The $120 Billion Dual IPO Race: DeepSeek vs Moonshot

*Two IPOs, one quarter: DeepSeek is heading for Shanghai's STAR Market while Moonshot AI targets Hong Kong. Together they are worth more than $120 billion on paper. (Image: Unsplash)*

Two Companies, One Deadline

On the morning of October 6, 2026, Bloomberg published two stories within hours of each other. The first: DeepSeek, the Hangzhou-based lab that stunned Silicon Valley with R1 in January 2025, was closing a funding round of at least ¥80 billion — roughly $12 billion — that could approach ¥100 billion once every signed term sheet converts. The second: Moonshot AI, the Beijing startup behind the Kimi assistant and the 2.8-trillion-parameter Kimi K3, had completed its final private round at a valuation of about $50 billion and was preparing to list in Hong Kong.

The symmetry was hard to miss. Both rounds effectively priced during China's National Day holiday week. Both companies are now restructuring their corporate entities for public listings. And both have told bankers to target the same window: the first quarter of 2027.

One will list on the Shanghai Stock Exchange's STAR Market, the board Beijing built for hard-tech champions. The other will list on the Hong Kong Stock Exchange, the venue Chinese tech giants have used for two decades to reach global capital. One raised more money in a single round than almost any private company in Chinese history. The other raised money more times in a single year — eight rounds in ten months — than most startups do in a lifetime.

The question investors in Shanghai, Hong Kong, and New York are quietly debating is not whether these listings will happen. It is which one matters more — and what each says about how China's AI industry plans to pay for the next decade of compute.

The Reluctant Capitalist: DeepSeek's ¥100 Billion Turn

Two years ago, Liang Wenfeng told anyone who would listen that DeepSeek did not need outside money. The lab was incubated inside High-Flyer Quant, his hedge fund, and early compute was effectively subsidized by trading profits. "The problem isn't money" became the quote attached to every profile of the company — a badge of purity in an industry burning tens of billions of dollars a year.

That era is definitively over. According to Bloomberg's October 6 report, DeepSeek's second external round — barely four months after its first — has swollen from an initial target of ¥50 billion to at least ¥80 billion committed, with signed term sheets pointing to a final tally near ¥100 billion ($14–15 billion). Tencent and CATL, the battery giant that has spent two months building an energy-and-compute investment portfolio, are the largest committed investors.

The sequence deserves a closer look, because the round's growth is the story. The Information reported on September 24 that DeepSeek had closed what it called the B round at roughly $7 billion and a $75 billion valuation. Twelve days later, Bloomberg's figures were $12 billion and counting, with the same ~$75 billion valuation band. The most plausible reading, confirmed by the term-sheet language in Bloomberg's reporting, is that the round's first close filled quickly — and then investor demand kept arriving. In a market where Chinese venture fundraising has contracted for three straight years, one AI lab absorbed something close to a full year of national VC deployment.

The money is not going to marketing. DeepSeek's financials, disclosed in the course of fundraising, show the classic frontier-lab profile at Chinese scale: ¥475 million in revenue for the first seven months of 2026 against a ¥715 million net loss. Revenue comes almost entirely from API calls and enterprise services. What the balance sheet does not show is the asset DeepSeek is actually building: a 160,000-accelerator data center in Inner Mongolia running on Huawei Ascend chips, which would rank among the largest domestic-chip clusters ever assembled. Days before the funding news landed, DeepSeek open-sourced its entire Ascend-facing infrastructure stack — compiler, compute libraries, and distributed communication tools — signaling that the Inner Mongolia build is a template, not a one-off.

Liang Wenfeng's personal stake tells its own story. His net worth roughly doubled to about $36 billion after the June round, and he contributed again to the October round — unusual for a founder in any market, and a signal that control matters to him even as the shareholder list lengthens. He has told investors, reportedly in at least one meeting this year, that open-source model development remains non-negotiable even as he pursues artificial general intelligence. The STAR Market listing, with CITIC Securities advising, would make DeepSeek the most valuable pure-play AI modeler ever to go public in China — and test whether public-market investors will underwrite a company whose founder has explicitly deprioritized monetization.

The Serial Fundraiser: Moonshot's Eight Rounds to $50 Billion

If DeepSeek's capital story is about restraint finally yielding to scale, Moonshot's is about momentum engineered with near-mechanical precision. Founded in April 2023, the company launched the Kimi assistant and closed a $200 million angel round within two months. Since then it has raised continuously — and 2026 turned the tempo up to something without precedent in Chinese venture capital.

Between January and July 2026, Moonshot completed six distinct rounds. January and February brought three back-to-back closings — $500 million, $700 million, $700 million — lifting the valuation from $10 billion to $18 billion with Alibaba, Tencent, and Wuyuan Capital writing repeat checks. In May, Meituan's Long-Z fund led a $2 billion D round that pushed the post-money valuation past $20 billion and the cumulative total above ¥37.6 billion. June's E round priced at $31.5 billion pre-money. July's F round brought in more than $3.5 billion led by a national-level AI industry fund, landing at $35 billion.

Then, on October 6, came the finale: a final pre-IPO round at roughly $50 billion — a $15 billion step-up in under three months.

What investors are buying is easier to see on an income statement than DeepSeek's pitch. Moonshot's annualized revenue run-rate has reached $1 billion — roughly double DeepSeek's annualized pace — and is expected to hit $2 billion by December, driven by Kimi's consumer subscription and enterprise API businesses. The July launch of Kimi K3, a 2.8-trillion-parameter open-weights model that third-party evaluators rank alongside Google's Gemini 3.8 Flash and xAI's Grok 4.7 on intelligence benchmarks, converted that revenue into a global growth story.

The IPO machinery is already moving. Bank of America is lead coordinator, with CICC, Deutsche Bank, and Goldman Sachs as joint underwriters. Early-look investor meetings could begin this month, and the company is exploring raising up to $5 billion — a figure that would make it one of the largest Hong Kong tech listings in years. Behind the scenes, Moonshot is negotiating revenue-share agreements that would let Microsoft, Amazon, and Google host Kimi K3 on their clouds — a striking inversion in which American hyperscalers could become distribution channels for a Chinese frontier model.

Where DeepSeek is vertically integrating into silicon and electrons, Moonshot is horizontally integrating into every cloud and every screen.

Head-to-Head: Two Playbooks, One Market

The valuation gap — roughly $20–25 billion — is smaller than it looks once you account for what each number represents. DeepSeek's ¥500 billion is a private-market consensus reached during an oversubscribed round with industrial investors who want strategic access to models and compute. Moonshot's $50 billion is the price of the last shares sold before a public listing, underwritten by bankers who must defend it to global institutional investors within six months.

A more revealing comparison is capital efficiency per dollar of revenue. Moonshot is valued at roughly 25 times its projected December run-rate — aggressive but legible to anyone who has priced a software company. DeepSeek is valued at several hundred times current revenue, which means STAR Market investors are being asked to price something closer to a national research asset: the lab that cracked efficiency frontiers twice, open-sourced its way into global developer loyalty, and is now the single largest buyer of Huawei's Ascend silicon.

Their investor lists encode the divergence. DeepSeek's two biggest backers are an internet conglomerate that wants to embed its models across social, gaming, and cloud products, and a battery manufacturer building a compute-energy components business. Moonshot's cap table reads like a sovereign-capital reunion: a national AI industry fund, Alibaba, Tencent, Meituan, China Mobile — plus a syndicate of international backers watching the HKEX window.

*Hong Kong's HKEX gives Moonshot a path to global institutional money; Shanghai's STAR Market gives DeepSeek a domestic investor base aligned with the national tech agenda. (Image: Unsplash)*

Why Two Exchanges Matter More Than Two Tickers

The listing venues are not a logistics detail. They are the strategic payload.

Shanghai's STAR Market was created in 2019 to finance hard technologies Beijing considers strategic — semiconductors, aerospace, biomedicine. A DeepSeek listing there would be the board's crowning validation: a frontier AI lab, long self-funded and famously skeptical of capital markets, choosing domestic public investors to finance its trillion-parameter ambitions. STAR Market's retail-heavy investor base and state-adjacent funds are far more likely to underwrite a loss-making lab whose value is tied to national capability than any foreign institution would be.

Hong Kong is the opposite instrument. HKEX has spent two decades serving as the bridge between Chinese growth stories and global capital — Tencent, Alibaba, Meituan all passed through it. Moonshot's choice of venue, reportedly with Bank of America steering the process, signals ambition to be priced as a global AI company, not a domestic champion. Hosting Kimi K3 on American clouds would reinforce that framing: revenue in dollars, distribution everywhere, geopolitics managed rather than avoided.

Dual-tracking through two exchanges also de-risks the political economy. If US-China tension tightens around Hong Kong capital flows, DeepSeek's Shanghai listing proceeds regardless. If domestic liquidity tightens, Moonshot's Hong Kong book is global. Beijing gets a domestic flagship; Chinese AI gets a global one.

The Global Context: Public Money Versus Private Burn

Zoom out and the dual race becomes a structural statement. America's frontier labs are still burning private cash at unprecedented rates — OpenAI and Anthropic together are expected to consume over $100 billion in compute commitments next year, financed by sovereign funds, mega-cap partners, and debt. Their IPO timelines keep slipping. China's two most valuable model companies, by contrast, are racing to *public* markets while the window is open — and doing it with a combined $30-plus billion of fresh private capital already in the bank.

The backdrop makes the timing rational. China's generative AI user base crossed 700 million in the first half of 2026 — a penetration rate above 50%, meaning one in two Chinese citizens now uses generative AI. Intelligent compute capacity hit 2,185 EFLOPS, up 177% year-on-year, and the first fully domestic 100,000-card supercluster entered service. All six of the most-called large models on global leaderboards are now Chinese-built. When a market's adoption curve looks like that, public investors will pay for exposure — and both companies are moving before the pricing window narrows.

The risks are equally symmetrical. DeepSeek must prove that open-source economics can sustain a ¥500 billion valuation when the moat is reputation and the competition — Zhipu's GLM series, StepFun's Step models, Alibaba's Qwen — gives the software away for free.

*Roughly half of DeepSeek's raise is earmarked for compute infrastructure, anchored by a 160,000-chip Huawei Ascend cluster in Inner Mongolia. (Image: Unsplash)*

Moonshot must convert $1 billion of ARR into margins while funding 2.8-trillion-parameter training runs and sharing revenue with the very clouds it depends on. Both are betting that being first to list is itself a moat: the company that sets the public-market reference price for Chinese AI will shape every valuation, acquisition, and talent package that follows.

Liang Wenfeng, the quant who insisted money was not the problem, is now sitting on a company that raised roughly $20 billion in five months. Whether the problem was money or not, the solution clearly is — and in Q1 2027, the world gets to buy a piece of both answers.

What Chinese AI Circles Are Saying

Zhihu — @投资人老周 (investment professional, 48K followers):

"两年前说'问题不是钱'，两年后单轮融资近千亿。DeepSeek的融资史就是中国AI的估值锚重置史。但注意一个细节：梁文锋本人继续出资。这在任何市场都不常见——控制权比估值重要。"

>

*"Two years ago he said 'money isn't the problem'; now he's raising nearly ¥100 billion in one round. DeepSeek's funding history is the re-pricing of China's AI anchor. But note the detail: Liang Wenfeng personally invested again. That's rare in any market — control matters more than valuation."*

X (Twitter) — @chinaaimarkets:

"Moonshot going to HKEX, DeepSeek to STAR. One wants global capital, one wants national strategic pricing. Together they're worth $121B. For context, that's more than the combined market cap of every US AI startup that's actually IPO'd."

>

*Direct post; bilingual account covering Chinese AI capital markets.*

Weibo — @科技财经小辣椒 (1.2M followers):

"月之暗面十个月八轮融资，阿里腾讯美团国家队全部到场，这不是融资是抢筹。K3确实能打，但2万亿参数的训练账怎么算？IPO募50亿美元也就够烧两年。"

>

*"Eight rounds in ten months for Moonshot — Alibaba, Tencent, Meituan, the national fund, all piling in. This isn't fundraising, it's a scramble for shares. K3 can really fight, but who pays the 2.8-trillion-parameter training bill? A $5B IPO raise only covers about two years of burn."*

Xiaohongshu — @AI产品观察员：

"别只盯着估值看。DeepSeek开源了昇腾全套基础设施，月之暗面在跟微软亚马逊谈云托管——一个往底层钻，一个往全球铺。路线不同，但都指向同一件事：中国模型公司开始掌握自己的命运了。"

>

*"Stop staring at the valuations. DeepSeek open-sourced its whole Ascend stack while Moonshot negotiates cloud hosting with Microsoft and Amazon — one drilling down, one spreading global. Different routes, same destination: Chinese model companies are starting to control their own destiny."*

Douban — @量化交易员小陈：

"幻方出来的团队最懂的就是风险和杠杆。梁文锋用对冲基金的方式做AI公司：自己出钱占大头，外面的是流动性不是控制。科创板散户多，这个结构上市后才算真正考验。"

>

*"A team from a quant fund understands risk and leverage best. Liang is running an AI company like a hedge fund: his own money keeps the controlling stake; outsiders are liquidity, not control. STAR Market is retail-heavy — the real test comes after listing."*

GitHub — @open-model-dev (comment on DeepSeek Ascend toolchain repo):

"Everyone's debating valuations and nobody's reading the TileLang commit history. DeepSeek just made 160k Ascend chips programmable with the same ergonomics as CUDA. That changes the open-source hardware game more than any IPO filing."

>

*Posted as a discussion comment on the open-sourced Ascend infrastructure components; representative of global developer reaction.*

Related Coverage:

- China's AI IPO Gold Rush: DeepSeek, Moonshot, and the New Capital Markets

- Moonshot AI's $50 Billion Blueprint: Sovereign Capital Meets Kimi K3

- DeepSeek's Silence Before V4-Pro: What China's Quietest AI Lab Is Really Doing

- DeepSeek V4.1-Flash and the Open-Weights Disruption

- The 3% Gap: Bloomberg's October Reckoning on US-China AI