Why Is DeepSeek Raising Another $7 Billion Just Months After Its First Round?

Separate closed rounds from headlines before you update a board deck

DeepSeek spent years saying it would not raise, list, or commercialize. The narrative flip is sharp. Before treating "$70B" as a settled fact, keep five frictions in view:

• Round 2 is still unsigned: Amount, valuation, and timeline come from anonymous dealmakers and Caijing-style coverage. Terms can still move.
• Pre-money vs post-money mix-ups: Round 1 is usually quoted as post-money above 350B yuan; Round 2 as pre-money ~500B yuan. The ~43% step-up compares those two bases — not a single mark-to-market snapshot.
• ARR and margin are unaudited: Reported ARR of $400–500M and gross margin above 50% are media-sourced, not formal company disclosures.
• Most investors do not get a vote: In Round 1, most outside capital entered through a Liang-controlled LP — no voting rights, five-year lock-up. The National AI Industry Investment Fund was the exception.
• Product cadence is a parallel track: Fundraising sits beside V4 Flash GA and compute buildout. Do not collapse API price wars and private-market multiples into one conclusion.

Timeline: from "no fundraising" to a $70B valuation in four months

• April 2026: A corporate filing shows registered capital increased and founder Liang Wenfeng personally subscribed, raising his direct stake from 1% to 34%. Combined with an entity he controls, total control reached roughly 84.29% — the ownership structure that later shaped how outside capital entered. The same month DeepSeek opened its first external round and previewed V4.
• June 2026: Round 1 closed at roughly 50 billion yuan (~$7.4 billion), post-money above 350 billion yuan (reported $52–59B across sources) — the largest first-round raise in Chinese AI history. Liang personally contributed 20 billion yuan; Tencent put in 10 billion yuan, CATL 5 billion yuan, with JD.com, NetEase, IDG Capital and China's National AI Industry Investment Fund also participating.
• July 14–17, 2026: Outlets reported STAR Market IPO prep and second-round talks at a pre-money valuation of roughly $71 billion (~480 billion yuan) — about 37% above Round 1 post-money. ARR of $400–500 million, mostly from API tokens, became public for the first time.
• July 25–26, 2026: Talks paused. Bloomberg and others cited Liang's frustration that closed-door investor remarks had circulated online; some standby investors were told to hold off on signing.
• August 4–5, 2026: Dealmakers cited by Caijing said the round restarted, still targeting 50 billion yuan at ~500 billion yuan (~$70 billion) pre-money, with signing expected in late August. Both sides reportedly want to keep the round low-profile.

The numbers at a glance: Round 1 closed vs Round 2 in talks

Inside the deal: compute bills, voting rights, and a 148x P/S multiple

The real bill is compute, not headlines. Shortly after Round 1, DeepSeek said it would double headcount across data-center and AI-agent teams, and Reuters reported it was hiring chip-design engineers for in-house AI inference chips. Analysts estimate that for every 10 billion yuan raised, roughly 7 billion yuan goes into compute — chips, data centers, bandwidth, liquid cooling. The cadence is a race to keep pace with buildout, not a valuation brag contest.

Most investors don't get a vote. Round 1 capital largely flowed through a limited partnership controlled by Liang Wenfeng: no voting rights, five-year lock-up. The exception: China's National AI Industry Investment Fund invested directly with voting rights and no lock-up. Liang's control stays near 84%. Outlets like Forbes have flagged governance and state-alignment questions around that cap table.

A 148x P/S is either a future bet or a red flag. At ~$70B pre-money against $400–500M ARR, implied P/S sits around 140–150x — well above OpenAI's ~65x and Anthropic's ~21x. One dealmaker's line, as translated in Chinese coverage: pricing a foundation-model company is an options bet, not a cash-flow valuation. Investors are pricing the chance DeepSeek becomes infrastructure-level in China's compute stack and enterprise agent market.

How DeepSeek stacks up against Moonshot, Zhipu, and MiniMax

DeepSeek and Moonshot — still private — both carry P/S multiples around 140–150x, above what listed Zhipu and MiniMax trade at in the secondary market. Private-market investors are paying a steeper premium for labs that have not yet faced public-market scrutiny.

The controversy: a leaked transcript, voting rights, and bubble warnings

1. A leaked closed-door transcript stalled the deal. The July pause was reportedly triggered by Liang's frustration that first-round investor-meeting remarks spread online — a reminder that a growing investor base makes a low profile harder to keep.
2. Voting-rights structure draws outside scrutiny. Most external investors have no vote and a five-year lock-up; only the state-backed National AI Industry Investment Fund gets direct voting rights with no lock-up. DeepSeek has not publicly settled the governance questions this raises.
3. The valuation-to-revenue gap remains unresolved. A 140–150x P/S is extreme even for high-growth SaaS (often 30–50x). Whether the mark holds depends on converting technical lead into scaled enterprise revenue after a STAR Market listing — still untested by public markets.

Deal size, valuation, and ownership details above come from anonymous-sourced reporting (Caijing, Reuters, Bloomberg, Forbes, among others). Treat specific Round 2 numbers as reported-but-unconfirmed until a formal announcement.

Why it matters: STAR Market rules, compute self-reliance, and the global AI funding race

• Listing rules just shifted for unprofitable AI companies. On June 17, 2026, the Shanghai Stock Exchange expanded the STAR Market's "fifth listing standard" to cover AI — companies need not be profitable or even have significant revenue to file if technology is strong enough. That backdrop makes a late-2026 IPO filing (2027 listing target) plausible.
• A five-year "no fundraising, no IPO, no commercialization" policy ended. DeepSeek was funded by Liang's quant fund High-Flyer until Round 1 in June 2026. Rivals Zhipu and MiniMax are already listed in Hong Kong; Moonshot keeps raising at speed.
• Global re-pricing of frontier labs continues. OpenAI was reportedly valued at $300B in 2025; Anthropic's mark reportedly surpassed OpenAI's by June 2026. Paying a steep premium for a globally competitive Chinese lab is partly a bet on technical pace plus market scale.
• Compute self-reliance is the subtext. Reports of in-house AI inference chips and owned data centers mirror a broader push to pair frontier models with domestic compute — which also explains why modest reported revenue still requires frequent, large raises.

Six steps to stress-test the funding story before you act on it

Headlines about "$70B" and "$14B combined" are easy to forward. Use these six steps to turn the story into something you can cite — or discard.

1. Tag each figure's status: Closed / in talks / rumor. Round 1's ~50B yuan is closed. Round 2's amount and ~500B yuan pre-money are media-cited talks.
2. Normalize valuation basis: Fix pre-money or post-money, and currency (yuan vs USD). Keep "about $100B → about $70B pre-money path" separate from "350B post → 500B pre" yuan accounting.
3. Recompute P/S yourself: ~500B yuan pre-money ÷ (ARR $400–500M converted to yuan) should land near 140–150x. Compare to OpenAI ~65x and Anthropic ~21x instead of stopping at "expensive."
4. Map the cap table: LP indirect stakes (no vote, five-year lock) vs National Fund direct stake. Check whether "founder control" and "governance transparency" can both be true in the same deck.
5. Align product and capital calendars: Place funding nodes next to V4 Flash GA, OpenRouter usage share, and chip/data-center reporting. Ask whether the cash is buying compute or narrative.
6. Run API/agent A/B on a wipeable Mac: Capital news does not change SLAs. If production depends on DeepSeek or a peer switch, blind-score 20–50 real tasks on a root-capable, wipeable Mac — not a shared CI image full of stale SDKs.

pre_money_cny = 500_000_000_000
arr_usd_low, arr_usd_high = 400_000_000, 500_000_000
usd_cny = 7.1
for arr_usd in (arr_usd_low, arr_usd_high):
    arr_cny = arr_usd * usd_cny
    print(f"ARR ${arr_usd/1e8:.1f}B → P/S ≈ {pre_money_cny / arr_cny:.0f}x")

Citable figures and sources

• Round 1: ~50B yuan raised; post-money above 350B yuan; backers include the National AI Industry Investment Fund, Tencent (10B yuan), CATL (5B yuan), and others listed above.
• Round 2 (media-reported, unconfirmed): Target ~50B yuan at ~500B yuan (~$70B) pre-money; late-August 2026 signing planned; combined total would exceed 100B yuan (~$14B) if it closes.
• ARR / P/S: ARR ~$400–500M (media); implied Round 2 P/S ~140–150x.
• Control: Liang's direct stake rose to ~34%; combined control ~84.29%; most outside investors entered via LP with no vote and a five-year lock-up.
• STAR Market: Fifth listing standard expanded to AI on June 17, 2026; reports cite a filing target by end-2026 and a 2027 listing goal.

Funding and product status can change quickly — treat official filings and company statements as authoritative. Figures in this article are current as of August 6, 2026.

Chinese financial coverage (relay and original reporting leads):

Sina Finance / Caijing — DeepSeek restarts fundraising at ~500B yuan pre-money

Sina Finance — DeepSeek second round: 500B yuan pre-money, 50B yuan target raise

English and third-party coverage:

Forbes — DeepSeek Just Raised $7.4 Billion. Here's The Catch.

Gate News — DeepSeek Reopens Second Funding Round

FAQ

The Round 2 numbers are striking, but "in talks, not signed," mismatched voting rights, a 140–150x P/S, and unaudited ARR are real frictions. Capital narrative does not fix API jitter, long-horizon agent reproducibility, or peer model switches in production — those need a wipeable, root-capable Apple Silicon box. KVMFLUX Mac mini M4 rentals start by the day with real hardware and minute-level SSH, so you can finish DeepSeek / Kimi / Qwen sample suites before you chase the fundraising headline.