DeepSeek is valued at 400 billion, with only twenty employees, relying on 50 billion in financing to invest in computing power, overturning traditional business organizational structures.
CATL and Tencent have used 50 billion RMB in hard cash to forcibly bridge the chasm between virtual computing power and heavy industrial manufacturing. When DeepSeek's first-round financing valuation blew straight through the 350-billion-to-400-billion RMB range, the classical investors in the primary market who were still crunching price-to-sales ratios were left utterly speechless.
This is by no means a simple algorithm frenzy. If you take a closer look at this extremely lavish payment list, Tencent's investment stems from an instinct for cloud defense—an effort to lock a model base with such terrifying throughput capacity firmly inside its own server network. But CATL's entry tears open the coldest industrial reality behind this sky-high financing round.
As the undisputed global titan of power batteries, CATL shelling out serious money to buy a large language model that writes code and does reasoning reflects a deeper interest: paying rent in advance for an intense craving for industrial brains and embodied intelligence. On the eve of full automation sweeping heavy-industry assembly lines, relying solely on rigid script commands can no longer control increasingly complex robotic collaboration arrays. A new-energy workshop needs an extremely intelligent, ultra-low-cost open-source hub to take over the robotic arms and unmanned factories in the physical world. This massive sum is, at its core, a digital brain reserve that a manufacturing mogul is stockpiling for next-generation productivity.
This grand infrastructure ambition stands in suffocating contrast to the company's extremely skeletal physical form.
If you trace through the underlying corporate registry grid on Tianyancha and peel back the layers, you will see the true revenue-generating structure of the new-generation algorithm giant, which defies common sense. Hangzhou DeepSeek Artificial Intelligence Basic Technology Research Co., Ltd. just completed a very small capital expansion in April of this year, with its registered capital crawling from 10 million RMB to a mere 15 million. Even more striking data hides in the social security participation records—shrinking from four people in 2023 to twenty people in 2024.
Twenty people propping up a 400-billion-RMB valuation. That is a dimensionality reduction massacre against traditional business organizational structures.
In the classical internet era, a giant with a hundred-billion market cap needed tens of thousands of ground-force salespeople, sprawling administrative middle offices, and office towers scattered across the country. But inside DeepSeek's system, those twenty natural persons paying social security are not employees in the traditional sense—they are hyper-concentrated nodes of pure intellect. This newly raised 50 billion RMB is absolutely not meant to expand management layers or build a lavish CBD headquarters. Its ultimate destination is brutally cold—entirely converted into GPU arrays on computing clusters, into electricity blazing through ultra-large liquid-cooled data centers, into the physical fuel that pushes large models across the threshold to next-generation reasoning.
This extreme thinness in headcount coupled with extreme heaviness in computing consumption has completely overturned the rules of tech entrepreneurship over the past few years. Liang Wenfeng and his team, through their overwhelming ability to cut model-inference costs, have driven those middlemen—who tried to fine-tune open-source models and resell API access with teams of a few hundred people—straight into a dead end.
This hunt, led by top-tier industrial capital, declares the exit of pure financial investors from the big-model betting table. In this cycle where the purity of computing power and the closed loop of industry determine the final outcome, a model capable of swallowing astronomical capital must find an extremely hard landing scenario. That social security roster with only twenty names on it is not just a spectacle—it is a liquidation manifesto written to the old business era. It stands as cold, hard proof that in the ledger of the next industrial revolution, capital no longer pays a premium for mediocre human resources. It pays unconditionally only for the few sharpest human minds—and for silicon-based computing power that can take over the physical world.
