Take the money, keep the control — DeepSeek's strange first round

Here's the deal: DeepSeek just took outside money for the first time ever. The round is about 50 billion yuan, roughly $7.4B. That single raise pushed its valuation above $50B, making it the most valuable AI startup in China. Stop there and it reads like "another Chinese AI megaround." But the real story isn't the number — it's the structure.

Normally, when a giant like Tencent or CATL writes a check this big, they get equity and voting rights in return. Not here. Investors don't get DeepSeek shares directly. Instead they put money into a separate limited partnership controlled by founder Liang Wenfeng, and that money is locked up for five years. On top of that, only China's National AI fund received voting rights. Everyone else funds the thing and stays quiet.

So DeepSeek made a very blunt statement: "We need outside money, but we will not hand anyone the power to steer this company." Today let's unpack who DeepSeek is, why the structure is built this way, what each side gets, and what it signals for the AI landscape.

The players — DeepSeek, Liang Wenfeng, and the silent whales

DeepSeek is a Hangzhou-based AI lab founded in July 2023. Its parent is High-Flyer, a quant hedge fund, and DeepSeek built its own models on that fund's money and stockpiled GPUs. In early 2025 it shook US AI stocks by shipping strong open-weight models — a "wait, you can get a good model this cheap?" shock to the market. Famously, it had never taken a cent of outside capital before this.

Liang Wenfeng is the founder of both High-Flyer and DeepSeek. In this round he personally put in about 20 billion yuan — roughly $3B. The key isn't a big outside whale; it's that the founder is the single largest backer. Usually taking investment dilutes a founder, but Liang poured in more of his own money to tighten his grip on control.

The whales funded the rest. Per reporting, Tencent considered/contributed about 14 billion yuan, battery giant CATL about 5 billion yuan, and China's National AI Industry Investment Fund put in roughly $150M directly. Ordinarily these players would demand equity and a voice. This time most accepted "fund it and stay locked for five years." Why they agreed is the real point of this story.

By the numbers

Item Detail
Round type First-ever external raise
Size ~50B yuan (≈ $7.4B)
Valuation $50B+ (China's highest-valued AI startup)
Largest backer Founder Liang Wenfeng (≈ $3B)
Key participants Tencent (≈$1.4B), CATL (≈$700M), National AI Fund (≈$150M)
Structure Investors fund an LP controlled by Liang, not DeepSeek equity
Terms 5-year lock-up; only the state fund holds voting rights

Read it line by line and the strategy is clear. First, the founder is the largest backer. Controlling a company needs equity and votes, and Liang didn't want outside money shaking that. So he wrote the biggest check himself and nailed down "I'm the captain of this ship."

Second, investors don't get direct DeepSeek equity. They sit in an LP Liang manages, with Liang as a layer between the company and the money. Outside capital comes in but can't touch the board or decisions. Cash flow and control flow are deliberately split.

Third, a 5-year lock-up and state-only voting. Money that can't be pulled for five years filters out quick-flip speculators — only long-haul partners welcome. Giving votes solely to the National AI fund is a deft balance: keep alignment with the government in a Chinese context while blocking private whales from steering.

What each side gets — why everyone accepted the concession

DeepSeek and Liang win cleanly. They secured $7.4B of outside capital while keeping 100% of control. AI is an astronomically expensive game of GPUs, power, and talent, and one hedge fund's treasury only goes so far. So they took the money but kept outsiders out of the calls that matter — model direction, open-weight strategy. Money and freedom at once.

Why did Tencent and CATL accept such an unfavorable-looking deal? Because the seat itself is scarce. DeepSeek had never taken outside money, so this round was nearly the only chance to get into China's priciest AI startup. No votes, sure — but if the valuation climbs, the LP stake climbs with it, so the financial upside is intact. They bet on future value instead of control.

The National AI fund's win is different in kind. It alone got votes and is free of the lock-up. This isn't a plain financial bet; it's China institutionally securing a tie to its flagship AI lab. The three interests meet neatly at one point — money apart, power apart, government alignment intact — and that's the strength of the structure.

Echoes of the past — the upside and shadow of founder control

A founder taking outside money while keeping control is familiar in Silicon Valley. Google, Meta, and Snap went public with dual-class structures: founder shares carry multiplied votes, so even diluted, the founder keeps control. In the success cases this let them push a long-term vision without flinching at short-term shareholder pressure.

But there's a shadow. When control concentrates too heavily in one person, the checks vanish if that person is wrong. Outside holders can't correct the leadership, so a bad direction is hard to stop — which is why some dual-class companies faced governance fights. DeepSeek's structure goes a step further: investors don't even hold direct company equity, so the checks are weaker still.

DeepSeek's context differs from the Valley, though. In China, geopolitics — government alignment, export controls — is always underneath. Giving votes only to the state fund reads as a Chinese-style balance: block private control, keep the government relationship. The same "founder control" structure means something quite different depending on the soil it's planted in.

Competitor counter-play — US big labs and rivals at home

This round didn't appear in a vacuum. US big labs like OpenAI and Anthropic raise tens of billions and run models on vast capital. To stay in that arms race, DeepSeek needed ammo — and $7.4B is the ammo. But unlike US labs that spread equity and influence widely to investors, DeepSeek took the opposite road and clamped down on control. It's a declaration: "We do this our way."

Competition at home is fierce too. Alibaba- and Tencent-linked efforts and a wave of new labs are vacuuming up AI capital. DeepSeek printing a $50B+ valuation in that mix signals where Chinese AI capital is flowing. Tencent funding a rival despite being a competitor shows the whales' logic: if you can't beat them, at least ride along.

For latecomers and other labs, the calculus sharpens. DeepSeek showed you can close a megaround even on "take the money, keep control" terms, so any founder with leverage will eye the same card. Labs without leverage still trade equity and power for cash. In the capital markets, "who can keep control" is becoming a new class line.

So what actually changes

If you're a developer or company using AI models, the immediate change is small. But $7.4B means DeepSeek can keep investing aggressively in open-weight models, so cheap, capable open options are more likely to keep coming. Good news if your team is cost-sensitive.

If you're an investor or market watcher, the structure itself is the lesson. If "take money, give no equity or votes" proves it works, startups with leverage may copy it. For capital providers, whether to put in "powerless money" becomes a new judgment call.

If you watch geopolitics, note the state-fund-only votes. China's institutional tie to its flagship AI lab just got tighter. How that plays into the US–China AI race and export-control picture is too early to call. But it's clear that geopolitics is woven into even a capital structure.

One step further — what "separating money from control" asks of AI capitalism

More than the $7.4B, what's worth chewing on is the design itself: splitting money from control. In normal capitalism, whoever provides money takes a matching share of power — equity is voice, and big checks buy board seats and a hand on direction. DeepSeek broke that formula head-on. "Take the money, get zero power" — and giants like Tencent accepted it. That, paradoxically, shows how scarce the AI asset is. When the seat itself is rare, concessions you'd never imagine become possible.

Why it matters: it previews the power structure of the AI era. The core talent that builds models and whoever holds the data and compute becomes the "buyer," while the capital providing money becomes the "seller" — a reversal. In the dotcom or mobile booms, capital picked startups; at the top AI labs the opposite can appear. Teams that can truly build great models are few, and capital wanting in is plentiful. DeepSeek's structure is the most blatant institutionalization of that supply-demand imbalance.

But the structure carries clear risk. When control concentrates too heavily in one person, there's no brake if their judgment is wrong. With investors not even holding direct equity, the mechanism to correct management effectively vanishes. If Liang keeps making good calls, the structure shines; one big misstep and it flows on unchecked. From a governance lens, that's the weakest point — and how investor frustration surfaces when the five-year lock-up ends is its own variable.

There's also the geopolitical meaning baked in. Giving votes only to the National AI fund reads not as a mere capital design but as a device tying the Chinese government to its flagship AI lab institutionally. With US–China AI rivalry and export controls intensifying, this balance — block private whales, keep government alignment — sketches one model of Chinese-style AI capitalism. Copying it wholesale in Korea or the US would be hard, but the question "how do you protect control of a critical-tech company" is valid everywhere.

Finally, the implication for markets like Korea is worth noting. A Korean AI startup with leverage might study the DeepSeek-style card of "take money, keep control." Conversely, for Korean VCs and conglomerates providing capital, whether to put in "powerless money" becomes a new judgment call. In the end, this round is more than a funding headline — it re-asks "who controls what" in the AI era. The answer is still open, and whether other labs run similar experiments in their rounds is the thing to watch.

🥄 Three Things You're Probably Wondering

— So what does this mean for me? Almost no direct impact. But DeepSeek is now well-funded, so cheap, decent open-weight models are more likely to keep arriving. Good if you're a developer watching costs.

— Why fund it if you get no equity and no votes? Because the seat was scarce. DeepSeek had never taken outside money, so this was nearly the only way in. No control, but if the valuation rises the LP stake rises with it — financially worth the bet.

— Is it ahead of the US big labs? Too early to say. On raw capital, OpenAI and Anthropic still lead; DeepSeek's edge is efficiency — good open models for less. Different roads, so a straight comparison is hard, and we'll have to watch.

References

Numbers and criteria are as of announcement and may change. Investment calls are yours to make!