CATL, the largest manufacturer of electric-vehicle batteries in the world, would like to buy a piece of DeepSeek. That is a slightly strange sentence. CATL makes batteries. DeepSeek makes a chatbot that is unusually good at math and has spent more than a year making OpenAI and Anthropic nervous about their pricing. A battery company and a frontier AI lab are not, on the face of it, in the same line of work, and yet here is CATL, joining the round.
The round is worth pausing on, because it has grown. DeepSeek’s first step toward outside capital was a fairly modest thing, around $300 million at a valuation above $10 billion. The lab is now reportedly raising something closer to $7 billion to $10 billion at a valuation between $45 billion and $50 billion, and its founder, Liang Wenfeng, has begun describing the company in terms of AGI, which is the sort of thing that happens to valuations. The rest of the investors are the expected crowd: a national AI investment fund, the semiconductor sector’s “Big Fund”, Tencent, Alibaba, Hillhouse. An internet conglomerate investing in an AI lab is barely news. A battery manufacturer is the news.
The explanation is that CATL has been doing this for a while, just less visibly. Investors linked to CATL agreed to pay nearly $1 billion for a 38 percent stake in VNET, one of China’s oldest data-center operators. CATL itself spent roughly $600 million on a 49 percent stake in Hangzhou Zhongheng, which makes the high-voltage direct-current equipment that data centers use to keep from wasting electricity. It is selling its sodium-ion batteries as storage for power grids and data centers. Somewhere in there CATL looked hard at artificial intelligence and concluded, not unreasonably, that underneath the software it is an electricity business, and decided to own the parts of the electricity business that the software cannot run without.
This is a defensible view. An AI model is a machine for turning electricity into tokens. A battery is a machine for holding electricity until something wants to turn it into something else. DeepSeek and CATL are running the same trade from opposite ends, and once that is clear, a battery company investing in an AI lab stops looking like a category error and starts looking like vertical integration. CATL is less interested in DeepSeek’s chatbot than in DeepSeek’s power bill, which CATL would very much like to supply.
This is also, for what it is worth, the conclusion almost everyone reaches after thinking about AI infrastructure for long enough. It is why SpaceX wants to put data centers in orbit and charge them for sunlight. Power is the input that runs short first, the chips and the algorithms having turned out to be the comparatively tractable part, and the companies worth watching are the ones quietly buying the power. CATL has gotten there by way of batteries, which is a perfectly good way to get there.
Which leaves DeepSeek’s cap table, a document that has become unexpectedly readable. Its investors now include two state funds, the two companies that run most of the Chinese internet, a large investment firm, and the national battery champion. That looks less like a roster of people betting DeepSeek’s API will out-earn OpenAI’s than like a roll call of Chinese industrial capacity, arranged around the national AI lab the way chairs are arranged around a table. American AI labs raise from hyperscalers and venture funds. European ones increasingly raise from government gigafactory programs. DeepSeek raises from China, more or less the whole of it, which is a slightly unusual way to run a venture round and a very deliberate way to run an industrial policy.
–
By the Control Plane Editorial Team