by holding” if they will forge ahead of their US equivalent. [See Where foreign capital fits in China’s parallel tech system.]
Industrial policy
Investors should see the AI boom in the context of the US/China tech race, he said. This is important because it distinguishes today’s digital revolution from previous leaps forward in industrialisation. Both republican and democratic parties are using industrial policy to compete with China in a strategy that short-circuits free markets.
Like the CHIPS Act that boosts domestic semiconductor production, or the US government’s 10 per cent stake in Intel, for example. Elsewhere, the government has pledged to build nuclear projects to support the energy demands of AI, and a recent trade deal with Japan will also focus on AI.
Shah urged the CalPERS board to understand the significance of this state intervention.
“The US government is also another huge provider of capital to AI. The US government is using its balance sheet to participate,” he said.
However, although US industrial policy is focused on supporting AI infrastructure to compete with China, domestic US politics increasingly dances to a different tune. Shah highlighted research that shows the public concern with the negatives of AI like labour dislocation and power prices.
“Fifty per cent of Americans are more concerned than excited about increased AI use,” he said. “Republican lawmakers know if they are pro-AI they will pay for it at the ballot box.”
Public discontent is stopping – or postponing – the construction of multiple data centres, he said. Elsewhere, US states continue to enact AI laws despite federal pre-emption efforts to limit regulation around AI . California’s new Transparency Frontier Artificial Intelligence Act introduces clauses asking corporates to publish a risk framework on how they will manage “catastrophic harms” from AI and whistleblower protections, for example.
China is winning the race
In recent years, China has taken the lead in critical technologies in a trend that became apparent when China revealed DeepSeek a year ago. The AI start-up made headlines and caused equity markets to sink when it revealed its own model, developed at a fraction of the cost of rivals like OpenAI’s ChatGPT. (See DeepSeek triggered rout highlights diversification dilemma.)
Shah said that although the narrative around China winning the race is downplayed in the US “we have a DeepSeek moment every month.” He said the US has advantages like Nvidia, which has the most advanced chips, and America’s computing infrastructure is deeper than China, however China has no issues around building data centres or any of America’s power constraints.
Even though many public US pension funds have no exposure to China, he urged investors to understand how their exposure to AI is at risk because of Chinese competitors. US allocators need to understand what the same AI-focused companies are doing in China and assess if they will forge ahead of their US equivalent.
“Many of our clients are doing this holding by holding,” he said.
Not many businesses USe AI yet
Importantly, over 80 per cent of U.S. businesses report not using AI in the production of their goods and services yet. This matters, he explained, because the pace at which AI is rolled out in the economy and integrated into business models is inextricably linked to GDP growth and equity markets: 60 per cent of GDP growth in the US currently derives from AI capex.
“You need [AI] companies drawing revenues that are significant enough that they can finance the associated infrastructure,” Shah said.
Because US corporations are still at the early stages of integrating AI into their goods and services, it raises the question of where and when the revenues will come from to pay for the massive expansion in AI, and drive demand.
“Someone has to pay for it,” said Shah.
One reason for corporate reticence could be trust. He noted that although AI is increasingly matching human experts, people still trust humans more to carry out complex tasks. Areas where AI is gaining traction include coding, back-end professions, self-driving cars and robotics. He added that demographic challenges, particularly in China, will drive robotics.
Power prices
AI attracts hostility in the US because people are concerned about power prices. But Shah warned that the increase in power prices in the US so far has little to do with AI.
“Those increases are still to come,” he said.
He said power prices won’t come down until the US fixes its permitting system. The “single biggest” challenge facing the US power grid is a permitting system that brings layers of complexity to transmitting power across the country. Getting permission for new transmission lines can take decades, requiring different states and environmental approvals.
“Can the administration, in partnership with states and cities, have permitting reform that would have a net benefit to us all?” he asked.
Despite the current administration’s cheerleading of fossil fuel production, over 90 per cent of the incremental power in the US comes from renewables and storage, according to its own data. In a ringing endorsement of the wisdom of the decision by the CalPERS board to invest in renewables, Shah urged board members to continue not to “just follow headlines,” but look instead “at the real economy”, which shows that trillions of dollars continue to plough into solar and battery technology.
Shah warned that more risk is creeping into AI. For example, hyperscalers have begun raising debt finance.
These companies have used their profits to finance AI infrastructure but they are now going to the debt markets to tap private credit and the bond market in a notable shift. These companies have little leverage and do have the capacity to take on debt, but he warned that adding leverage to the system increases uncertainty and leaves these companies in the hands of forces outside their control.
For allocators, investing in debt involves a different risk profile and a different type of underwriting and he flagged the importance of diversification across the AI ecosystem. It requires a systems-level approach that invests across the spectrum from hyperscalers to the companies set to benefit from AI efficiencies.
Shah concluded with another risk: what if machine learning runs out of data? Experts in the field have flagged that at some point the data could dry up. One way ahead could result in more data being collected from AI watching and listening to how humans behave physically.
