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

Who: DeepSeek. When: talks restarted August 4, 2026, with signing targeted for late August. What: a second external round aiming for ~50 billion yuan. Scale: ~500 billion yuan pre-money, about 43% above the first round’s ~350 billion yuan post-money. Stakes: if it closes, combined fundraising tops 100 billion yuan (~$14B) in under five months — and accelerates a company that once insisted on “no fundraising, no IPO, no commercialization” into the center of China’s AI capital story.

• April 2026: A corporate filing shows DeepSeek increased registered capital; 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 same month, DeepSeek opened its first external round and previewed V4.
• June 2026: Round 1 closed at roughly 50 billion yuan (~$7.4B), with post-money valuation 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 10 billion yuan; CATL 5 billion yuan; JD.com, NetEase, IDG Capital, and China’s National AI Industry Investment Fund also participated. See the supercycle breakdown.
• July 14–17, 2026: Outlets reported STAR Market IPO prep plus second-round talks at roughly $71B pre-money (~480 billion yuan), about 37% above Round 1 post-money. ARR of ~$400–500M (mostly API tokens) surfaced publicly for the first time.
• July 25–26, 2026: Talks paused. Bloomberg and others said Liang was unhappy that closed-door investor remarks had circulated online; some standby investors were told to hold off signing.
• August 4–5, 2026: Dealmakers cited by Caijing said the round restarted, still targeting 50 billion yuan at ~500 billion yuan (~$70B) pre-money, with late-August signing. Both sides reportedly want a low-profile process.

• Treating “in talks” as “closed”: Round 2 has not signed; size and valuation can move.
• Mixing pre- and post-money: Round 1 is often quoted post-money (>350B yuan); Round 2 rumors are pre-money (~500B yuan).
• Assuming retail access: Institutional-only structures, five-year lock-ups, and limited voting rights dominate Round 1.
• Ignoring the real bill: Analysts estimate ~70% of each 10B yuan raised goes to compute — chips, data centers, bandwidth, liquid cooling.
• Applying SaaS P/S instincts: A 140–150x sales multiple is an options bet, not cash-flow underwriting.
• Reading IPO dates as commitments: Late-2026 filing / 2027 listing targets are media-reported goals, not official schedules.

The real bill is compute, not headlines

Shortly after Round 1, DeepSeek said it would double headcount across data-center and AI-agent teams. Reuters reported hiring for in-house AI inference chips. Industry analysts estimate that for every 10 billion yuan raised, roughly 7 billion yuan goes into compute — chips, data centers, bandwidth, liquid cooling. Fundraising cadence is a race against the buildout, not a valuation trophy. That lines up with custom inference-chip coverage.

Most investors don’t get a vote

In Round 1, most outside capital flowed through a limited partnership controlled by Liang Wenfeng: 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. The structure keeps Liang near ~84% control — and drew governance scrutiny from outlets such as Forbes.

A 148x P/S is either an options bet — or a red flag

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 a chance DeepSeek becomes infrastructure-level in China’s compute ecosystem and enterprise agent market. For global context, see Anthropic’s Series H / IPO path.

DeepSeek and Moonshot — still private — both carry P/S multiples around 140–150x, above already-listed Zhipu and MiniMax. Private markets are paying a steeper premium before public-market scrutiny.

• Leaked closed-door remarks stalled the deal: The July pause was reportedly triggered by Liang’s frustration that first-round investor-meeting comments had spread online.
• Voting-rights structure draws 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. Forbes and CIW flagged governance questions; DeepSeek has not publicly settled them.
• Valuation-to-revenue gap remains unresolved: A 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 Market listing — still untested.

To be clear: deal size, valuation, and ownership details come from anonymous-sourced reporting (Caijing, Reuters, Bloomberg, Forbes, among others). Treat specific numbers as reported-but-unconfirmed until a formal announcement.

• Listing rules loosened for unprofitable AI firms: On June 17, 2026, the Shanghai Stock Exchange expanded the STAR Market’s “fifth listing standard” to AI — profitability or large revenue is not required if technology is strong enough. That is the backdrop for a reported late-2026 IPO filing targeting 2027.
• Five-year “no fundraising / no IPO / no commercialization” ended: DeepSeek was funded by Liang’s quant fund High-Flyer until Round 1 in June 2026. Peers Zhipu and MiniMax already listed in Hong Kong; Moonshot keeps raising fast.
• Global re-pricing of frontier labs: OpenAI was reportedly valued at $300B in 2025; Anthropic’s valuation reportedly surpassed OpenAI by June 2026. Premiums for a globally competitive Chinese lab are part of that race.
• Compute self-reliance is the subtext: Reports of in-house inference chips and owned data centers explain why modest ARR still demands rapid capital. Product context: V4 Flash official benchmarks.

Fundraising headlines do not change today’s token bill — but they do change vendor-risk assumptions and compute-capacity planning. Use this sequence:

1. 01
   Label Round 2 as unconfirmed — keep internal notes tagged “anonymous dealmaker / pre-signing”; do not bake figures into contracts or public decks.
2. 02
   Re-check supplier concentration — if production traffic leans on DeepSeek APIs, document failover routes (multi-model gateway + caching).
3. 03
   Budget for the compute race — treat the “~70% of raise → compute” analysis as a capacity signal, not a stock tip; pre-book eval and inference headroom.
4. 04
   Compare peers on ARR / P/S / listing status — DeepSeek vs Moonshot vs Zhipu vs MiniMax, so a single “$70B” headline does not dominate the memo.
5. 05
   Decouple agent hosts from model choice — models swap; long-running agents, CI, and eval fleets need a stable dedicated plane first.
6. 06
   Validate with a real bill — size cloud Mac / agent-host cost on the pricing page, then trial 7×24 sessions via the order page; decide on measured latency and disconnects, not funding headlines.

If Round 2 closes as reported, DeepSeek will have raised over $14B in under five months at ~$70B pre-money. The driver is compute buildout plus a STAR Market path — not valuation vanity alone. Voting rights, leaked closed-door remarks, and a 140–150x P/S remain the three live controversies. Until DeepSeek confirms terms, keep every figure marked as rumored and in negotiation.

While capital narratives move fast, agent, eval, and CI workloads still need an auditable, low-jitter compute plane. Shared minute pools, oversubscribed VPS, and under-desk Macs often lose the token savings to bandwidth jitter, noisy neighbors, and dropped long sessions. For a more stable production surface, NUKCLOUD multi-region bare-metal Mac / cloud Mac nodes offer dedicated Apple Silicon and clear tenant boundaries — compare specs on the pricing page and trial via the order page.

Sources as of 2026-08-06: Caijing (via Sina Finance / Wall Street CN); The Standard (HK), Gate News, ChainCatcher; Forbes “DeepSeek Just Raised $7.4 Billion. Here's The Catch.”; SCMP, Caixin Global, Reuters, Bloomberg; CIW on cap-table structure; DeepSeek API docs, TechCrunch, Hugging Face on V4; 36Kr / TMTPost peer comps. Most figures are anonymously sourced media reports — verify before republishing.