China's LLM Funding Frenzy: Who's Next to Be Eliminated?

In May 2026, a seemingly insane funding competition is unfolding in China's AI large language model (LLM) sector.

On May 7, Moonshot AI (Kimi) completed a funding round of approximately $2 billion, pushing its post-money valuation beyond $20 billion. The round was led by Meituan Longzhu, with participation from China Mobile and CPE.

On May 8, DeepSeek initiated its first external funding, planning to raise 50 billion yuan (about $7 billion). Founder Liang Wenfeng personally contributed 20 billion yuan to lead the round, with the National Integrated Circuit Industry Investment Fund (the "Big Fund") negotiating as the main investor, and Tencent expressing intent to invest 6 billion yuan. The post-money valuation skyrocketed from an initial $10 billion to $51.5 billion, jumping fivefold in just 21 days.

Simultaneously, Zhipu AI completed a new round worth tens of billions of yuan, reaching a pre-money valuation of 200 billion yuan. Baichuan Intelligent closed a Series A round of 5 billion yuan, with state capital from Beijing, Shanghai, and Shenzhen entering. StepFun raised over 5 billion yuan, with Tencent and Xiaomi participating... In just one month, the total funding from major companies alone exceeded 100 billion yuan.

Capital is flooding in, and valuations keep being updated. However, behind this frenzy, an unavoidable question arises: In this funding rush, who will be the next to fall?

1. A Torrent of Funds Like a "Market"

Let's revisit this funding feast in May. DeepSeek's round is the largest standalone round in Chinese AI history and highly unusual. For the past two years, the founder had told all investors that "VC money is a burden," yet suddenly opened the doors.

The investment structure is noteworthy. Liang Wenfeng personally contributed 20 billion yuan, accounting for 40% of the total raise. Tencent put in 6 billion yuan for a 2% stake, with the Big Fund leading. This is not just fundraising; it's a "repricing" of DeepSeek.

According to analysis, Liang Wenfeng internally stated, "We're not raising because we lack funds. To solidify the value of employee stock options, we need a clear valuation anchor." Over the past three years, DeepSeek's employee options were among the hardest-to-value assets in China's AI sector. Despite top-tier technical prowess, without funding history, published valuations, or an IPO schedule, the card of "technical ideals" no longer sufficed as competitors like ByteDance, Tencent, and Xiaomi offered definitive, million-yuan annual salaries to poach talent.

In the past six months, at least five core researchers have left DeepSeek. Luo Fuli, a key contributor to V3, joined Xiaomi; Guo Daya, a core researcher for R1, went to ByteDance; and Ruan Chong, a multimodal technology expert, moved to autonomous driving firm QCraft... With under 200 total employees and about 100 in core research, each departure directly severs a technical pipeline.

Thus, the 20 billion yuan personal investment wasn't mere fundraising, but a powerful signal: "This company has this much value, and the founder is all-in."

2. $20 Billion vs. $51.5 Billion: The 2.5x Gap in Valuation Logic

In this wave, the valuation difference between Kimi and DeepSeek is intriguing. Kimi's annual recurring revenue (ARR) has surpassed $200 million, with monthly active users (MAU) reaching 9 million—the numbers justifying its valuation to investors.

Meanwhile, DeepSeek's MAU is 127 million, 14 times Kimi's, but its API pricing is one-tenth of OpenAI's, with V4-Flash cache hits at 0.02 yuan per million tokens—world-lowest levels. Actual revenue remains undisclosed.

Logically, the one with revenue should be valued higher. Yet the result was the opposite: Kimi, with $200 million ARR, was valued at $20 billion, while DeepSeek, with undisclosed revenue, soared to $51.5 billion. The former earns through a "product," while the latter is priced on a "narrative."

This shift in valuation logic is telling. Kimi is a typical market-driven AI startup, with Alibaba as the largest shareholder at about 36% (partly paid via Alibaba Cloud compute credits), and Tencent following. Multiple internet giants on one cap table is now common, almost standard in AI.

DeepSeek is unique in being seen as a "national-level technical asset." After V4's release, it completed adaptation to Huawei's Ascend platform. NVIDIA CEO Jensen Huang remarked frankly that if DeepSeek releases first on Huawei's platform, it would be "devastating" for the U.S. When a model company is imbued with strategic significance for "compute self-sufficiency," its valuation logic transcends traditional price-to-earnings ratios.

However, there's always a "gap of time's validation" between valuation and true value.

3. Securing Funds Doesn't Guarantee Survival

LLMs are a classic business of "high investment, long cycles, uncertain returns." Securing funding merely grants a seat at the table; staying seated till the end is another matter.

Look at the numbers. Zhipu AI, already listed in Hong Kong, reported 2025 revenue of 724 million yuan against a loss of 4.7 billion yuan. Revenue grew 131%, but losses expanded, and gross margin fell from 56% to 41%. Yet its market cap exceeds 370 billion Hong Kong dollars.

"Revenue 700 million, loss 4.7 billion, market cap 370 billion." These three numbers reflect the reality of the current LLM industry. Capital pays for the "future," but that future must prove it's more than just a story.

Globally, OpenAI's cumulative funding is $122 billion, Anthropic's $30 billion. DeepSeek's 50 billion yuan ($7.35 billion) is record-breaking domestically but merely a fraction of OpenAI's. Its $51.5 billion valuation is about 6% of OpenAI's ($852 billion).

Global AI capital concentrates on a handful of top firms, with funding density rapidly declining elsewhere.

This isn't a positive signal. From the 1990s internet bubble to the 2015 O2O wave and 2018 blockchain craze, whenever capital shifts from "broad and shallow" to "top-concentrated," industry elimination accelerates.

4. Three Danger Signals

In VC, the saying goes: "In tailwinds, compete on speed; after the tide recedes, compete on underlying strength."

Following this logic, these three types of companies are most at risk:

First: Those without differentiated technical barriers. If model capabilities only "keep up with open-source levels," there's no irreplaceability in the industry. With Meta's Llama, Alibaba's Qwen, and Zhipu's GLM going open-source and continuously improving, moats built solely on "fine-tuning open-source models" are near-zero.

Second: Those lacking self-sustaining revenue and surviving on fundraising. Losses come in "strategic" and "consumptive" types—the former invests in the future, the latter buys time. Zhipu AI has 700 million yuan revenue against 4.7 billion yuan loss; Kimi has $200 million ARR but spends hundreds of billions on operational costs. Everyone is burning cash, but as long as the next round succeeds, they can continue; if it fails, it's immediate exit. When global AI capital contracts, companies with closed funding windows face immense survival pressure.

Third: Those with diluted founder control and unstable core teams. LLMs are a long game; team stability is a core competitive advantage. Liang Wenfeng spent 20 billion yuan personally to retain control, while MiniMax's Yang Zhilin designed an AB share structure for dual-class voting. Though methods differ, both aim to "not hand core decision-making power to capital." Conversely, teams where founder stakes have diluted below 10% through repeated funding, with core members continuously departing, will struggle to endure an industry winter.

5. Industry Endgame: 3-5 Survivors, Most Eliminated

A common view among AI practitioners is that "fewer than five companies will ultimately survive in China's LLM industry."

Those five likely include:

• National representatives: DeepSeek (compute self-sufficiency + Big Fund backing)
• Direct subsidiaries of internet giants: ByteDance (Doubao), Alibaba (Tongyi), Tencent (Hunyuan) (with massive traffic and capital)
• Listed platform companies: Zhipu AI, MiniMax (with access to secondary market funding)
• Domain-specific winners: Companies deep in sectors like healthcare, law, or finance, building industry barriers

Most others, though currently glittering, will undergo painful "bubble bursts." They'll be acquired or simply shut down.

History proves it. Of thousands of O2O startups in 2015, fewer than 10 survived. Most of the hundreds of blockchain firms from 2018 vanished, and 2021's community group-buying platforms consolidated to a few.

LLMs burn money faster than any of these industries, meaning elimination cycles will be equally rapid.

Final Thoughts

I once heard from a middle manager at an LLM company. On the day they completed a new funding round last year, there was no celebratory mood; the conference room was silent.

The CEO's first words were: "The money has arrived. But the real test begins now."

At the time, it sounded dramatic, but in retrospect, it might have been the most honest reaction.

The May 2026 funding rush marks a milestone for China's LLM industry. Yet it's not the finish line—it's the starting gun for the true battle.

Astonishing amounts of money, absurdly high valuations, intense competition. But everyone knows: securing a seat at the table now doesn't mean you'll leave with a smile.

When tailwinds blow, even pigs can fly, but when the wind stops, it's the pigs that crash.

The answer lies not in valuations, but in technology, business models, and the validation of time.