DeepSeek: Analysis of the 7.4 Billion Raise and Liang Wenfeng's Strategy
- DeepSeek is seeking new capital to reach a valuation of 74 billion dollars, a strong increase compared to the 50 billion from the June round.
- Ultra-centralized governance: CEO Liang Wenfeng maintains total control through a Limited Partnership that deprives partners of voting rights.
- Critical dependence on Nvidia hardware and cost pressure: the startup has increased API rates to support the expansion of data centers and AI agents.
- Financial roadmap: the goal is listing on the Shanghai Star Market by 2027, supported by investors such as Tencent, CATL, and state funds.
DeepSeek, the AI entity born as a division of the hedge fund High-Flyer Capital Management, is undergoing a phase of unprecedented financial acceleration. After a first external round in June 2026, the company has resumed a massive capital raise that redefines its positioning in the global artificial intelligence market.
The obsession with control: Liang Wenfeng's unusual governance structure
DeepSeek's management is characterized by a strongly asymmetrical power structure, orchestrated by CEO and founder Liang Wenfeng to neutralize the influence of external investors. According to Chinabizinsider, citing a report from The Information, the June funding round did not see capital injected directly into DeepSeek, but rather through a Limited Partnership mechanism controlled by Liang.
The implications of this structure are radical:
- Absence of voting rights: All external investors, with one exception, do not possess voting rights within the company.
- Lockup period: Partnership participants are subject to a five-year lockup period, which prevents the sale of shares on the secondary market, an unusual practice in venture capital.
- Limited privileges: Investors have access to privileged financial disclosures and priority rights for future rounds, but no decision-making leverage.
Strategic Analysis: This configuration indicates that Liang Wenfeng does not consider external capital as a strategic partner, but as pure financial fuel. By protecting governance, Liang prevents pressures for short-term profit from interfering with the research-first approach that led to the success of the R1 model in 2025.
From 50 to 74 billion: the valuation surge between the June round and the current mega-raise
The evolution of DeepSeek's valuation in a few months highlights an exponential growth in the startup's perceived value. Cross-referenced data between Investire.biz and Chinatechnews show a significant quantitative jump:
| Period | Capital Raised / Goal | Company Valuation | Round Status |
|---|---|---|---|
| June 2026 | 7.4 billion USD (50 bln yuan) | Over 50 billion USD | Closed |
| August 2026 | 7.4 billion USD (50 bln yuan) | 74 billion USD (500 bln yuan) | Ongoing / Resuming |
The new round, initially suspended, is expected to be finalized by the end of the month with minimal public discretion. The valuation premium compared to June reflects the company's determination to scale operations and technological capabilities, despite operational challenges.
The billionaires club: who is financing DeepSeek's expansion among tech giants and state funds
DeepSeek's investor map reveals a strategic mix of high-profile private capital and Chinese government support. Chinabizinsider details the contributions of the debut round, highlighting Liang's role as the primary financier.
- Liang Wenfeng: 20 billion yuan (the largest single contributor).
- Tencent: 10 billion yuan.
- CATL (Contemporary Amperex Technology): 5 billion yuan.
- JD.com, NetEase, IDG Capital: 3 billion yuan each.
- National Artificial Intelligence Industry Investment Fund: 1 billion yuan.
It is fundamental to note the exception granted to the National Artificial Intelligence Industry Investment Fund: unlike the others, this state fund invests directly in DeepSeek, is exempt from the five-year lockup, and holds voting rights. Geopolitical Analysis: This detail suggests that, although Liang controls operations, the Chinese State maintains a direct decision-making anchor, ensuring that AI development is aligned with national interests.
The silicon trap: hardware dependencies and access to Nvidia GPUs
Behind the image of low-cost model efficiency, a critical vulnerability emerges. A transcript of a closed-door meeting by Liang Wenfeng, leaked online and reported by Chinatechnews, revealed DeepSeek's deep dependence on Nvidia hardware.
Key points emerging from the leaks include:
- Hardware preferences: Liang candidly admitted a preference for Nvidia chips and detailed purchasing strategies to bypass or manage restrictions.
- Supply risk: There are concrete concerns regarding access to restricted silicon (silicon subject to export restrictions).
- Priority investments: The new capital raise is largely destined for the acquisition of GPUs and the expansion of data centers to support the training of more complex models.
Business Analysis: DeepSeek finds itself in a paradox: while promoting algorithmic efficiency (models that require fewer resources), its industrial scalability depends entirely on a supply chain controlled by the United States. Any further tightening of GPU export controls could compromise the company's technological roadmap.
Low-cost models or profit strategy: analysis of API rate increases
DeepSeek built its global reputation by offering high-efficiency AI models at contained costs. However, recent data indicates a change in course. Chinatechnews reports that the company has announced a second consecutive increase in API rates.
The comparison between the previous and current approach highlights a strategic tension:
- Low-Cost Strategy (Past): Focus on market penetration and demonstrating the technical superiority of the R1 model. Goal: rapid acquisition of users and integrations.
- Profit/Sustainability Strategy (Present): Rate increases to offset rising computing costs and aggressive competition for talent acquisition (which has already led to talent attrition phenomena).
'The pricing adjustments reflect the company's evolving business model and competitive positioning in the AI market'
Analysis: The price increase suggests that algorithmic efficiency is not sufficient to cover the operational costs of a global scale. DeepSeek is moving from a phase of growth at all costs to one of financial sustainability, necessary to attract investors in view of the listing.
Toward the Shanghai Star Market: financial milestones for the 2027 listing
The ultimate goal of Liang Wenfeng and his investors is entry into the public market. According to Investire.biz, the capital raising strategy is calibrated to culminate in an IPO on the Shanghai Star Market (the market dedicated to tech and innovative companies in China).
Expected timeline of events:
- June 2026: Closing of the first external round (7.4 bln USD) to stabilize the financial base.
- August 2026: Mega-raise at a valuation of 74 bln USD to finance data centers and AI agents.
- 2026-2027: Phase of offer diversification (expansion into AI agents and new services) and margin optimization via API.
- 2027: Expected listing on the Shanghai Star Market.
Verifiable Indicator: The success of this timeline will depend on DeepSeek's ability to maintain API revenue growth despite the price increases. A key signal will be the publication of the first quarterly financial reports showing the ratio between computing costs and API revenues.
The challenge of Chinese low-cost models: impact for EU companies and the AI Act
The rise of DeepSeek and its ability to produce high-performance models at reduced integration costs represent both an opportunity and a risk for the European market.
Impact on integration costs for EU companies
For Italian and European companies, access to low-cost Chinese AI models could accelerate the adoption of generative AI in low-margin sectors. However, price instability (as demonstrated by DeepSeek's recent increases) introduces a risk of vendor lock-in with unpredictable variable costs.
Compliance and AI Act
The integration of models like those from DeepSeek raises critical compliance issues:
- Transparency: The EU AI Act requires rigorous transparency standards on training data and risk management. The closed nature of Liang Wenfeng's governance and the poor public disclosure could make it difficult for EU companies to certify the compliance of systems based on DeepSeek.
- Data sovereignty and NIS2: The use of Chinese APIs for critical business processes could collide with NIS2 directives on the resilience of network and information systems, especially regarding data localization and software supply chain security.
Summary for the Italian entrepreneur: The adoption of DeepSeek offers a competitive advantage in terms of immediate costs, but requires a rigorous analysis of legal risk (AI Act) and strategic risk (dependence on hardware/software subject to geopolitical tensions). The indicator to watch will be the eventual release of EU-compliant versions or the opening of data centers on European territory.
FAQ
Why did DeepSeek's valuation go from 50 to 74 billion dollars in such a short time?
The surge reflects the company's determination to rapidly scale operations, the expansion toward data centers and AI agents, and the attractiveness for new investors despite governance restrictions.
What is Liang Wenfeng's role in capital management?
Liang maintains almost absolute control through a Limited Partnership; external investors have no voting rights and are subject to a five-year lockup, with the exception of the Chinese state fund.
What is DeepSeek's main operational risk?
The critical dependence on Nvidia hardware and limited access to restricted silicon, as emerged from leaked internal documents.
When is DeepSeek's stock market listing expected?
The strategic goal is listing on the Shanghai Star Market by 2027.
Sources: Investire, Chinatechnews, Chinabizinsider · by glacom.news AI
Scrivila qui: Susanna, l assistente AI di glacom, ti risponde via email con un approfondimento gratuito.
Nessuna consulenza personalizzata (finanziaria, legale o medica): solo informazione e fonti. Email usata solo per rispondere.
.png)
