# DeepSeek took outside money, and the cap table is the story

URL: https://www.thedeepfeed.ai/posts/2026-06-11-deepseek-first-external-raise/
Category: Business
Published: 2026-06-11
Author: the-deep-feed
Tags: deepseek, china, fundraising, tencent, catl, open-weights, liang-wenfeng
Kind: deep

> The lab that refused capital is closing a $7.4B first round at up to $59B. Why a battery maker leads it, why Tencent's check is defensive, and why Liang kept 40%.

## TL;DR

- **DeepSeek** is closing its first-ever external round: roughly **50 billion yuan (~$7.4B)** at a post-money valuation of **350–400 billion yuan ($52–59B)**, per [SCMP](https://www.scmp.com/tech/big-tech/article/3355818/deepseek-nears-us7b-haul-first-ever-funding-round-backing-tencent-catl) and [CNBC](https://www.cnbc.com/2026/06/03/deepseek-slated-to-draw-7-billion-in-maiden-fundraising-sources-say.html). That is roughly a **six-fold** rerating from a ~$10B mark weeks earlier.
- Founder **Liang Wenfeng** is personally writing ~**20 billion yuan**, about **40%** of the round, keeping him the largest single shareholder. He is buying control, not just capital.
- The two anchor outside checks are **Tencent (~10B yuan)** and battery giant **CATL (~5B yuan)**. CATL does not want a model. It wants the **power contract** for the data centers the model runs in.
- The syndicate is **all-domestic, fewer than 10 names, no foreign capital** — by [TechNode](https://technode.com/2026/06/04/deepseek-in-talks-to-raise-7-billion-from-tencent-catl-and-other-investors/)'s count. The lab that built its brand on neutrality is now bound to China's platform incumbents.
- This is the capital side of the same story we covered when [open-weight Chinese models passed Claude](https://thedeepfeed.com/posts/2026-06-04-open-weight-coding-frontier-caught-claude/). The money arrives exactly as the usage charts turn.

For two years, the most interesting thing about **DeepSeek** was the money it would not take. The Hangzhou lab was the rarest kind of frontier shop: profitable parent, no venture round, no board of outside investors, no pressure to monetize the weights it kept giving away for free. Its founder told Chinese media it wanted to chase artificial general intelligence without the distortions that capital brings. Investors who circled were turned away. The lab that "couldn't be invested in" was a deliberate posture, and it held.

It does not hold anymore. According to the [South China Morning Post](https://www.scmp.com/tech/big-tech/article/3355818/deepseek-nears-us7b-haul-first-ever-funding-round-backing-tencent-catl), DeepSeek is finalizing its first external funding round: about 50 billion yuan, roughly $7.4 billion, at a post-money valuation between 350 and 400 billion yuan, or [$52 billion to $59 billion](https://invezz.com/ng/news/2026/06/03/deepseek-nears-dollar74-billion-raise-at-valuation-of-up-to-dollar59-billion/). [CNBC](https://www.cnbc.com/2026/06/03/deepseek-slated-to-draw-7-billion-in-maiden-fundraising-sources-say.html) and [Axios](https://www.axios.com/2026/06/03/china-deepseek-billion) confirmed the same figures from separate sourcing. Bloomberg called it one of the largest startup financings in Chinese history.

The headline is the number. The story is the cap table.

![A line-art ledger with one bold red entry standing out among neat columns, the moment a self-funded lab admits outside names to its register](/post-images/2026-06-11-deepseek-first-external-raise/cap-table-ledger.jpg)

# The lab that did not need a round

To understand why this raise matters, you have to understand why DeepSeek spent two years not raising.

DeepSeek was spun out of High-Flyer, a quantitative hedge fund Liang Wenfeng co-founded, and the fund's profits bankrolled the lab's compute. That arrangement gave DeepSeek something almost no other frontier lab had: independence from the venture treadmill. It could release V3 and R1 under permissive licenses, price inference near cost, and absorb the margin compression, because no outside shareholder was demanding a return. The open weights were not charity. They were a competitive weapon a self-funded lab could afford to swing.

That independence was also the brand. When Western analysts argued that Chinese labs would inevitably be captured by state or platform interests, DeepSeek was the counterexample. It answered to no one. The reporting on this round names the tension directly. The neutrality is the thing being spent.

> Taking outside capital changes the company's character. It formalises DeepSeek as a commercial entity with investors to answer to.
>
> — The Next Web, [on DeepSeek's maiden fundraising](https://thenextweb.com/news/deepseek-7-billion-maiden-fundraising), Jun 3, 2026

So the question is not "why does a frontier lab raise $7.4 billion." Frontier labs raise that before lunch. The question is why *this* lab, which made a point of refusing it, decided the cost of staying pure now exceeds the cost of taking the check.

# Reading the cap table

Start with who is in the round, and for how much. The figures below are the per-investor commitments reported by [36Kr](https://eu.36kr.com/en/p/3837092707693698) and corroborated by [TechNode](https://technode.com/2026/06/04/deepseek-in-talks-to-raise-7-billion-from-tencent-catl-and-other-investors/).

| Participant | Reported commitment | What they actually are | Read |
|---|---|---|---|
| Liang Wenfeng (founder) | ~20B yuan (~40%) | The CEO, via personal capital | Buys control, not exposure |
| Tencent | ~10B yuan | Platform + WeChat distribution | Defensive — can't let a rival lock DeepSeek |
| CATL | ~5B yuan | World's largest EV battery maker | An energy-infrastructure play, not an AI bet |
| National AI fund | undisclosed | State industrial capital | The official imprimatur |
| NetEase, JD.com | undisclosed | Platform incumbents | Distribution and cloud demand |

The first thing the table tells you is that this is not a normal financing. A normal $7.4 billion round at a $55 billion valuation would have a lead venture or sovereign fund anchoring 30 to 40 percent and a long tail of co-investors. This round has the *founder* anchoring 40 percent with his own money, two strategic corporates writing the next-largest checks, and a state fund providing cover. TechNode reports the syndicate is capped at [fewer than 10 investors, all domestic, with no foreign capital](https://technode.com/2026/06/04/deepseek-in-talks-to-raise-7-billion-from-tencent-catl-and-other-investors/).

That is not a cap table optimized to raise the most money at the best price. It is a cap table engineered for control and alignment. Every choice in it sacrifices financial efficiency for something else. Walk the three load-bearing names and the logic comes into focus.

# Why a battery maker wants in

CATL is the strangest name on the list, and it is the one that explains the most. Contemporary Amperex Technology is the world's largest manufacturer of electric-vehicle batteries. It does not build models, run inference, or employ AI researchers at frontier scale. On paper, a battery company taking a 5-billion-yuan stake in a language-model lab is a category error.

It is not. CATL is not buying a piece of DeepSeek's models. It is buying a seat next to DeepSeek's electricity bill.

The [reporting on CATL's plan](https://cnevpost.com/2026/05/22/catl-plans-to-invest-in-deepseek/) is explicit: the company is moving to sell power equipment to the large AI data centers whose energy demand is being driven by the surge in compute. A frontier lab's biggest non-chip cost is power, and power for AI is a problem of generation, storage, and conditioning at the megawatt scale. CATL makes exactly that hardware. Its stake in DeepSeek is a supplier relationship dressed as an equity check, the same maneuver [The Information first flagged](https://www.theinformation.com/articles/chinese-ev-battery-giant-catl-plans-invest-deepseek) when the plan surfaced.

And it fits a pattern CATL has been building for months. Since early April the company has taken an indirect stake in a Chinese supplier of data-center power-conversion gear and agreed to a roughly $942 million deal for a slice of a Nasdaq-listed Chinese data-center operator, [per Vector's reporting on the strategy](https://vector.news/catl-closes-the-loop-on-ai-compute/). The DeepSeek stake closes the loop: own the battery, own the power-conversion layer, own a piece of the data-center operator, and own equity in the model company that fills the racks. CATL is positioning itself as the energy substrate of Chinese AI, and the DeepSeek check is the marquee customer relationship at the top of that stack.

The signal underneath this is the one US hyperscalers have been screaming about for a year. The binding constraint on frontier AI is shifting from chips to electricity. When a battery giant decides the highest-return adjacency to its core business is an AI lab, it is pricing a future where compute is gated by power, not silicon. DeepSeek, famous for training a frontier-class model on a fraction of the usual compute budget, is an odd place to make that bet. Unless the bet is that even the most efficient lab in the world is about to spend like a utility.

![A charcoal sketch of a battery cell wired directly into a server hall, one red current line running between them, energy as the new gating resource](/post-images/2026-06-11-deepseek-first-external-raise/battery-to-rack.jpg)

# Tencent's defensive check

Tencent's 10 billion yuan is the largest outside commitment, and it is the easiest to misread as conviction. It is closer to insurance.

Tencent runs WeChat, the distribution layer through which a billion Chinese users will meet AI, and it has its own model effort. It does not need DeepSeek's weights; they are open, and Tencent can run them on its own cloud whenever it likes. What Tencent cannot tolerate is a rival platform locking up a preferential relationship with the country's most-used open-weight lab. ByteDance's Doubao is the competitive pressure here. If DeepSeek's next generation became the default reasoning engine inside a competitor's super-app while Tencent sat outside the round, that is the expensive outcome.

So the 10 billion yuan is a hedge against exclusion. The [analysis of the round](https://aiweekly.co/alerts/deepseek-raises-7b-led-by-tencent-and-catl) frames it cleanly: Tencent's stake is defensive, and non-participation would have been the costlier move. This is what a strategic investor does when the asset is too important to a competitor to leave unhedged. The check is not a verdict that DeepSeek wins. It is a verdict that Tencent cannot afford for DeepSeek to win for someone else.

That distinction matters for how to read the valuation. A $55 billion mark backed by genuine financial conviction is one signal. A $55 billion mark backed substantially by founder capital and defensive corporate hedging is a different one. The number is real, but the demand curve underneath it is bent by strategy, not by a clean auction of belief.

# The 40 percent that keeps Liang in control

The most telling line in the cap table is the founder's own. Liang Wenfeng is committing roughly 20 billion yuan of personal capital, about 40 percent of the round, which keeps him the [largest single shareholder after the raise](https://aiweekly.co/alerts/deepseek-raises-7b-led-by-tencent-and-catl).

Founders almost never do this. The entire point of raising outside money is to fund the company with *other people's* capital while diluting yourself as little as necessary. Liang is doing the opposite: he is putting in the single biggest check so that the outside money cannot, in aggregate, outvote him. He can afford it, because High-Flyer made him wealthy independent of DeepSeek. And he is spending that wealth to buy the one thing the round threatens, which is his unilateral control over a lab that was, until now, answerable to no one.

Read against DeepSeek's two-year refusal to raise, the 40 percent is a tell. Liang did not decide that DeepSeek suddenly needs outside investors. He decided that DeepSeek needs the *resources* those investors bring, the power relationships, the platform distribution, the state cover, while conceding as little governance as the structure allows. The personal check is the price of keeping the lab's direction in one person's hands after the register fills with names that will, for the first time, expect something back.

It works only because the rest of the syndicate is built to be passive. Strategic corporates investing for supply contracts and defensive hedges do not push for board control or product direction the way a financial sponsor would. The cap table is internally consistent: passive strategic money around a controlling founder. The result is a company that is now commercial and shareholder-bound on paper, but still steered by the same person, in the same direction, as when it took nothing at all.

# The money arrives as the charts turn

![A schematic worksheet on cream paper titled "CATL closes the loop": a four-tier energy-to-AI stack — battery cells, power-conversion gear, data-center operator, model company — connected by conduit lines, with a single red loop-arrow running from battery to model and back, labeled own every layer the electrons touch](/post-images/2026-06-11-deepseek-first-external-raise/catl-closes-the-loop.jpg)

Timing is the part that turns a financing story into a thesis. DeepSeek did not take its first outside capital at a low point. It took it at the exact moment Chinese open-weight models seized the usage charts.

We documented that reversal directly: in a three-week window, [open-weight Chinese models caught the coding frontier](https://thedeepfeed.com/posts/2026-06-04-open-weight-coding-frontier-caught-claude/), and Chinese systems took roughly half of all tokens flowing through OpenRouter. The lab that was called uninvestable is capitalizing precisely as the category it leads tops global usage. This is not a rescue round. It is a victory lap with a balance sheet attached, and the valuation rerating tells you the market agrees. The reported [six-fold jump from a ~$10 billion mark to $52–59 billion](https://www.cnbc.com/2026/06/03/deepseek-slated-to-draw-7-billion-in-maiden-fundraising-sources-say.html) in a matter of weeks is not a financing artifact. It is the market repricing a lab whose models just won the usage war.

And DeepSeek is not alone in the timing. The same week its round leaked, a separate Chinese AI lab incubated by the same Beijing research academy that once seeded Zhipu reached unicorn status, raising [1 billion yuan just 10 months after inception](https://eu.36kr.com/en/p/3837232087561601). Capital is flooding into Chinese frontier and applied AI at once, and it is overwhelmingly domestic capital chasing companies that build in the open. The funding map and the usage map are converging on the same coordinates.

That convergence is the real signal. For two years the Western critique of Chinese open-weight labs had a financial escape hatch: sure, the models are good and cheap, but the labs are unfunded curiosities that cannot sustain frontier R&D without a business model. This round closes that hatch. DeepSeek now has $7.4 billion, a state-blessed syndicate, a battery giant wiring up its power, and a founder who spent his own fortune to keep the wheel. The open-weight strategy is no longer a passion project running on hedge-fund overflow. It is a capitalized, aligned, nationally-backed institution.

# The round buys durability, not a margin business

The instinct is to read this as DeepSeek selling out, trading its independence for a $55 billion price tag. That misreads the structure. The cap table was built so that the founder keeps control, the strategic investors get supply and distribution relationships rather than governance, and the lab keeps doing the open-weight thing that made it valuable in the first place. Nobody in this round is buying the right to make DeepSeek act like OpenAI.

What the money buys is durability. It buys CATL's power infrastructure so the next model generation can train and serve at scale. It buys Tencent's distribution and the state fund's cover, which in China is a survival input, not a luxury. It buys the resources to keep giving the weights away while everyone else has to charge for them. The genius and the risk of the structure are the same fact: DeepSeek took outside capital without surrendering the thing outside capital usually demands, which is a path to extracting returns.

That is the bet the whole syndicate is making, and it is worth naming plainly. Tencent, CATL, and the Chinese state did not fund DeepSeek to turn it into a margin business. They funded it to make sure the most-used open-weight lab in the world stays Chinese, stays open, and stays aligned with the infrastructure they sell and the platforms they run. The cap table is not a financing. It is industrial policy with a founder's signature at the top, and the price of admission was letting Liang Wenfeng keep the pen.

:::editor[Update — June 22, 2026]
The round closed. What was a leaked "nearing completion" story on June 11 is now done: DeepSeek finalized its first external raise at **$7.4 billion**, making it [China's most valuable AI startup](https://www.livemint.com/global/deepseek-becomes-china-s-most-valuable-ai-startup-after-over-7-4-billion-fundraise-11781609546128.html). The closing details sharpen the cap-table thesis above rather than revise it. [Forbes](https://www.forbes.com/sites/anishasircar/2026/06/17/deepseek-just-raised-74-billion-heres-the-catch/) reports the deal was structured so that, despite the breadth of the syndicate, voting power concentrates almost entirely in state-aligned hands — control was the point, exactly as the structure predicted, only more so. And the first demand-side consequence has already surfaced: Microsoft is [weighing adding DeepSeek to its enterprise Copilot Cowork tool](https://en.sedaily.com/international/2026/06/17/microsoft-weighs-adding-chinas-deepseek-to-copilot) to cut inference costs. A capitalized, state-blessed open-weight lab is now a procurement option for the largest Western software vendor. The hatch this round closed is staying closed.
:::

## Sources

- [South China Morning Post — DeepSeek nears US$7b haul in first-ever funding round, with backing from Tencent, CATL](https://www.scmp.com/tech/big-tech/article/3355818/deepseek-nears-us7b-haul-first-ever-funding-round-backing-tencent-catl)
- [CNBC — DeepSeek slated to draw $7 billion in maiden fundraising, sources say](https://www.cnbc.com/2026/06/03/deepseek-slated-to-draw-7-billion-in-maiden-fundraising-sources-say.html)
- [36Kr — DeepSeek's financing nearing completion: Liang Wenfeng 20B, Tencent 10B, CATL 5B](https://eu.36kr.com/en/p/3837092707693698)
- [TechNode — DeepSeek in talks to raise $7 billion from Tencent, CATL and other investors](https://technode.com/2026/06/04/deepseek-in-talks-to-raise-7-billion-from-tencent-catl-and-other-investors/)
- [Axios — China's DeepSeek reportedly raising $7 billion](https://www.axios.com/2026/06/03/china-deepseek-billion)
- [The Information — Chinese EV Battery Giant CATL Plans to Invest in DeepSeek](https://www.theinformation.com/articles/chinese-ev-battery-giant-catl-plans-invest-deepseek)
- [CnEVPost — CATL reportedly plans to invest in Chinese AI star DeepSeek](https://cnevpost.com/2026/05/22/catl-plans-to-invest-in-deepseek/)
- [Invezz — DeepSeek nears $7.4 billion raise at valuation of up to $59 billion](https://invezz.com/ng/news/2026/06/03/deepseek-nears-dollar74-billion-raise-at-valuation-of-up-to-dollar59-billion/)
- [36Kr — Xingyuanzhi: a new unicorn, $1B raised 10 months after inception](https://eu.36kr.com/en/p/3837232087561601)

---

Canonical: https://www.thedeepfeed.ai/posts/2026-06-11-deepseek-first-external-raise/
Site: https://www.thedeepfeed.ai
Full corpus: https://www.thedeepfeed.ai/llms-full.txt