When Jensen Huang praised OpenClaw final week, the ripples reached Hong Kong inside hours. Shares in MiniMax and Zhipu AI jumped greater than 20% after the Nvidia CEO declared in a CNBC interview that the quickly spreading open‑supply agent framework was “definitely the next ChatGPT.”
Overseas traders as soon as dismissed China’s AI push as a constrained, second-tier effort. But now, strategists argue that the nation could also be higher positioned for AI, due to cheaper energy, rising capital spending, and a swarm of open-source builders. Those self same analysts are additionally questioning whether or not the U.S. AI growth, after years of sky-high valuations and information middle spending, is working out of steam.
“We’ve actually reduced our exposure to U.S. tech,” Mohit Kumar, Jefferies’ international macro strategist, informed Fortune on the funding financial institution’s Asia Discussion board in Hong Kong final week. “We believe that China is the big winner in this tech war for a number of reasons: valuation, wider adoption of AI, an advantage in power generation.”
“China basically has unlimited access to cheap energy, whereas the U.S. has this massive energy bottleneck,” Jefferies’s international head of fairness analysis Chris Wooden mentioned to Fortune.
The nation could have roughly 400 gigawatts of spare energy capability by 2030, equal to a few instances what the world might want to meet information middle calls for, in keeping with Goldman Sachs. In distinction, the U.S. is scuffling with getting old infrastructure and a scarcity of producing capability, resulting in spiking energy costs in information middle states like Virginia.
In China’s western provinces like Ningxia and Gansu, electrical energy can price as little as 5 cents per kilowatt-hour, versus 25 cents in Beijing or Shanghai, or 40 cents in some components of the U.S, in keeping with Baidu’s chief finance officer Henry He, who famous for attendees that energy could make up about 35% of inference prices.
China’s manufacturing sector additionally helps its AI sector, notably in purposes that work together with the bodily world. For instance, He famous that Baidu’s Apollo robotaxis can cost one yuan ($0.15) per mile and nonetheless be working at break-even within the metropolis of Wuhan.
Within the U.S., robotaxis producers have had to decide on between LiDAR sensors or visible cameras as a result of price. However He argued that, in China, “we don’t need to make that difficult choice,” as each are reasonably priced due to Chinese language manufacturing.
The identical logic extends to autonomous aviation. EHang, for instance, depends on a home provide chain that may present reasonably priced batteries and digital elements to construct its aerial autos. “We have an extremely competitive component cost, and we can turn it into a very competitive selling price,” CFO Conor Yang defined to Fortune.
A peaking U.S. AI cycle?
Wooden’s optimism in China is contrasted by his questions in regards to the U.S. The Jefferies strategist thinks that traders are beginning to ask questions in regards to the sum of money being spent within the U.S. on AI infrastructure, and expects U.S. capital spending to peak this 12 months.
In an interview with Fortune, Wooden described the U.S. personal fairness market as affected by “a giant case of financial constipation,” with tens of 1000’s of portfolio firms ready to IPO. As well as, personal fairness has lately poured cash into the software program sector, proper as AI now threatens to erode their enterprise fashions.
Personal credit score has been examined because the chapter of auto components provider First Manufacturers Group final 12 months, which spooked retail traders and pushed some asset managers, like Blue Owl, to limit withdrawals. This 12 months’s “AI scare trade” has solely made issues worse as retail traders attempt to pull their capital from fund managers, main different fund managers, like BlackRock and Morgan Stanley, to additionally cap withdrawals.
Monetization remains to be troublesome
Nonetheless, China’s AI firms might battle to cost a lot for his or her merchandise within the nation’s hyper-competitive atmosphere, the place labs compete on each value and efficiency.
“It is very difficult for a single company to always [have] a top model globally,” He, from Baidu, mentioned. (Baidu, an early mover within the area with its ERNIE mannequin, has since misplaced floor to Chinese language rivals like Alibaba and DeepSeek)
Startup MiniMax, for instance, generated $79 million in income final 12 months, but reported a $1.8 billion internet loss. (Traders don’t appear to care, driving shares up by over six instances because the startup’s IPO in early January.)
The issue for AI labs, whether or not within the U.S. or in China, is that any lead in efficiency may disappear in a matter of months, as different labs catch up by providing fashions with near-frontier-level efficiency, typically on an open-source foundation.
Wooden, from Jefferies, thinks that giant language mannequin suppliers will find yourself like utilities: capital-intensive, commoditized, and unlikely to earn sustainable returns. As an alternative, he argued that China’s AI growth “is going to be in applications—cheap applications—made possible by open‑source models and very cheap power.”
Agentic AI could possibly be the subsequent battleground. Chinese language firms giant and small are rapidly rolling out their very own OpenClaw frameworks, and native governments are providing subsidies to “one-person companies” constructing their very own AI agent startups.
Chinese language tech firms have already got sturdy shopper platforms to place AI brokers in entrance of customers. Tencent’s WeChat, for instance, has over 1.3 billion month-to-month lively customers and hosts thousands and thousands of mini packages spanning commerce, transport, way of life, and finance.
“We’re going to see some interesting innovations in China. This is the only area where the U.S. and the West are going to be behind,” Michael Bruck, founding accomplice at 71 Capital, informed Fortune.
“Is it a huge stretch of imagination to think the next version of a mini program is going to be an agent built into WeChat?”
