Tencent and Alibaba circle DeepSeek as funding talks reach $20B valuation

What happened

The Information reported on April 22, 2026 that Tencent and Alibaba are in talks to invest in Hangzhou-based DeepSeek at a valuation exceeding US$20 billion ¹. Bloomberg corroborated the talks, noting that DeepSeek's founder Liang Wenfeng remains the controlling shareholder and that the round is being led through hedge fund High-Flyer Capital Management ². Earlier in 2026, DeepSeek published research papers signalling a push to train larger models at lower cost using reinforcement-learning innovations and a refined mixture-of-experts architecture ³. Its widely anticipated V4 multimodal model was released around the March 4 "Two Sessions" parliamentary period ⁴.

What it means

A $20 billion valuation would place DeepSeek roughly an order of magnitude below Anthropic and OpenAI, but well above where any independent Chinese AI lab has previously priced. The unusual feature is that both Tencent and Alibaba, who compete fiercely in cloud and consumer AI, are reportedly bidding into the same round. That pattern suggests neither hyperscaler is willing to let the other secure exclusive distribution, and that DeepSeek's leadership is using competitive tension to preserve independence.

For international entrepreneurs, the round matters because it signals that China's open-weight model ecosystem now has a credible commercial cap-table, not just academic-style releases. DeepSeek's models have been downloaded heavily by developers in Southeast Asia, the Middle East, and Latin America, where US export-control friction makes US frontier models more expensive or operationally awkward. A funded DeepSeek with cloud-distribution support from Tencent and Alibaba could become the default open-weight backbone outside the OECD core.

Historical parallel

The closest parallel is Yandex's mid-2010s positioning in Russia and the CIS, where it served as a regional alternative to Google in search and AI without trying to compete head-on in the US market ⁵. A more recent and more directly relevant comparison is the 2017 Tencent-Alibaba duelling investments in bike-sharing rivals Mobike and Ofo, where the two giants forced rival fundraising rounds at escalating valuations ⁶. That cycle ended badly for both targets, but it illustrates the pattern of Chinese hyperscalers willing to fund a strategic asset they cannot acquire outright. A more constructive parallel is Mistral AI's 2024 fundraises in France, where strategic backing from Microsoft and Nvidia helped a non-US lab build a credible open-weight alternative without ceding control ⁷.

What to watch next

• Whether the round closes with both Tencent and Alibaba on the cap-table, or only one (signalling a strategic alignment).
• DeepSeek's planned successor to V4, particularly any R2 reasoning model, and whether training shifts back to Nvidia GPUs from Huawei Ascend hardware.
• Adoption metrics in Southeast Asia and Middle East cloud markets, where DeepSeek competes with Llama and Qwen.
• Whether the US Treasury's Outbound Investment Security Program adds DeepSeek to a restricted-investment list, which would block US LP capital.
• Any IP or export-control action by the US government targeting open-weight Chinese models.

Implications for global entrepreneurs

For founders building AI products outside the US-China core, the emergence of a well-capitalised DeepSeek expands the menu of frontier-grade backbones meaningfully. A founder in Toronto or London considering whether to build on Claude, GPT, Gemini, Llama, Qwen, or DeepSeek now faces a real procurement question rather than a defaulted choice. Open-weight models in particular reduce single-vendor risk, but introduce new compliance work: any company processing EU or Canadian regulated data on Chinese-origin weights will face heightened scrutiny under data-residency and supply-chain rules.

For Chinese diaspora founders, the round also signals that the Beijing policy environment is willing to let private capital lead AI scaling, rather than forcing consolidation under state-owned vehicles. That is a meaningful contrast to the 2021-2023 platform-economy crackdown and may persuade some founders who left for Singapore or Vancouver to consider partial repatriation of R&D activity.

Risks and caveats

The reporting comes from a single primary source (The Information) with limited corroborating disclosure from DeepSeek itself. Valuation figures in private Chinese rounds historically have been negotiated downward at close. There is also a non-trivial risk that US authorities expand entity-list coverage to DeepSeek or to Tencent and Alibaba's cloud arms in ways that would force cap-table restructuring. Finally, DeepSeek's rumoured plan to retain Huawei Ascend hardware for at least some training runs continues to expose the company to chip-software immaturity that delayed its 2025 model roadmap.

---

Sources