April to August 2026: first round, IPO prep, pause, and restart
DeepSeek's fundraising arc moved fast. In April 2026, a registered capital increase pushed founder Liang Wenfeng's stake from 1% to 34%, giving him roughly 84.29% total control. The same month the company opened its first-ever funding round and previewed V4. By June 2026, that first round closed at roughly 50 billion yuan (~$7.4 billion) at a post-money valuation above 350 billion yuan (roughly $52–59 billion). Tencent committed 10 billion yuan, CATL 5 billion yuan, and the National AI Fund, JD, NetEase, and IDG also participated. Liang personally invested 20 billion yuan.
In mid-July, DeepSeek began STAR Market IPO preparation while simultaneously opening second-round talks at ~$71 billion pre-money. Revenue metrics surfaced for the first time: ARR of $400–500 million. On July 25–26, talks paused — Bloomberg reported Liang was unhappy about comments from a closed-door meeting circulating publicly. By August 4–5, Caijing cited a restart: 50 billion yuan at 500 billion yuan pre-money, with a late-August signing target. Sources say DeepSeek wants to keep the process low-profile.
| Date | Event | Key detail |
|---|---|---|
| April 2026 | Cap table shift + first round opens | Liang stake 1%→34%, total control ~84.29%; V4 preview ships |
| June 2026 | First round closes | ~50B yuan (~$7.4B); post-money >350B yuan ($52–59B); Tencent 10B, CATL 5B, National AI Fund, JD, NetEase, IDG; Liang 20B yuan personal |
| July 14–17 | STAR Market IPO prep + round-two talks | Second round at ~$71B pre-money; ARR $400–500M first reported |
| July 25–26 | Talks paused | Liang unhappy about closed-door meeting comments circulating (Bloomberg) |
| August 4–5 | Round-two restart | Caijing: 50B yuan at 500B yuan pre-money; late-August signing; low-profile approach |
| August 6 (today) | Not yet closed | Second round still in negotiation; no official DeepSeek confirmation |
Treating Caijing numbers as confirmed: Every valuation and round size in this article comes from anonymous dealmakers — DeepSeek has not issued an official statement.
Assuming the July pause killed the deal: Talks restarted within two weeks; the Bloomberg-reported friction was about leaked comments, not fundamental terms.
Equating pre-money with post-money: Round one closed at post-money >350B yuan; round two is discussed at ~500B yuan pre-money — a 43% step-up before new capital lands.
Ignoring the IPO parallel track: Fundraising and STAR Market filing are running simultaneously — round-two capital may fund compute before IPO proceeds arrive.
Expecting investor voting power: Most round-one participants reportedly hold no voting rights through an LP structure controlled by Liang — a detail that may resurface in round two.
Comparing to US multiples without revenue context: DeepSeek's ~140–150× P/S looks extreme next to OpenAI (~65×) and Anthropic (~21×), but ARR is still sub-$500M.
Round one vs round two — and how DeepSeek stacks against peers
| Metric | Round 1 (June 2026) | Round 2 (Aug 2026, reported) |
|---|---|---|
| Raise size | ~50B yuan (~$7.4B) | ~50B yuan (~$7B) |
| Valuation basis | Post-money >350B yuan ($52–59B) | Pre-money ~500B yuan (~$70B) |
| Valuation step-up | First-ever institutional round | ~43% above round-one post-money |
| Key investors | Tencent, CATL, National AI Fund, JD, NetEase, IDG; Liang 20B yuan | Not yet disclosed (talks ongoing) |
| Status | Closed | Restarted; late-August signing target |
| Cumulative raised (if closes) | ~$7.4B | Over $14B in under 5 months |
| Company | Valuation | ARR (reported) | P/S multiple | MAU / scale | Notes |
|---|---|---|---|---|---|
| DeepSeek | ~500B yuan pre (~$70B) | $400–500M | ~140–150× | 100M+ MAU | Round-two talks; gross margin reportedly strong |
| OpenAI | ~$852B (Mar 2026) | — | ~65× | — | Reference multiple for frontier labs |
| Anthropic | $65B Series H (May 2026) | — | ~21× | — | Lower P/S at higher revenue base |
| Moonshot AI | ~$20B (May 2026) | ~$200M | — | — | Kimi K3 open-weight rival |
| Zhipu AI | ~350B yuan (HK listed) | — | — | — | Public-market comp in China |
| MiniMax | ~210B yuan (HK listed) | — | — | — | Public-market comp in China |
At a reported ~148× price-to-sales, round-two investors are not buying current revenue — they are buying an options bet on DeepSeek's compute scale, in-house chips, and a potential STAR Market IPO in 2027. The multiple is roughly 2.3× OpenAI's ~65× and 7× Anthropic's ~21× on comparable framing.
Where the money goes: compute, voting rights, lockups, and STAR Market context
Compute is the real bill. Sources close to the deal say roughly 70% of every 10 billion yuan invested flows directly to compute — data centers, GPU clusters, and the infrastructure behind DeepSeek's 100M+ monthly active users. That explains why a company that just raised $7.4 billion is back for another $7 billion within months: training and serving frontier models at scale burns capital faster than ARR can replace it.
Voting rights are the controversy. Most round-one investors reportedly received no voting rights, participating through an LP structure controlled by Liang Wenfeng, with a five-year lockup. The National AI Fund is the notable exception — it reportedly holds voting rights. Whether round-two terms repeat this structure is unresolved and may be one reason Liang paused talks in late July.
Policy backdrop shifted in June. On June 17, 2026, the STAR Market expanded its fifth listing standard to cover AI companies, opening a domestic IPO path that did not exist when DeepSeek was founded. DeepSeek also abandoned a long-standing five-year no-fundraising policy, signaling a strategic pivot toward external capital and compute self-reliance — including investment in in-house chips to reduce dependence on Nvidia supply chains.
Note: All deal terms, investor rights, and valuation figures in this section come from anonymous dealmakers cited by Caijing and Bloomberg. DeepSeek has not published official confirmation of round-two size, valuation, or governance structure.
Six-step checklist: evaluating DeepSeek's second round as an observer or builder
Separate fundraising from product milestones: V4-Flash shipped July 31; round-two capital funds infrastructure, not the next model release. See V4-Flash benchmarks for API status.
Track official vs reported numbers: $70B pre-money and 50B yuan raise are Caijing-sourced — wait for a DeepSeek or regulatory filing before treating them as settled.
Model total capital, not round size: Round one plus round two exceeds $14B in under five months — benchmark against Moonshot (~$20B valuation, ~$200M ARR) and listed peers Zhipu and MiniMax.
Read governance terms, not just valuation: No voting rights plus five-year lockup means most investors are passive LPs — only the National AI Fund reportedly breaks that pattern.
Map the IPO timeline: STAR Market filing expected end-2026, debut likely 2027 — retail access for international investors will be limited even after listing.
Plan your own compute stack: If you build on DeepSeek APIs or run open-weight models locally, dedicated hardware matters for 24/7 agent workloads — see the help center.
Three hard numbers, unverified-deal caveats, and what builders should do next
DeepSeek's second round is less a surprise raise than a acceleration of a strategy set in motion in April: take external capital, scale compute, pursue domestic chip independence, and file for a STAR Market IPO — all while keeping Liang Wenfeng in effective control. The July pause over leaked meeting comments shows the founder values discretion as much as valuation. The August restart at a higher pre-money suggests investors still want in, even at ~148× sales.
For developers and operators, the funding round changes little about API access today — but it signals where DeepSeek is investing: data centers, headcount, and silicon. If you depend on DeepSeek for production workloads, the risk is not valuation hype; it is whether compute expansion keeps pace with 100M+ MAU growth and whether governance terms affect long-term strategic decisions.
$14B / 5 months: Combined round-one and round-two capital if the August deal closes — unprecedented pace for a Chinese AI lab that had never raised institutional money before April 2026.
148× / 43%: Reported price-to-sales at ~$70B pre-money on $400–500M ARR, with valuation up 43% from round-one post-money — investors pay for growth optionality, not current revenue.
70% / 84.29%: Roughly 70% of each 10B yuan tranche goes to compute; Liang holds ~84.29% total control after April's cap-table shift — capital and control stay concentrated.
Caution: Round-two size, $70B pre-money valuation, ARR figures, investor rights, and IPO timing all come from anonymous sources (Caijing, Bloomberg) — not official DeepSeek disclosures. Treat every number as provisional until confirmed by the company or a regulatory filing.
Stacking alternatives: waiting for IPO access means missing 12–18 months of product iteration; betting on API stability alone ignores compute-scale risk; running agent workloads on a personal Mac hits sleep, network, and queue limits. For iOS CI/CD, multi-model A/B testing against Kimi K3 and Qwen3.8-Max, and production-grade agent automation, KVMNODE dedicated Mac Mini M4 cloud rental is usually the better fit: exclusive Apple Silicon, sudo access, multi-region nodes, daily/weekly/monthly billing. See the pricing page and order page.
Data as of August 6, 2026 · Sources: Caijing round-two restart report, Bloomberg July pause coverage, June 2026 first-round close reporting, STAR Market fifth listing standard expansion (June 17, 2026)