DeepSeek Raises About $15 Billion in One of China’s Biggest-Ever Startup Rounds, Bloomberg Reports

DeepSeek, the Chinese artificial intelligence company whose models have become the most credible challenger to America’s leading AI labs, has raised about 100 billion yuan — roughly $15 billion — in its latest funding round, according to Bloomberg News, in what would rank among the largest private financings in Chinese corporate history.

The report, published on Thursday and citing people familiar with the matter, says the Hangzhou-based firm more than doubled its original fundraising ambition during negotiations, lifting its target at least twice — first from 50 billion yuan to 80 billion yuan, then to the final figure. Reuters carried the Bloomberg report as well, while noting that it could not independently verify it.

The money arrives at a pivotal moment for DeepSeek. The company is racing to convert its open-weight models into durable commercial advantage, stockpiling computing power and restructuring internally ahead of a possible stock-market listing. The scale of the raise puts China’s best-known AI start-up within sight of the funding firepower of its American rivals: OpenAI is currently raising $30 billion at a reported $1.4 trillion valuation, and Anthropic’s draft S-1 filing disclosed a first-half loss of $80.7 billion alongside a staggering $518 billion compute commitment, according to Reuters.

Industrial capital leads the round

What distinguishes this round is not just its size but its composition. The investor roster reads less like a venture-capital convention and more like a roll call of Chinese industrial power. Contemporary Amperex Technology Co. Ltd. — better known as CATL, the world’s largest electric-vehicle battery maker — and Tencent Holdings, the WeChat operator, were among the biggest investors, according to the Bloomberg report.

Chinese financial media outlet Caijing reported earlier this month, via multiple sources familiar with the matter, that CATL plans to invest an additional roughly 10 billion yuan in this round, exceeding the approximately 5 billion yuan it committed in the previous one. Tencent and IDG Capital are also participating in the current transaction, the Caijing report said. New investors named include Stony Creek Capital, China Fortune-Tech Capital, CITIC-affiliated CPE, Legend Capital and Geely Group — all described as capital sources with close ties to manufacturing, semiconductors or government-backed funds.

According to the Caijing report, new and existing shareholders are each expected to account for roughly half of the total investment. The pre-money valuation remains unchanged at approximately 500 billion yuan — about $75 billion — even as the round itself doubled from its original plan, which investment-banking sources attributed to pent-up demand from investors who missed allocation in the previous round and to DeepSeek’s voracious appetite for capital to fund model research and computing infrastructure.

A war chest for the Huawei wager

Where the money goes matters as much as who provides it. DeepSeek has placed an enormous bet on China’s domestic chip supply chain, and the fundraise is widely seen as the fuel for that strategy.

Bloomberg reported in September that DeepSeek plans to deploy at least 160,000 Huawei Ascend 950DT chips at a large data centre in Ulanqab, Inner Mongolia — a project on the scale of roughly one gigawatt. The site, about 350 kilometres northwest of Beijing, was chosen for its cold climate and cheap power. Partial operations could begin in late 2027 or early 2028, according to the reports.

The company has been methodically preparing that transition. In April, DeepSeek adapted its V4 model to run on Huawei hardware; Huawei said its chips had been used in part of the training for the smaller V4-Flash model. On September 30, DeepSeek open-sourced a free toolkit for Huawei’s Ascend accelerators — including TileLang, described by industry watchers as a domestic counterpart to Nvidia’s CUDA — handing Chinese developers the means to migrate their own models to domestic silicon. Earlier this month, on October 17, DeepSeek released an updated checkpoint, DeepSeek-V4.1-Flash-1017, with open MIT-licensed weights, continuing the V4 family’s rapid release cadence.

According to The Information, DeepSeek chief executive Liang Wenfeng told investors in September that expanding the use of domestic chips, including Huawei’s, for AI model training was now a major priority, with Huawei expected to begin delivering training chips to DeepSeek as early as the fourth quarter of 2026 and into the first quarter of 2027, as reported by TrendForce. The same reports say DeepSeek is currently training a 2-trillion-parameter model and aims to eventually build an 8-trillion-parameter system.

IPO countdown begins

Behind the financing sits a corporate transformation. With the funds, the company plans to expand its computing infrastructure significantly and finalise an internal restructuring in preparation for a prospective initial public offering on Shanghai’s STAR Market — potentially as early as 2027, according to industry reporting.

The listing race is on elsewhere too. Rival Chinese lab Moonshot AI is reportedly planning a Hong Kong listing in the first quarter of 2027, targeting a $50 billion valuation. China’s AI labs, once content to be research operations, are racing to become listed, capitalised companies.

The geopolitical backdrop adds friction. This week, China’s foreign ministry demanded the United States “withdraw” restrictions on DeepSeek after the Department of Homeland Security ordered federal civilian agencies to remove the company’s products from their IT systems — an order originally issued in July warning that Beijing could exploit DeepSeek to obtain sensitive data. “China will firmly safeguard the legitimate rights and interests of Chinese enterprises,” foreign ministry spokeswoman Mao Ning said on Tuesday, according to China Daily. The DHS order covers products from other companies that rely on DeepSeek’s open-source models.

Analysis: Why It Matters

1. This is industrial capital, not venture capital

The investor list is the real headline hiding inside the number. When a battery manufacturer, a car company, and state-adjacent funds write the biggest cheques, they are not buying lottery tickets — they are buying a supply chain. CATL, Geely and their peers make things; they see AI not as a software bet but as an input to manufacturing, logistics and energy. That changes the texture of the funding: it is patient, strategic, and deeply tied to the physical economy. Silicon Valley’s AI rounds are financed by funds optimising for financial return; DeepSeek’s is financed, in large part, by companies optimising for industrial position. Both are rational — but they produce different kinds of companies.

2. The money is for Huawei silicon, and that is the ballgame

Strip away the valuation talk and this is, at bottom, a compute story. DeepSeek’s 160,000-chip Inner Mongolia plan is one of the largest commitments to Chinese-manufactured AI hardware on record, and the training-chip deliveries expected from late this year will be the hardest test yet for China’s domestic chip stack. The open-source toolkit released in September is the tell: DeepSeek is not just buying Huawei chips, it is building the software ecosystem — the CUDA moat’s Chinese-language answer — so that every Chinese lab can follow. If the 2-trillion-parameter model trains successfully on Ascend silicon, the export-control logic that assumes chip restrictions can indefinitely slow China starts to look optimistic. If it stalls, the sceptics are vindicated and DeepSeek is back in line for Nvidia. The $15 billion is, in effect, a bet on that outcome.

3. Open weights are the strategy, not the marketing

DeepSeek’s cadence of open releases — V4-Flash, V4.1-Flash-1017 with MIT-licensed weights, the TileLang toolkit — is easy to misread as generosity. It is better read as economics. Open weights buy developer adoption, and adoption buys the ecosystem that makes Huawei silicon viable; viable silicon makes the compute affordable; affordable compute makes the next model possible. Each open release is a subsidy paid in software to lower the barrier for everyone else. That flywheel, if it turns, is worth far more than the licence fees DeepSeek forgoes — and it is a direct inversion of the closed, subscription-first playbook being refined in the West.

4. Decoupling has not starved DeepSeek — it has rearmed it

There is an irony in this week’s news cycle worth sitting with. On Monday, the US government was scrubbing DeepSeek from federal computers; on Thursday, Bloomberg reported the company closing one of the largest startup rounds in history. The DHS order and the fundraise are two faces of the same phenomenon: as official channels between the two AI ecosystems narrow, the Chinese one is being flooded with domestic capital and talent. Restrictions were supposed to slow China down. Instead, they appear to have concentrated Chinese industrial capital behind the single most credible frontier lab. Policymakers in Washington betting that pressure would weaken DeepSeek should reckon with the opposite reading of this week’s evidence.

5. The STAR Market listing will be the verdict

The prospective 2027 STAR Market IPO is the moment all of this gets priced by strangers. A 500-billion-yuan pre-money valuation is a private-market number agreed among believers and strategists; a public listing is a number agreed with the world. Between now and then, DeepSeek must show that the compute works, that the models keep pace, and that open weights can convert into revenue at industrial scale. The doubling of the round — from 50 billion to 100 billion yuan — suggests the believers think it can. The market, eventually, gets the final vote.

What to Watch Next

1. The formal close: Bloomberg’s report rests on people familiar with the matter, and Reuters could not independently verify it. Watch for confirmation from DeepSeek, CATL or Tencent filings — particularly any Hong Kong or Shenzhen exchange disclosures of Tencent’s and CATL’s commitments.
2. The Inner Mongolia build: The 160,000-chip Ulanqab data centre is the physical test of this money. Watch for construction milestones and for Huawei’s training-chip deliveries from the fourth quarter of 2026.
3. The next model: DeepSeek’s V4 family has kept a rapid cadence. If the rumoured 2-trillion-parameter training run is on Huawei silicon, its results will be the single most important technical signal of the year for China’s AI hardware push.
4. The Washington response: The DHS DeepSeek order and Beijing’s “resolute countermeasures” warning are still fresh. Watch whether the US extends restrictions to investors or to the open-weight models themselves.
5. The IPO filings: A STAR Market prospectus, when it lands, will reveal DeepSeek’s revenue, margins and — crucially — how much of the $15 billion is already committed to hardware.

Sources

• Bloomberg — “DeepSeek Raises $15 Billion in Challenge to Leading US Labs” (October 22, 2026)
• Reuters — “DeepSeek raises $15 billion in challenge to leading US labs, Bloomberg News reports” (October 22, 2026)
• BigGo Finance (Caijing) — DeepSeek funding round details, investors and valuation (October 13, 2026)
• Background: DeepSeek V4, Huawei chips and the Inner Mongolia data centre plan