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

Product context: DeepSeek V4-Flash benchmarks. Compute self-build rumors: DeepSeek custom AI chip reporting.

DeepSeek, the Chinese AI lab behind the open-weight R1 and V4 models, has restarted talks for a second funding round targeting roughly 50 billion yuan (~$7 billion) at a pre-money valuation of about 500 billion yuan (~$70 billion) — up 43% from its first round two months ago. If it closes, DeepSeek will have raised over $14 billion in under five months.

01 · Three decision pitfalls: treat Round 2 as reported, not signed

1. Stage mismatch: amount, valuation, and timeline for Round 2 come from anonymous dealmakers cited by financial media — DeepSeek has not issued an official confirmation.
2. Valuation vs revenue gap: at ~$70B pre-money against ~$400–500M ARR, implied P/S sits around 140–150x, far above OpenAI (~65x) and Anthropic (~21x) per dealmaker estimates.
3. Capital narrative ≠ runtime readiness: fundraising cadence tracks compute buildout; it does not automatically mean API SLAs or agent frameworks are production-ready. Validate in isolation.

02 · Timeline: from "no fundraising" to a $70 billion valuation in four months

• April 2026: Corporate filing shows registered capital increased; founder Liang Wenfeng raised his direct stake from 1% to 34%. Combined with an entity he controls, total control reached roughly 84.29%. First external round opened; V4 series previewed.
• June 2026: First round closed at roughly 50 billion yuan (~$7.4 billion), post-money above 350 billion yuan (reported $52–59B across sources) — largest first-round raise in Chinese AI history. Backers included the National AI Industry Investment Fund, Tencent (10B yuan), CATL (5B yuan), JD.com, NetEase, IDG Capital and others.
• July 14–17, 2026: Outlets reported STAR Market IPO prep and Round 2 talks at roughly $71B pre-money (~480B yuan). ARR of $400–500M (mostly API tokens) surfaced publicly 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.
• August 4–5, 2026: Caijing-cited dealmakers said the round restarted, still targeting 50B yuan at ~500B yuan pre-money (~$70B), with signing expected in late August. Both sides want a low-profile process.

Caveat: every Round 2 figure above — amount, valuation, timeline — comes from anonymous dealmakers, not an official DeepSeek statement. Terms can still shift.

03 · The numbers at a glance

Hard number #1: If Round 2 closes as targeted, cumulative fundraising exceeds 100 billion yuan (~$14B) in under five months.

Hard number #2: Round 2 pre-money (~500B yuan) is about +43% above Round 1 post-money (~350B yuan).

Hard number #3: Analysts estimate that for every 10 billion yuan raised, roughly 7 billion yuan goes into compute-related spend.

04 · Inside the deal: compute bills, voting rights, and option pricing

4.1 The real bill is compute

Shortly after Round 1, DeepSeek said it would double headcount across data-center and AI-agent teams. Reuters reported chip-design hiring for in-house AI inference chips. Fundraising cadence is a race to keep pace with compute buildout, not a vanity valuation contest.

4.2 Most investors don't get a vote

In Round 1, most outside capital entered through a limited partnership controlled by Liang, with no voting rights and a 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% — a structure that has drawn governance scrutiny from outlets such as Forbes.

4.3 A 148x P/S is an options bet

One dealmaker's line, translated from Chinese coverage: pricing a foundation-model company is fundamentally an options bet, not a cash-flow valuation. Investors are pricing the chance DeepSeek becomes infrastructure-level in China's compute ecosystem and enterprise agent market.

05 · How DeepSeek stacks up against Moonshot, Zhipu, MiniMax

Private-market investors are, for now, paying a steeper premium for the two still-private frontrunners than secondary markets assign to already-listed peers.

06 · Controversy: leaked transcript, voting rights, bubble warnings

1. Leaked closed-door transcript stalled the July talks — a reminder that a larger investor base makes a low profile harder.
2. Voting-rights structure leaves most external investors without a vote while the state-backed fund sits as the exception; DeepSeek has not publicly resolved the governance questions.
3. Valuation-to-revenue gap remains unresolved: 140–150x P/S is extreme even versus high-growth SaaS (often 30–50x). Whether it holds depends on converting technical lead into scaled enterprise revenue after a STAR listing — still untested.

07 · Why it matters: STAR rules, compute self-reliance, global re-pricing

• STAR Market fifth listing standard expanded to AI on June 17, 2026 — unprofitable or low-revenue AI firms can file if technology is strong enough. That underpins DeepSeek's reported late-2026 filing / 2027 listing path.
• End of the five-year "no fundraising, no IPO, no commercialization" policy funded solely by Liang's quant fund High-Flyer until Round 1 closed in June 2026.
• Global backdrop: OpenAI was reportedly valued at $300B in 2025; Anthropic's valuation reportedly surpassed OpenAI by June 2026. Steep premiums for competitive Chinese labs fit a wider frontier re-pricing.
• Compute self-reliance is the subtext behind chip and data-center buildouts — and why modest revenue still implies heavy fundraising cadence.

08 · Five-step Mac isolation: don't let headlines pick your model ID

1. Provision an isolated Mac node with only the tools you need (Cursor / Claude Code / OpenClaw). Skip personal iCloud and company SSO.
2. Store the DeepSeek API key in node-local env vars — never sync it to your daily laptop.
3. Run the same agent task suite against deepseek-v4-flash and peers (Kimi K3 / Qwen3.8-Max); log latency, failure rate, and cost per task.
4. Document hard constraints headlines never change: rate limits, regional availability, key rotation, billing alerts.
5. Destroy the node after sign-off; capture conclusions in an internal selection memo, not a bookmark folder of valuation articles.

09 · FAQ

Has DeepSeek's second funding round actually closed?
Not yet. Still in negotiation, targeting late August 2026. Final terms may differ from reports.

Why raise again so soon?
Rapid data-center, in-house chip, and headcount buildout is outpacing what Round 1 covered, per multiple reports.

Is the $70B valuation confirmed?
No. Anonymous dealmakers via Caijing and peers — not an official DeepSeek statement.

Do investors get more control?
Not necessarily. Most Round 1 outside capital had no vote and a five-year lock-up; only the National AI Industry Investment Fund got direct voting rights.

When might DeepSeek go public?
Reportedly filing for STAR Market by end of 2026, targeting 2027. This private round remains institutional-only.

10 · Rent a Mac to validate: reading deal news is not a production plan

You can skim Caijing summaries and flip a model ID on your personal laptop, but a daily driver is the wrong place for multi-vendor key rotation and agent bake-offs. Common limits: company credentials mixed into the same Keychain, Windows/Linux cloud boxes that cannot fully exercise Apple Silicon–adjacent tooling next to Xcode sidecar work, and temporary VPS images that miss native macOS fidelity. If you want reproducible comparisons, day-based billing, and real Apple hardware, an isolated Mac rental is usually the cleaner path. See bare-metal macOS pricing.

11 · Sources

• Caijing magazine reporting, as relayed by Sina Finance and Wall Street CN
• The Standard (HK), Gate News, ChainCatcher coverage of the restarted round
• Forbes: "DeepSeek Just Raised $7.4 Billion. Here's The Catch."
• South China Morning Post, Caixin Global, Reuters, Bloomberg reporting
• CIW on the Round 1 cap-table / voting structure
• DeepSeek API docs, TechCrunch, Hugging Face write-ups on DeepSeek-V4
• 36Kr, TMTPost coverage comparing Moonshot, Zhipu, and MiniMax

Most figures come from anonymous sources and media reports rather than official disclosures. Figures as of August 6, 2026 — verify before publishing.