Even *Black Myth* Isn’t as Wild as This! DeepSeek Is Aiming for $10 Billion

I'd give it a 7. Because on the surface this sounds like capital market buzz, but at its core, it’s really about a Chinese homegrown tech company finally daring to play big.

One moment they were saying they weren’t in a hurry to raise money, and the next there were reports that the first funding round would target at least $300 million, with a valuation of no less than $10 billion. You want to ask whether that contrast is huge? Of course it is. But this doesn’t feel like the kind of overhyped product launch where some companies promise the moon and crash on delivery. It feels more like a veteran player spotting a small-studio console from the PS1 era and suddenly deciding to upgrade it into a development platform capable of running blockbuster titles—an aggressive move, with a very clear goal.

First, look at the foundation. DeepSeek was incubated by High-Flyer Quant, and Liang Wenfeng oversees both High-Flyer Quant and DeepSeek. That background matters a lot, because this isn’t the kind of AI startup that only knows how to tell stories and burn through investors’ money. According to public information, High-Flyer Quant’s average return in 2025 was 56.55%, ranking second among China’s quantitative private funds with assets over 10 billion yuan. Its average returns over the past three and five years were 85.15% and 114.35% respectively, and its assets under management have already exceeded 70 billion yuan. Put simply, they were never short on money before—they just weren’t in a rush to take outside funding.

That also explains Liang Wenfeng’s earlier remark that “we are not in a hurry to raise funds right now.” The reason is very practical: concern that outside investors would interfere in decision-making. That concern is far from unnecessary. We’ve seen too many projects change flavor the moment funding comes in—much like how a lot of Japanese studios back in the day gradually shifted from making the games they actually wanted to make into producing service-style products that looked good as just another line in a financial report. Players can smell that instantly.

That said, if they really do raise this round, the timing isn’t bad either. As early as March 2025, tech giants such as Tencent and Alibaba were already looking to discuss potential cooperation with DeepSeek. So if they’re launching their first external funding round now, it’s clearly not about survival, but about scaling up for the next stage. At this point in the AI race, running on momentum alone is no longer enough—computing power, ecosystem, and chip adaptation all require real money.

What’s even more worth watching is the product cadence. DeepSeek’s next move is to launch the brand-new multimodal large model V4, the first major version update since the R1 reasoning model in January 2025. Public information shows that V4 natively supports image, video, and text generation, and has already completed deep optimization and adaptation with leading domestic AI chip makers such as Huawei and Cambricon, with a focus on compute scheduling and inference efficiency. There’s nothing wrong with that direction, because the AI competition today is no longer just about comparing parameter counts. It’s more like the development wars at the end of the PS3 era—whoever can squeeze the most out of the hardware looks like the real deal.

Then look at what it has already delivered. In December 2024, DeepSeek officially launched and open-sourced DeepSeek-V3. According to the article, V3 comprehensively outperformed ChatGPT-4o in mathematics, coding ability, and Chinese knowledge Q&A. Even more striking was the cost: V3 used only 2000 GPU chips, with a total training cost of $5.576 million; by comparison, GPT-4o used about 25000 GPUs, with a total training cost close to $100 million. That gap doesn’t feel like a minor iterative improvement between console generations—it feels more like taking a well-tuned old engine and getting frame rates out of it that even the shiny new machine next door can’t necessarily sustain reliably.

So if this funding round goes through, my view is simple. First, it will turn DeepSeek from a “technical dark horse” into a true “industry player.” Second, whether outside capital can resist the urge to meddle will determine whether it goes on to become something like Black Myth—a breakout hit that smashes through expectations in one blow—or ends up as a pile of half-finished products with wavering direction. What annoys player communities most right now? It’s not that you raised money. It’s that after taking the money, you start talking in empty slogans, putting on airs, and slowing down updates.

If you normally follow China’s domestic AI scene, this is worth keeping a close eye on. Not because the $10 billion figure is so shocking, but because companies in China are finally starting to prove that competitive models do not necessarily have to be built by brute-forcing with astronomical spending. The real challenge is getting efficiency, cost, and ecosystem right at the same time—and that, more than any trending slogan, is what real hard power looks like.