DeepSeek Reportedly Closes $15B Round Days After US Ban — One of China's Largest Ever

Reported: ¥100 billion — DeepSeek is closing a new funding round

On Thursday, October 22, Bloomberg News first reported that DeepSeek is wrapping up a new funding round, having raised about 100 billion yuan (approximately $14.98 billion) from existing and new investors. Reuters the same day cross-confirmed the figure, citing four people with knowledge of the matter. If confirmed, it would be one of the largest private financings in Chinese history.

A note on scope before we go on: two days earlier (October 20), the U.S. Department of Homeland Security ordered federal agencies to remove DeepSeek technology from their systems — the scope, wording, and impact of that move are covered separately in "DHS Orders Federal Agencies to Remove DeepSeek — Third-Party Wrappers Included" and won't be rehashed here. This article is about the capital side of the story: two days after the ban, DeepSeek secured one of the largest private investments in Chinese AI history.

One caveat that matters: as of publication, DeepSeek has not commented on the round, and every figure, valuation, and investor name below comes from media reports citing unnamed sources. One source told Reuters the round "is set to be complete in the coming days, but financial details and the investor lineup are not ironclad and could still change." Treat every number below as "reported."

The numbers: a doubled target, a ¥500 billion valuation

Item	Reported figure	Attribution
Round size	~¥100 billion (≈$14.98 billion)	Bloomberg first; Reuters cross-confirmed
Original target	~¥50 billion	Reuters: actual raise doubled it
Pre-money valuation	~¥500 billion (≈$74–75 billion)	Reuters reported ¥500B; WSJ reported ~$74B
Number of investors	More than 15	Reuters, citing sources
Lead investors	CATL, Tencent	Bloomberg/Reuters in agreement
New investors	Midea, Meituan, Geely, GigaDevice	Reuters
Other investors	Boyu Capital, CPE, Legend Capital, IDG Capital (existing), etc.	Reuters; WSJ also named Monolith, Stony Creek Capital, others
Founder stake	Liang Wenfeng holds ~81.6% of the domestic operating entity	TechCrunch, citing WireScreen ownership data

The ban-vs-funding contrast actually tells us two things

TechCrunch's headline put the timing front and center: "just two days after" the U.S. order. The contrast is tempting to read as a rebuttal, but the more useful read is that two unrelated logics are at work.

First, the DHS ban governs U.S. federal systems, not DeepSeek's business. The order targets DeepSeek technology inside U.S. federal civilian agencies — in plain terms, Washington telling its own staff not to use it. DeepSeek's commercial battlefield is elsewhere: according to the Wall Street Journal via PYMNTS, founder Liang Wenfeng told investors in September that annualized revenue had reached $1 billion, driven largely by cloud API access; gross margins on its V4 flagship API were reportedly 70%–80%. In August, DeepSeek introduced peak-hour pricing that raised flagship model rates roughly fourfold — while remaining cheaper than major U.S. rivals. The ban hurts its brand and channels in America; it doesn't touch the "cheap and good enough" fundamentals.

Second, this ¥100 billion is essentially a collective vote by Chinese industrial capital. Look at the lead and new-investor list: CATL (batteries), Tencent (social + cloud), Midea (appliances), Meituan (local services), Geely (autos), GigaDevice (semiconductors). This isn't a traditional financial-VC syndicate — it's a jigsaw of "Chinese manufacturing + internet + chips." Reuters put it bluntly: many investors are "backed by the state and big corporations," against the backdrop of China racing to get ahead of the U.S. in AI.

Chinese-style mobilization vs. American-style fundraising: two capital logics

Set against the global AI funding map, the contrast is sharp. The U.S. lab playbook is "cloud giants + sovereign funds + VCs": Microsoft, Amazon, and Google supply compute and cash, Middle Eastern sovereign funds write the giant checks, top VCs follow — money buying an "AGI option." DeepSeek's round runs on a different logic: "industrial chain + corporate capital" — battery makers, carmakers, appliance makers, chipmakers, and internet giants betting together.

Each logic has its own arithmetic. The American bet is on a capability singularity — money in, gambling on a future model breakthrough delivering monopoly-scale returns. The Chinese industrial bet is more pragmatic: AI models are the "water and electricity" of the next decade, and batteries, cars, appliances, and phones will all be remade by AI. Better to buy into the player most likely to make models cheap and open than to be a passive buyer later. How much of the ¥100 billion is return-driven versus strategic positioning, outsiders can't know; but the roster itself says DeepSeek is being funded as "national AI infrastructure," not as an ordinary startup.

One detail worth noting: the round reportedly started with a ¥50 billion target and ended up doubling to ¥100 billion. Oversubscribed twofold — after the DHS ban landed. At least in the Chinese capital context, the U.S. restriction reads as reverse certification: "this company is worth restricting."

The investor map: a guest list for "AI remakes every industry"

Spread the new and existing investor list flat and a pattern emerges: nearly all of them are industry leaders in sectors about to be remade by AI.

• CATL: the world's largest EV battery maker. Battery manufacturing is a classic high-yield, high-consistency process industry where AI's value in process optimization, defect detection, and materials R&D shows up directly in earnings. CATL has also been moving into humanoid robots and embodied AI — whose "brain" is the large model.
• Tencent: WeChat + Tencent Cloud. China's largest consumer traffic gateway and a major cloud provider — itself a big customer and distribution channel for DeepSeek's API. Tencent's investment hands DeepSeek's models a distribution network that never runs dry.
• Midea: the global appliance leader, reframing "smart home" as "AI home robotics." The next form factor of appliances is the agent; Midea needs a cheap, locally deployable brain.
• Meituan: a scheduling machine for local services. Food delivery and instant retail are, at bottom, massive real-time optimization problems — and models plus agents are rewriting its fulfillment stack.
• Geely: autos. Smart cockpits, autonomous driving — no carmaker dares fall behind on AI.
• GigaDevice: memory and MCU chips. On-device AI can't land without cheap storage and compute silicon — an upstream supply-chain position.

The subtext of the list: China's most profitable manufacturers and internet companies are using equity to "pre-order" AI capability for the next decade. They're not buying DeepSeek's current P&L; they're buying the option that says "if model capability reshuffles my industry one day, I want a seat at the table." The established names — Boyu, CPE, Legend Capital, IDG — are the financial investors pricing that option. This "industrial strategy + financial follow-on" mix is the standard AI financing structure in China's primary market over the past two years.

The global picture: two money-printing machines

TechCrunch's report helpfully listed a set of Chinese AI funding coordinates worth viewing side by side: Moonshot AI (Kimi) raised $2 billion at a $20 billion valuation in May; Manus raised over $500 million earlier this month; Zhipu (Z.ai) and MiniMax went public in Hong Kong in January with market caps of roughly $29 billion and $7.2 billion. And DeepSeek's own June round — over $7 billion — was already among the largest in the world at the time, in dollar terms.

The coordinates across the Pacific are a different order of magnitude: we previously reported Anthropic's Q3 revenue at $16.8 billion, profitable on an adjusted basis for two straight quarters, with a November IPO targeting a $2 trillion valuation. The American AI capital loop is "high revenue → high valuation → more fundraising → more compute" — an internally generated flywheel where models genuinely make money, so capital dares to bet bigger.

DeepSeek's loop runs on the opposite logic: "cheap open models → ecosystem capture → industrial capital infusion → more compute." Its API undercuts U.S. rivals by an order of magnitude; $1 billion in annualized revenue (per the WSJ figure) is a rounding error next to Anthropic's quarterly number — but it buys real adoption across the global developer ecosystem, including countless indie projects in the vibe coding community. The ¥100 billion is, in essence, industrial capital paying for that ecosystem capture, betting on scale first and monetization later.

Neither model is superior — only path-dependent. U.S. labs lean on self-sustaining cloud and subscription revenue; Chinese AI companies lean on an industrial base willing to pay for strategic positioning. The risks differ too: the former fears "the AI bubble popping and revenue being debunked," the latter fears "industrial capital running out of patience and the IPO window closing." Which is why DeepSeek reportedly hired CITIC Securities in September to prepare a STAR Market listing: however much private money there is, it needs a public-market exit, fast.

The 81.6% stake: why Liang Wenfeng dares take industrial money

An easily overlooked number: per TechCrunch citing WireScreen's ownership analysis, Liang Wenfeng still holds roughly 81.6% of DeepSeek's domestic operating entity after the earlier round. Founder supermajority control is why DeepSeek dares take money from 15 industrial investors — capital can join the cap table, but it can't rewrite the technology roadmap.

It's a subtle contrast with U.S. lab governance: OpenAI's nonprofit parent structure, Anthropic's long-term benefit trust — elaborate firewalls designed so capital can't hijack the AGI roadmap. DeepSeek's firewall is far simpler: the founder holds over four-fifths of the votes, and investors are financial backers and strategic partners, nothing more. It's also part of why industrial capital is willing to come — what it wants was never control, just a seat.

Supermajority control has its costs, of course: with 15 industrial shareholders of diverse backgrounds on board, disclosure and related-party scrutiny through the IPO process will be more complex than with pure financial investors. The STAR Market's "hard tech" review bar, plus DeepSeek's High-Flyer-incubated origin story, means the compliance homework for listing won't be light. That may be exactly why DeepSeek recruited former VC Yan Wentao as CFO in September — per the Journal, his first item on the job list is translating the company into a language public markets understand.

A third top-up in four months: capital is accelerating

Zoom out and DeepSeek's financing cadence is visibly accelerating:

• June 2026: reportedly raised over $7 billion at a ~$50 billion valuation;
• July 2026: reportedly sought another $1.5 billion at a ~$71 billion valuation;
• September 2026: reportedly hired CITIC Securities to prepare a STAR Market IPO and named Yan Wentao CFO;
• October 20, 2026: the U.S. DHS ordered federal agencies to remove DeepSeek;
• October 22, 2026: the ¥100 billion round reportedly closed, at a ~¥500 billion pre-money valuation.

From a $50 billion to a $75 billion valuation in four months. The September IPO-prep moves (CITIC + a new CFO, per the Journal) show DeepSeek putting a listing on the calendar — possibly as early as early 2027 on the STAR Market. PYMNTS even suggests it could "reach public markets before OpenAI." Many of the 15-plus investors in this round are, not coincidentally, buying the last ticket before the IPO.

What it means for the vibe coding ecosystem: a "certainty premium" for the open route

For this site's readers, the practical question is: what does ¥100 billion have to do with vibe coding? The answer: a lot — and it's good news.

The basic facts first. DeepSeek made its name in early 2025 with the open-weight R1 reasoning model — free, open, cheap — and single-handedly knocked a zero off global AI API pricing. It has since remained one of the most important open-model providers in the vibe coding toolchain: countless indie developers and small teams run their AI coding assistants and agent workflows on DeepSeek's API, because it's the best balance of "good enough" and "cheap." Liang's 81.6% stake (per WireScreen) also means the company's direction won't easily be bent by capital.

¥100 billion buys, first and foremost, certainty of compute and talent. DeepSeek is reportedly expanding domestic compute infrastructure and shifting more training workloads onto homegrown chips such as Huawei's Ascend (we previously covered its open-sourcing of an Ascend-targeted TileLang). For the developer ecosystem, that means the open-weights, high-value-for-money API route is unlikely to dry up over the next year or two — exactly what the vibe coding ecosystem fears losing most: a stable, cheap, continuously improving open-model foundation.

But the other side of the coin deserves equal weight. More money means more monetization pressure. August's fourfold peak-hour price hike was already a signal: once the valuation hits ¥500 billion and 15 industrial investors are waiting on IPO returns, how long DeepSeek's generosity with "free and open" lasts is a genuine question. What the open-source community should actually watch isn't how much was raised, but whether the next flagship model's weights stay open and whether API prices keep climbing. Those two things are the real test of whether this ¥100 billion flows to the ecosystem or to the income statement.

One level deeper: the ban and the funding are two sides of "decoupling"

Put the October 20 DHS ban and the October 22 ¥100 billion raise on the same chart and they describe one reality: Chinese and American AI are accelerating into two parallel markets. The U.S. market logic is "security review first" — get DeepSeek out of federal systems, then talk. China's is "capital mobilization first" — ¥100 billion to make sure its homegrown models don't fall behind on compute and talent.

For developers, parallel markets mean rising cost of picking sides. A vibe coder today can prototype an agent on DeepSeek's API at near-zero cost — but if the target users are in the U.S., the spread of DHS-style restrictions (state- and agency-level limitation discussions were already underway) puts "DeepSeek-wrapped" products at compliance risk — precisely the "third-party wrappers included" lethality we unpacked in the ban article. Conversely, for domestic or Global South markets, DeepSeek's value and open ecosystem remain without substitute.

So what this round really changes isn't "who wins" — it's that the investment certainty of both markets got reinforced at once: Washington telling its labs "expand with confidence, I'll guard the market," Beijing telling DeepSeek "spend with confidence, I've got your back." For indie developers caught in the middle, the best strategy isn't picking a side but staying "model-swappable" — DeepSeek as the workhorse, an aggregation layer as backup — letting the ¥100 billion work for you without being locked in by either side's policy risk.

Which "reported" figures deserve a discount

A final transparency audit of every key figure's attribution:

• The "~¥100 billion / $14.98 billion" raise: first reported by Bloomberg, cross-confirmed by Reuters citing four sources — the hardest number available; but DeepSeek, CATL, and Tencent all declined to comment, so the official figure is zero;
• "¥50 billion target, doubled": single-thread Reuters sourcing;
• "~¥500 billion pre-money valuation": Reuters; the WSJ's figure was "~$74 billion" (≈¥494 billion at the prevailing rate) — broadly consistent, differences down to FX and rounding;
• "15-plus investors" and the roster: Reuters citing three/one sources; the WSJ's list (Monolith, Stony Creek Capital, etc.) doesn't fully overlap with Reuters', suggesting the final roster is still moving — consistent with "details not yet ironclad";
• "$1 billion annualized revenue, 70%–80% V4 API gross margins, ~4x August price hike": all WSJ-via-PYMNTS "reported" figures;
• "CITIC Securities hired for STAR Market IPO prep, Yan Wentao as CFO": Reuters (September) and WSJ sourcing;
• "Early-2027 listing, possibly before OpenAI": PYMNTS's framing — media inference, not company guidance.

In one line: Washington used a ban to keep DeepSeek out of federal systems; Chinese industrial capital used ¥100 billion to put it on the "national AI infrastructure" list. For vibe coders, the near term is good news — the open foundation's funding certainty just got stronger. The long-term watch item is whether this money changes DeepSeek's open-source generosity. Where the money flows decides how open the models stay.

Related: DHS Orders Federal Agencies to Remove DeepSeek — Third-Party Wrappers Included (the ban); DeepSeek's Annualized Revenue Tops $1 Billion (the revenue story).