CATL Invests 15.5 Billion Yuan in AI Energy Chain in Three Months
Alina Collins
CATL poured ¥15.5 billion into AI infrastructure in three months — power equipment, data centers, and a stake in DeepSeek — betting that the ultimate bottleneck for AI is not chips but electricity.
Where did the ¥15.5 billion go?
¥4.1 billion for a stake in Hangzhou Zhongheng Electric, a maker of high-voltage DC power equipment — the hardware that delivers grid power into a data center.
¥6.4 billion into GDS Holdings, a data-center operator that builds and runs server facilities.
CATL became DeepSeek's second-largest outside investor (behind only Tencent) in its first external funding round; Reuters reported the round valued DeepSeek at over $50 billion.
This means → three deals cover power delivery upstream, data-center operations midstream, and AI applications downstream — a full energy chain in one portfolio.
Why electricity, not chips?
Chairman Robin Zeng's long-term strategy: transform CATL from the world's largest EV-battery maker into a zero-carbon energy infrastructure provider.
The core thesis: GPUs determine computing performance, but power determines whether AI can scale affordably.
Nvidia CEO Jensen Huang has repeatedly called AI "ultimately an energy problem"; Elon Musk lists cheap electricity as AI's biggest constraint — both lines track CATL's direction closely.
In plain terms = chips are the brain; electricity is the blood supply. CATL chose to be the one pumping the blood.
Does this playbook look familiar?
In EV batteries, CATL built its dominance through vertical integration — controlling the chain from raw materials to cells to finished packs.
Now it is attempting the same playbook in AI energy: Zhongheng Electric handles power delivery, GDS handles the facilities, DeepSeek represents the end consumer of that power.
This reflects an industrial logic, not a financial one — CATL is wiring three links into a closed loop.
How big can the energy-storage business get?
CATL expects energy-storage products to account for roughly half of its global revenue by 2030.
Context: the IEA reports that global data-center investment nearly doubled between 2022 and 2024, reaching about $500 billion, with power now the sector's single largest constraint.
This means → the larger the AI models and the heavier the inference workloads, the closer stable power supply and storage come to rivaling chip performance in strategic importance.
Where are the risks?
Integrating power equipment, storage, and data-center operations requires cross-industry coordination — CATL's track record is in battery manufacturing, not grid infrastructure or facility management.
Large AI campuses typically need one to two years to reach full utilization, stretching the payback timeline.
In plain terms = vertical integration worked in batteries, but the same script does not automatically replay in a different industry — that is the central variable determining whether ¥15.5 billion of investment delivers on its thesis.
Content is for reference only, not financial advice.