• Spring 2026: China’s model labs stop playing small
• Mid‑April: DeepSeek’s modest raise that wasn’t
• Late April–early May: Beijing crashes the cap table
• Parallel track: Moonshot AI quietly becomes a revenue machine
• The open‑weight moment: from curiosity to asset class
• Beijing’s new doctrine: model capability as strategy
• Competing narratives: bubble, catch‑up, or power shift?
• What happens next

Chinese AI Lab Moonshot AI Raises $2 Billion at a $20 Billion Valuation Human Human coverage frames Moonshot AI’s $2 billion raise at a $20 billion valuation as part of a coordinated Chinese push to build national AI champions, often contrasted with DeepSeek’s potential $45 billion round. It highlights how state-backed capital, hardware constraints, and U.S. export controls shape both the scale of investment and the emphasis on efficient, domestically aligned AI models. @TC @TNW Chinese AI labs just turned the volume up on the global model race: one, Moonshot AI, has quietly hauled in $2 billion at a $20 billion valuation, while another, DeepSeek, is suddenly being courted at more than double that — and now sits at the center of Beijing’s AI strategy.

§Spring 2026: China’s model labs stop playing small

In early 2026, Chinese AI companies were still widely seen as capital‑constrained cousins to their U.S. counterparts, boxed in by U.S. chip export controls and far smaller funding rounds. But the performance of a new wave of large language models (LLMs) — cheaper, open‑weight, and increasingly competitive with OpenAI and Anthropic — began to flip the script.

DeepSeek broke out first. The lab “came to prominence in early 2025 after launching a large language model that trained on a fraction of the compute power and at a fraction of the cost of the big U.S. models like those from OpenAI and Anthropic.”¹ That technical flex — strong reasoning and coding at discount compute — positioned DeepSeek as proof that frontier AI could be built without Nvidia’s latest GPUs, a critical political point for Beijing.

Moonshot AI followed fast. Founded in 2023 by former Meta AI and Google Brain researcher Yang Zhilin, the Beijing‑based lab built its reputation on the Kimi series of open‑weight LLMs. Its Kimi K2.5 model “took the coding world by storm earlier this year, nearly topping benchmarks and posting performance figures close to that of Open AI and Anthropic’s models at the time.”² That was the moment global developers started to see Chinese open‑weight models as serious alternatives, not curiosities.

§Mid‑April: DeepSeek’s modest raise that wasn’t

By mid‑April 2026, DeepSeek’s story still looked more or less like a classic high‑growth AI startup. It was preparing “a $300m raise at a $10bn valuation, with Alibaba and Tencent talking”³ — big, yes, but in the same order of magnitude as plenty of Western AI rounds.

Crucially, DeepSeek had been an outlier in one respect: founder Liang Wenfeng, a hedge fund billionaire who controls nearly 90% of the company, “has not previously sought out investors,”¹ preferring to fund the lab internally. The pivot to outside money was defensive as much as ambitious. Faced with rivals raiding his staff, “Liang opted to raise funds in order to offer employees shares in the company.”¹

At this point, DeepSeek’s trajectory still felt like a big, but basically private‑sector, play: a founder trying to build a talent moat, cloud giants circling, China’s state capital watching from a polite distance.

§Late April–early May: Beijing crashes the cap table

Then the numbers exploded.

Within weeks, the prospective valuation leapt: “DeepSeek is in talks to raise its first round of venture capital, and in just a few weeks, its potential valuation has soared from $20 billion to $45 billion.”¹ The reason wasn’t just froth — it was politics.

China’s main state‑backed semiconductor investment vehicle, the China Integrated Circuit Industry Investment Fund — universally known as the “Big Fund” — stepped in. Bloomberg and the Financial Times reported that this fund “is now in talks to lead DeepSeek’s first external funding round at a valuation of approximately $45bn.”³

That matters because the Big Fund has, since 2014, been the engine room of China’s semiconductor ambitions, pouring “more than $50bn into Chinese chip-design, fabrication, packaging, and equipment companies.”³ Its “mandate was overwhelmingly focused on the silicon side of the AI stack” — fabs, memory, EDA tooling — not model labs.³

Backing DeepSeek is therefore a policy pivot disguised as a term sheet. As one analysis puts it bluntly: “DeepSeek’s $45bn valuation is also Beijing’s strategic statement.”³ In other words, the state has decided that in a world of constrained chips, model capability itself is now a national security asset.

DeepSeek, for its part, has engineered its stack for the new reality. The lab’s models are “optimized to run on chips made by China’s hardware giant Huawei Technologies,” and that combination “is considered a powerful duo for the nation to develop its own AI to rival the United States.”¹ If Washington won’t sell Nvidia, Beijing will bankroll Huawei‑plus‑DeepSeek.

Cloud giants Tencent and Alibaba are reportedly still in talks to participate too, underscoring that this is not just a state vanity project but a commercial anchor for China’s entire AI infrastructure.¹

§Parallel track: Moonshot AI quietly becomes a revenue machine

While DeepSeek’s valuation theatrics grab headlines, Moonshot AI has been running a different playbook: build open‑weight models, ship usage, then weaponize the numbers.

By late 2025, Moonshot had already raised heavily, hitting a $4.3 billion valuation and then jumping “to $10 billion following a $700 million raise” by early 2026.² But the real inflection came in the six months leading into April 2026. According to its financial adviser Huafeng Capital, the company “raised $3.9 billion over the past six months,”² and then went back to the well again.

In its latest round, Moonshot “has raised about $2 billion at a valuation of $20 billion,” led by Meituan’s venture arm, Long‑Z Investment, with Tsinghua Capital, China Mobile and CPE Yuanfeng also joining.² For a Chinese AI lab still ostensibly playing in the open‑source end of the pool, that is a staggering number.

Unlike many frontier labs, Moonshot isn’t just selling a dream of future AGI; it has tangible traction. Demand for its open‑weight models is booming as users and enterprises “don’t mind a performance hit in exchange for cheap inference.”² The result: “Moonshot’s annual recurring revenue topped $200 million in April, driven by rapid growth in paid subscriptions and API usage.”²

Its latest model, Kimi K2.6, is already “the second-most used LLM on distribution platform, OpenRouter.”² That detail matters: this isn’t a purely Chinese domestic story. It signals that developers globally are willing to integrate Chinese open‑weight models into their stacks — even as Washington tries to wall off China’s access to high‑end chips.

§The open‑weight moment: from curiosity to asset class

Moonshot’s mega‑round is part of a broader surge: “investor appetite for open-weight AI models made by Chinese labs” is “skyrocketing.”² While Western investors agonize over closed vs. open model economics, Chinese capital is effectively placing a portfolio bet across the open‑weight field.

Several of Moonshot’s rivals have already cashed in on that appetite by going public. Zhipu AI, listed in Hong Kong as Knowledge Atlas Technology, “ended Thursday with a market cap of HK$434.7 billion (roughly $55.9 billion), while MiniMax ended the day at HK$257.3 billion ($33 bil…).”² Public‑market multiples like that both validate and inflate the valuations now being attached to DeepSeek and Moonshot.

The logic is straightforward: cheap, reasonably strong open‑weight models are a perfect fit for developers and enterprises who are cost‑sensitive or wary of lock‑in to U.S. closed models. Chinese labs, forced by chip constraints to optimize aggressively, are turning that necessity into a selling point.

§Beijing’s new doctrine: model capability as strategy

Behind the venture frenzy is a deeper strategic re‑write in Beijing.

For years, China’s answer to U.S. export controls was to double down on the “silicon side of the AI stack” — funding fabs, foundries and tooling to catch up on chips.³ U.S. policy was clear: use semiconductor controls “to limit China’s access to leading-edge AI compute,” and thus its ability to train frontier models.³

The Big Fund’s move into DeepSeek is, as one observer notes, “the recognition that the response strategy now runs through model capability rather than purely through chip capability.”³ If China “cannot acquire Nvidia’s leading-edge GPUs at the volume required, it will, on this evidence, finance the model labs that have demonstrated they can produce frontier results without them.”³

DeepSeek embodies that bet; Moonshot shows how it can be monetized. One is being anointed as a quasi–national champion with state backing; the other is proving that open‑weight models can spin real revenue and international usage.

§Competing narratives: bubble, catch‑up, or power shift?

Different actors are now reading these moves through very different lenses.

• Chinese state and industrial planners see DeepSeek’s “$45bn valuation” not primarily as a market froth but as “Beijing’s strategic statement” that AI models themselves are now critical infrastructure.³
• Investors (both private and state‑adjacent) are chasing the thesis that Chinese open‑weight labs can deliver “frontier results” on cheaper hardware³ and convert that into recurring revenue, as Moonshot’s $200 million ARR and OpenRouter dominance suggest.²
• Global developers and enterprises are treating Kimi‑class models pragmatically: “no shortage of interest from those who don’t mind a performance hit in exchange for cheap inference.”² In other words, utility over geopolitics — at least for now.
• Skeptics will argue that valuations of $20–45 billion for labs barely three years old, operating under severe hardware constraints, smell like a bubble inflated by national‑champion logic and hot money fleeing into whatever can plausibly rival OpenAI.

What’s new is not the hype but the alignment: state funds, tech giants, and public markets are all converging on the same story — that China’s path around U.S. controls runs through aggressive funding of model labs optimized for domestic hardware and open‑weight distribution.

§What happens next

In the span of a few weeks, DeepSeek has gone from self‑funded curiosity to potential state‑backed flagship, its “first round of venture capital” poised to value it at $45 billion.¹ Moonshot has turned a hit open‑weight model into a $20 billion business with serious recurring revenue and one of the most‑used LLMs on a major global platform.²

For Washington, this is an uncomfortable proof point: chip controls have not slowed Chinese AI so much as redirected it. For investors, it’s a signal that the center of gravity in open‑weight AI may be drifting east. And for everyone else, it’s a warning that the next wave of AI competition won’t just be about who has the best GPUs — it will be about who can do the most with less.

Story coverage