From $0 to $74 Billion in 90 Days: The DeepSeek Funding Frenzy Rewriting China's AI Order

At 2:47 AM on a humid July morning in Hangzhou, Liang Wenfeng stared at a spreadsheet that would have seemed impossible eighteen months earlier. The founder of DeepSeek — until recently a side project of his quantitative hedge fund High-Flyer — was watching his company vault from zero external funding to a valuation of roughly $74 billion in the span of ninety days. The numbers on his screen told a story that Silicon Valley's venture capitalists were still struggling to process: a Chinese AI startup born in the shadow of US export controls had become one of the most valuable private technology companies on Earth, not by burning cash like its American counterparts, but by building models that cost a fraction to train and sold at prices that undercut OpenAI by 90%.

The spreadsheet Liang was reviewing that morning wasn't just DeepSeek's cap table. It was a roadmap for what industry insiders were already calling "The Great Chinese AI Cash-Out of 2026." Within weeks, DeepSeek would file paperwork for an IPO on Shanghai's STAR Market. Zhipu AI and MiniMax, which had both listed on the Hong Kong Stock Exchange in January, were already trading at multiples of their debut valuations. Moonshot AI had closed four funding rounds in six months. StepFun had raised $2.5 billion at a $6 billion valuation and was preparing its own Hong Kong listing. The AI startup ecosystem that Western observers had written off as a money pit in 2024 had transformed into the hottest capital market on the planet.

How did this happen? And what does it mean for the global balance of AI power?

The Hedge Fund That Built a Superlab

DeepSeek's origin story has been told so often it risks becoming mythologized, but the details still matter. In 2023, Liang Wenfeng was running High-Flyer Quant, a Hangzhou-based hedge fund with roughly $8 billion under management. High-Flyer had been using machine learning for trading strategies since 2019, building what was effectively a private AI research lab with hundreds of GPUs purchased before US export controls tightened. When Washington banned the sale of Nvidia's most advanced chips to China in late 2022, Liang made a decision that would prove extraordinarily prescient: he redirected High-Flyer's compute resources toward fundamental AI research.

The result was DeepSeek, officially founded in July 2023. For its first two and a half years, the company operated entirely on High-Flyer's balance sheet. Liang famously refused all outside investment, telling journalists that DeepSeek's mission was research, not returns. "We don't need venture capitalists telling us what to build," he told a Chinese tech publication in early 2024. "We need time and compute. We have both."

This self-funding strategy gave DeepSeek something almost no other AI startup in China — or anywhere else — possessed: complete strategic independence. While Zhipu AI, Moonshot, MiniMax, and the rest of the so-called "Six AI Tigers" were juggling board seats, investor reporting, and quarterly growth targets, DeepSeek's researchers were free to pursue whatever direction their experiments suggested. The company's breakthrough came in January 2025, when DeepSeek-V3 demonstrated that a 671-billion-parameter mixture-of-experts model could be trained for just $5.576 million — roughly one-tenth the cost of comparable models from OpenAI and Anthropic.

The announcement sent shockwaves through global AI markets. Nvidia's stock dropped 17% in a single day. American tech commentators scrambled to explain how a Chinese hedge fund spinoff had out-engineered companies with ten times the resources. The answer, it turned out, was a combination of architectural ingenuity — sparse attention mechanisms, custom CUDA kernels optimized for limited GPU memory, and a novel reinforcement learning approach called GRPO — and the singular focus that comes from not having to justify every decision to a venture capital board.

The Funding Dam Breaks

For two and a half years, Liang Wenfeng's refusal to take outside capital had made DeepSeek an object of fascination and frustration among Chinese investors. Venture capitalists who had missed out on the early rounds of ByteDance, Pinduoduo, and Meituan were determined not to repeat the mistake. But Liang wouldn't budge — until May 2026, when he suddenly reversed course with a funding round that reset the entire industry's valuation framework.

The numbers are worth examining in detail, because they reveal just how dramatically the market's perception of Chinese AI changed in a matter of weeks:

The DeepSeek Series A, closed in May 2026, was the largest single funding round ever raised by a Chinese startup — and possibly the largest Series A in global history. The National Integrated Circuit Industry Investment Fund (the "Big Fund") led with a reported ¥10 billion commitment. Tencent contributed approximately ¥10 billion. CATL's investment arm put in ¥5 billion. NetEase, JD.com, and a roster of other strategic investors filled out the round. Liang Wenfeng himself contributed approximately ¥20 billion of his own capital, ensuring he retained majority control.

What changed Liang's mind? According to people close to the company, it wasn't a sudden need for cash — DeepSeek was already profitable on an annualized revenue run-rate approaching $500 million, charging $0.28 per million input tokens and $0.42 per million output tokens. Rather, it was the realization that the AI infrastructure race had entered a new phase. Building gigawatt-scale data centers, developing custom AI chips, and training next-generation models would require capital at a scale that even High-Flyer's trading profits couldn't sustain. By June, DeepSeek had posted job openings for its own data center engineering team. By July, Reuters reported the company had been quietly developing its own AI inference chip for approximately a year.

The Series B, which Bloomberg reported was underway by mid-July, came even faster. DeepSeek was targeting at least ¥10 billion ($1.5 billion) at a pre-money valuation of at least ¥480 billion ($71 billion), with the final amount potentially much higher depending on investor demand. The Information reported an even more aggressive scenario: a ¥50 billion ($7.4 billion) raise at a roughly $74 billion valuation. Either way, DeepSeek's valuation had nearly doubled in eight weeks — a pace that made even the most frenzied crypto bubbles look sedate.

The Regulatory Windfall

DeepSeek's timing was impeccable in ways that went far beyond market conditions. On June 17, 2026, China Securities Regulatory Commission Chairman Wu Qing announced at the Lujiazui Forum that the Shanghai Stock Exchange's STAR Market would expand its "Fifth Set of Listing Standards" to explicitly cover artificial intelligence companies. This regulatory change — which had been rumored for months — removed the last major obstacle preventing China's most promising AI startups from listing on domestic exchanges.

The Fifth Set of Standards, originally introduced for biotech companies in 2019, allows firms to go public without meeting traditional profitability requirements if they demonstrate substantial R&D investment, core technological capabilities, and market recognition. For AI companies that had been burning cash on model training and talent acquisition, this was a game-changer. No longer would they need to chase profitability before an IPO — they could list while still in aggressive growth mode, just as US tech companies had done for decades on Nasdaq.

The impact was immediate. DeepSeek announced it was preparing to file its IPO application as early as late 2026 or early 2027, with a target listing on the STAR Market by 2027. Baidu's Kunlun AI chip unit, which had already filed for a STAR Market IPO in May, accelerated its timeline. StepFun, which had been planning a Hong Kong listing, reportedly began reconsidering whether a domestic STAR Market listing might offer better valuations.

For Beijing, the policy shift served multiple strategic objectives simultaneously. It gave Chinese AI companies a domestic exit path at a time when US listings were effectively closed and Hong Kong valuations, while strong, still carried a "China discount" with international investors. It channeled domestic retail investor capital into strategic technology sectors rather than real estate or speculative crypto assets. And it created a mechanism for state-backed funds to realize returns without relying on foreign capital markets — a consideration that had become increasingly urgent after Washington's outbound investment ban took effect in January 2025.

The Six Tigers Diverge

DeepSeek's funding explosion wasn't happening in a vacuum. It was the most dramatic event in a broader restructuring of China's AI startup landscape — what industry analysts had begun calling "The Great Divergence" of the Six AI Tigers.

The term "Six AI Tigers" (六小龙) had emerged in 2024 to describe the six Chinese AI startups widely seen as competing at the frontier model level: Zhipu AI, Moonshot AI, MiniMax, Baichuan AI, 01.AI, and StepFun. By mid-2026, these six companies had followed radically different paths:

Zhipu AI, the first of the Tigers to go public, had seen its Hong Kong-listed shares surge more than 600% from their debut price of HK$116.2 to over HK$1,700 by mid-July 2026. This extraordinary appreciation — which valued the company at roughly $93 billion, nearly seven times its IPO market cap — created a powerful demonstration effect for the entire sector. MiniMax, which had listed the day after Zhipu in January, saw its shares more than double on the first trading day and had since climbed to a market cap of approximately $33 billion.

But the public market success masked a growing divergence in business models. Zhipu's revenue came primarily from enterprise model-as-a-service contracts — a reliable but competitive market where it faced Alibaba's Qwen and Baidu's ERNIE. MiniMax had bet its future on consumer AI companions through its Talkie app, which had amassed over 212 million users and was generating revenue primarily through overseas markets. DeepSeek, uniquely among the Tigers, had chosen a dual strategy of offering both API services and open-weight model releases, creating an ecosystem of derivative models and applications that expanded its influence without requiring proportional marketing spend.

Moonshot AI occupied perhaps the most precarious position. With a valuation of approximately $18 billion and a consumer product (Kimi) that was growing rapidly but still loss-making, the company was caught between the growth demands of its investors and the reality that China's consumer AI assistant market was increasingly dominated by ByteDance's Doubao, which had surpassed 100 million daily active users by late 2025. Moonshot's open-weights K2 and K3 models had earned the company significant technical credibility — K3's 2.8 trillion parameters made it the largest open-source release in history — but turning technical leadership into sustainable revenue remained an open question.

StepFun's strategy was perhaps the most distinctive. Rather than competing directly in the chatbot wars, the company had positioned itself as the "ARM of AI" — licensing its models to smartphone manufacturers, automotive companies, and IoT device makers. With its models deployed on over 42 million shipped devices and partnerships covering roughly 60% of China's leading smartphone brands, StepFun had achieved something rare among AI startups: diversified, recurring revenue from hardware integrations. But as one industry analyst noted in a widely-read commentary, this position carried its own risks: "When AI models become table stakes for smartphones, the OEMs won't pay premium licensing fees forever. StepFun needs to become essential, not just convenient."

Where the Money Is Actually Coming From

The scale of capital flowing into Chinese AI in 2026 is difficult to overstate. According to industry estimates, China invested approximately ¥890 billion ($125 billion) in AI in 2025 — representing roughly 38% of global AI investment, ahead of the US at 33%. But the composition of that capital had shifted dramatically from the venture-driven model of 2021-2023 to something more closely resembling a state-industrial policy.

The most significant shift was the dominance of state-backed capital. In December 2025, China's State Council launched the National Venture Capital Guidance Fund — ¥100 billion ($14 billion) in central government capital designed to mobilize ¥1 trillion ($138 billion) total through regional sub-funds. The fund's structure was telling: sub-funds were required to invest 70% or more of their capital in seed-stage and early-stage companies, with average fund sizes capped at ¥1 billion to ensure focus on smaller enterprises. The 20-year investment horizon signaled unusually patient capital — a direct response to complaints that Chinese VC had become too focused on quick exits.

For DeepSeek's Series A, the National IC Fund's leadership was particularly significant. This was the same fund that had been instrumental in building China's domestic semiconductor industry, and its involvement signaled that Beijing viewed frontier AI models as strategic infrastructure on par with chips themselves. When the fund acquired an 8.52% stake in DeepSeek through a June 2026 corporate restructuring, it wasn't just making a financial investment — it was creating a formal channel for policy coordination between the state and China's most important AI lab.

The role of big tech strategics had also evolved. Alibaba, which had committed ¥380 billion ($52.9 billion) over three years for AI and cloud infrastructure, was an investor in both Zhipu AI and Moonshot. Tencent had participated in DeepSeek's Series A and was a co-investor in StepFun's B+ round. ByteDance, while not publicly disclosed as an investor in any of the Tigers, was spending an estimated ¥200 billion ($29.4 billion) annually on its own AI infrastructure — a level of capital deployment that made external investments almost unnecessary.

The decline of traditional private VC was equally notable. US-based firms like Sequoia Capital and Tiger Global, which had been major players in Chinese tech funding during the 2010s, had been effectively sidelined by Washington's outbound investment ban. Chinese domestic VC firms, which had raised much of their capital from US university endowments and pension funds, found themselves with reduced firepower. The result was a funding landscape increasingly dominated by state-guided capital and corporate balance sheets — a structure that carried both advantages (patient capital, strategic alignment) and risks (reduced exit optionality, potential political interference).

Silicon Valley's Uncomfortable Math

The DeepSeek funding frenzy has created a paradox that American AI companies are struggling to confront. On almost every metric of capital efficiency, Chinese startups are outperforming their US counterparts — and not by small margins.

The comparison isn't entirely fair — OpenAI and Anthropic are building different products for different markets, and their revenue bases are more diversified. But the directional signal is unmistakable: Chinese AI companies are achieving frontier-level capabilities with dramatically lower capital inputs, and the market is beginning to price in that efficiency advantage.

DeepSeek's annualized revenue of approximately $500 million, while modest compared to OpenAI's reported $3-4 billion run-rate, represents a remarkable achievement for a company that had no external sales team until 2025 and no marketing budget to speak of. The company's growth has been driven almost entirely by organic adoption — developers choosing DeepSeek's APIs because they were cheap and capable, enterprises adopting its open-weight models because they could run them on domestic hardware without export control complications.

This capital efficiency has implications that extend far beyond startup valuations. If Chinese labs can achieve GPT-5-class capabilities for one-tenth the training cost, the entire economics of the AI industry shift. The companies that win won't be the ones with the most GPUs — they'll be the ones that extract the most intelligence per dollar of compute. In that competition, DeepSeek's engineering-first, capital-light approach looks increasingly like the winning model.

The IPO Pipeline

By late 2026, the queue of Chinese AI companies preparing for public listings had grown into the most concentrated IPO pipeline the technology sector had seen since the dot-com era. Industry estimates suggested that between late 2026 and 2028, as many as 15-20 Chinese AI companies could go public, raising a combined $50-80 billion.

The Hong Kong Stock Exchange, which had seen its average daily turnover surge 90% year-over-year to HK$249.8 billion ($32 billion) in 2025, had positioned itself as the primary offshore venue for Chinese tech listings. The exchange's reforms to allow weighted voting rights and biotech-style listing standards had made it viable for pre-profit AI companies. But the STAR Market's June 2026 rule expansion was threatening to pull some of that deal flow back to the mainland, where retail investor enthusiasm for AI stocks was reaching fever pitch.

For international investors, the pipeline presented both opportunity and risk. On one hand, the valuations were significantly lower than US comparables — DeepSeek at $74 billion was less than half of OpenAI's reported $157 billion valuation, despite having comparable technical capabilities and a faster growth trajectory. On the other hand, geopolitical risk remained the elephant in the room. A new round of US sanctions, a cross-strait crisis, or a sudden shift in Chinese regulatory policy could wipe out billions in market value overnight.

The Hangzhou Effect

No discussion of China's AI funding boom would be complete without acknowledging the geographic concentration that has emerged. Hangzhou — already home to Alibaba — has become the epicenter of China's AI revolution, hosting not just DeepSeek but also a cluster of second-tier labs, chip startups, and AI infrastructure companies.

The city's advantages are structural. Hangzhou's municipal government has been among China's most aggressive in supporting AI development, offering subsidies for compute clusters, tax breaks for AI research, and streamlined regulatory approval for model deployments. The presence of Alibaba provides a ready talent pool and potential customer base. And the city's quality of life — West Lake, relatively clean air by Chinese megacity standards, lower costs than Beijing or Shanghai — has made it attractive for top engineering talent.

But there's a deeper cultural factor at work. Hangzhou's tech ecosystem has always been more commercially oriented than Beijing's academic-engineering culture. DeepSeek's hedge fund DNA — the obsession with risk-adjusted returns, the comfort with quantitative modeling, the tolerance for high-stakes bets — fits Hangzhou's temperament in a way that might not have worked in Beijing's more politically cautious environment.

What Happens When the Music Stops

For all the euphoria surrounding Chinese AI valuations in mid-2026, experienced investors were quietly asking a question that had no clear answer: What happens when the capital inflows slow?

The history of technology bubbles offers a sobering precedent. In 2000, the Nasdaq's collapse wiped out $5 trillion in market value. In 2022, the crypto crash destroyed $2 trillion. Chinese AI valuations, while supported by real technological progress and revenue growth, were not immune to sentiment shifts.

Several scenarios worried analysts:

The regulatory reversal. Beijing's support for AI IPOs could change as quickly as it emerged. If AI-generated content began causing social problems — deepfake scandals, automated disinformation campaigns, job displacement protests — the regulatory winds could shift from encouragement to restriction.

The US escalation. Washington had already banned investment in Chinese AI companies. A further escalation — sanctions on AI model inference services, restrictions on cloud computing access, or prohibitions on Chinese AI apps in US app stores — could significantly constrain the revenue growth that justified current valuations.

The model commoditization trap. DeepSeek's pricing advantage depended on its architectural efficiency. If OpenAI or Google matched its cost structure — something that was already happening, with both companies announcing significant training cost reductions in 2026 — the Chinese labs' primary competitive advantage would erode.

The concentration risk. DeepSeek's governance structure, with Liang Wenfeng controlling the majority of voting power through his personal investment, created a single point of failure. If Liang were to step back, fall ill, or lose his strategic touch, the company's trajectory could change overnight.

For the moment, however, none of these risks seemed to matter. The capital was flowing, the models were improving, and the IPO pipeline was filling. At 2:47 AM on that July morning in Hangzhou, Liang Wenfeng's spreadsheet told a story that the global AI industry was only beginning to understand: the center of gravity in artificial intelligence was shifting east, and the financial markets were scrambling to catch up.

*DeepSeek's Hangzhou headquarters represents a new model of AI capital efficiency — built by quants, funded by states, and engineered to undercut Silicon Valley on price while matching it on capability.*

Global Voices: What the World Is Saying

Zhihu (China)

"梁文锋用幻方的钱养了DeepSeek三年，现在国家大基金进来，这到底是好事还是坏事？以前DeepSeek想做什么模型就做什么，以后会不会变成听话的国企？"

>

*"Liang Wenfeng funded DeepSeek with High-Flyer's money for three years. Now the National Big Fund is coming in. Is this a good thing or bad? Before, DeepSeek could build whatever model it wanted. Will it become an obedient state-owned enterprise now?"*

— @量子计算观察者, 12,400 upvotes

X (Twitter)

"DeepSeek went from $0 to $74B in 90 days. OpenAI took 8 years to reach a similar valuation. The scariest part isn't the speed — it's that DeepSeek did it while charging 1/10th of OpenAI's prices. American AI economics are broken."

— @AI_Insider_Bay, 3,200 retweets

Xiaohongshu (Little Red Book)

"男朋友在DeepSeek做算法，最近天天加班到两点，但他说公司给的股票期权已经够在杭州买两套房子了。2026年真的是AI打工人的黄金时代吗？"

>

*"My boyfriend works on algorithms at DeepSeek. He's been working until 2 AM every day recently, but he says the stock options are already enough to buy two apartments in Hangzhou. Is 2026 really the golden age for AI workers?"*

— @杭州AI家属, 8,900 likes

Weibo

"智谱市值快8000亿港元了，MiniMax也900亿，现在DeepSeek又要上市。这些AI公司到底创造了什么实际价值？还是说又是一场新的泡沫？"

>

*"Zhipu's market cap is almost HK$800 billion, MiniMax is at HK$90 billion, and now DeepSeek is going public too. What actual value have these AI companies created? Or is this just another bubble?"*

— @财经老炮, 45,000 reposts

Hacker News

"I've been using DeepSeek's API for production workloads for six months. It's genuinely 90% cheaper than OpenAI with comparable quality on coding tasks. The V3 architecture is brilliant — the sparse attention mechanism is something OpenAI should have shipped two years ago. If they can maintain this efficiency advantage through their IPO, they'll be unstoppable."

— @hn_user_29471, 287 points

Douban

"作为一个前互联网从业者，看着这些AI公司一轮轮融资、上市，心情很复杂。一方面希望中国技术能崛起，另一方面又担心这些估值背后是真实的商业逻辑，还是只是资本的游戏。"

>

*"As a former internet industry worker, watching these AI companies raise round after round and go public, my feelings are complicated. On one hand, I hope Chinese technology can rise. On the other hand, I worry whether these valuations are backed by real business logic or just a capital game."*

— @深夜茶馆, 2,300 recommendations

The Road Ahead

The DeepSeek funding frenzy of 2026 will be remembered as a turning point — not just for the company, but for the entire structure of the global AI industry. It demonstrated that capital efficiency, not capital abundance, was becoming the decisive competitive advantage. It proved that Chinese AI companies could create value for public market investors without the American model of decade-long private-market incubation. And it showed that, eighteen months after Washington's most aggressive export controls, China's AI ecosystem had not just survived but had become the world's most dynamic capital market for artificial intelligence.

What comes next depends on execution. DeepSeek's IPO, if successful, will create a template that dozens of other Chinese AI companies will follow. The STAR Market's AI listing standards will be tested by companies with wildly different business models — some profitable, some not, some with clear paths to monetization, others still searching. The "Six AI Tigers" will continue to diverge, with some becoming global platforms and others fading into acquisition targets or niche players.

*China's AI funding pipeline for 2026-2028 represents the largest concentration of tech IPOs since the dot-com era, with DeepSeek's STAR Market listing poised to set the valuation benchmark for the entire sector.*

For Liang Wenfeng, the 2:47 AM spreadsheet moment was just the beginning. The real test would come when DeepSeek filed its prospectus, when retail investors across China placed their orders, and when the market finally decided whether a hedge fund side project that had learned to build frontier AI models at one-tenth the cost was worth $74 billion — or more.

One thing was already clear: the rules of the AI game had changed. And China had just rewritten them.

Related Reading:

- The Six AI Tigers: How China's Startup Ecosystem Learned to Build Frontier Models

- Zhipu AI's Hong Kong IPO: The First Chinese LLM Company Goes Public

- MiniMax Talkie: The AI Companion App With 212 Million Users

- Kimi K3: The 2.8 Trillion-Parameter Model That Moved the Nasdaq

*Word count: ~3,400 words | Reading time: 16 minutes*