Opinion | Why CATL Wants to Be the Infrastructure of the AI Age

When the world's largest battery maker puts 5 billion yuan into an AI software firm, the transaction deserves more than a passing glance at the investment rationale. DeepSeek's first external funding round closed in June at 50 billion yuan — roughly $7 billion — drawing investors from Tencent to gaming companies. CATL's participation, however, signals something qualitatively different: not a financial bet, but a declaration of intent.

The intent, as CATL founder Zeng Yuqun articulated at this year's Summer Davos Forum in Dalian, is to turn China's 40 million electric vehicles into a distributed AI computing and power network. The concept is deceptively simple: those EVs are driven for roughly one hour a day. The other 23 hours, their batteries and onboard chips sit idle. Network them, and you have a vast, decentralized infrastructure capable of generating the AI tokens and supplying the electricity that the generative AI industry is consuming at a pace no existing grid is built to sustain.

"Our idea is very simple: connect the electric vehicles," Zeng told the Davos audience in English, without an interpreter. The simplicity of the framing is deliberate — and a little misleading. What Zeng described is a fundamental reimagining of what a car is. Not a transportation device. Not a depreciating consumer product. A node in a computing network. A revenue-generating asset.

This framing matters commercially as much as technically. Ni Jun, co-president of CATL's new energy engineering and R&D division, made this explicit in a conversation with Storm Media on the sidelines of the Dalian forum. A networked EV, he argued, converts car ownership from consumption into investment. The vehicle earns money while parked. That is a compelling sales argument for an industry that has long struggled to justify the price premium of electric over combustion — and it is an argument that only works if CATL can deliver on the infrastructure promise.

Whether it can is an open question, and the technical constraints are formidable. Standard industrial manufacturing targets Six Sigma quality — roughly two to three defects per million units. EV battery packs, which contain hundreds of individual cells, cannot tolerate even one failure; a single bad cell renders the entire pack unusable. For a networked system in which each battery becomes a grid node, shared among millions of vehicles in parking lots, that tolerance collapses further still — to approximately one defect per one billion units. That is closer to aerospace or medical-device standards than to conventional automotive manufacturing.

Chip limitations compound the challenge. Ni acknowledged that roughly one quarter of China's existing 40 million EVs lack the advanced driver-assistance hardware needed to contribute meaningful computing power. The near-term addressable fleet for the Car as Token model is therefore closer to 30 million vehicles — still a staggering number, but one that illustrates the gap between Zeng's vision and its current implementation.

CATL is not standing still while the technology matures. As of May 2026, it has installed 1,650 "Chocolate" battery swap stations across 127 Chinese cities, capable of exchanging battery packs in under 99 seconds. The Chocolate Swap Alliance counts GAC Group, Wuling, SAIC and SAIC Maxus among its members, with 3,000 stations targeted by year-end. These are positioned to become the physical nodes of the networked system Zeng envisions — not just swap points, but hubs for power distribution and AI compute on demand.

Europe has experimented with Vehicle-to-Grid (V2G) schemes that allow EV owners to sell stored electricity back to utilities. Commercial deployments exist in the UK, the Netherlands and Denmark. None has reached transformative scale, and none incorporates AI computing. What CATL is proposing would be the first system to bundle power arbitrage with token generation — and to do so at the scale of a national EV fleet.

The implications extend beyond China. If CATL succeeds in making its swap network a critical layer of AI infrastructure, it will have built a second globally dominant franchise alongside its battery manufacturing business. The first made it indispensable to every automaker that wants to compete in EVs. The second would make it indispensable to every company that wants to run AI at scale. That is not a diversification strategy. That is a bid to own the physical layer of the AI economy — and the DeepSeek investment is the software key that unlocks it.

（Related： DeepSeek V4 Matches U.S. AI Leaders While Cutting Memory Costs, Goldman Sachs Says ｜ Latest ）