DeepSeek V4 | $20B Valuation, Tencent Funding Talks

1. First Outside Capital | $300M at Under 3% Equity

DeepSeek is seeking to raise at least $300 million while selling no more than 3 percent of its equity, according to the South China Morning Post, citing investors with direct knowledge of the plans. The deal marks a turning point for the Hangzhou-based company, which has never accepted outside capital since its founding in 2023 by Liang Wenfeng's hedge fund, High-Flyer Capital Management.

The fundraising structure signals that DeepSeek intends to retain near-total control. A 3 percent equity ceiling at a $20 billion valuation implies the company is entering external markets for liquidity and benchmark purposes, not for operational capital. No deal has been finalized, and none of the companies involved have commented publicly as of April 25, 2026.

BY THE NUMBERS

$300M+

Target Raise

Max Equity Sold

$20B+

Valuation

2. Tencent, Alibaba, State Funds | Who Is Bidding

The Information first reported on April 17 that DeepSeek was raising outside money at a $10 billion-plus valuation. By April 22, the outlet reported the target had jumped above $20 billion, with Tencent proposing to acquire as much as a 20 percent stake. DeepSeek has resisted ceding that level of control, according to Bloomberg. Alibaba is also negotiating, though details of its offer have not been disclosed.

Tencent

Proposed 20% stake, highest disclosed offer

DeepSeek has resisted the level of control this would cede, per Bloomberg. Tencent operates WeChat and is China's largest gaming company.

Alibaba

Actively negotiating, terms undisclosed

Alibaba's cloud division competes directly with DeepSeek's model capabilities. Its participation is confirmed but deal size remains private. See ObjectWire's Alibaba AI coverage for context.

Big Fund III Affiliates

State-backed participation expected

China's National Integrated Circuit Industry Investment Fund III affiliates are expected to join the round, per South China Morning Post.

Additional Investors

Multiple unnamed parties beyond the tech giants

The Wall Street Journal reported that DeepSeek has been in discussions with multiple prospective investors, with funding expected to be yuan-denominated.

Yuan Denomination

The Wall Street Journal reports the funding round is expected to be denominated in Chinese yuan, consistent with a domestic capital structure and limiting exposure to US dollar-linked regulatory scrutiny.

Alibaba's stake in this round comes as the company has been aggressively expanding its own AI capabilities. ObjectWire has covered Alibaba's parallel AI push: its Qwen 3.6 model preview and developer SDK rollout, and an unusual incident in which an Alibaba AI agent autonomously mined cryptocurrency during a training run. For a full company profile, see the ObjectWire Alibaba hub.

3. 5 Researcher Departures | The Talent War Driving the Round

Between late 2025 and April 2026, at least five core R&D members left DeepSeek. The most prominent departure was Guo Daya, a lead researcher on the R1 model, who joined ByteDance's Seed AI team. Chinese outlet LatePost reported Guo's annual compensation at ByteDance at as high as 100 million yuan ($14.7 million).

Why the Round Helps Retention

The round is partly intended to establish a market benchmark for DeepSeek's valuation, giving employees greater clarity on the worth of their stock options. Without a public valuation benchmark, employees holding equity cannot assess the real-world value of their compensation, which weakens retention when competitors offer immediate cash. Establishing a $20B+ floor changes that calculus directly.

The retention risk is acute because DeepSeek's competitive advantage is concentrated in a small team of frontier researchers. The R1 model, which upended Western assumptions about AI development cost curves in January 2025, was built by a lean group. Losing further members to ByteDance Seed, Baidu, or international labs represents a structural threat to the company's model-quality roadmap.

Read more on the broader AI talent competition on ObjectWire Tech and the OpenAI coverage hub.

4. DeepSeek V4 | 1.6 Trillion Parameters, Open Source

On the same week as funding talks intensified, DeepSeek released preview versions of its V4 model on Friday, April 25. V4 is a 1.6 trillion-parameter open-source system. The company says it rivals leading American models in three core capability areas: coding, reasoning, and agentic tasks.

V4 vs. the Competition

According to CNN, DeepSeek stated that V4 surpasses other open-source models in world knowledge benchmarks but acknowledged the model still trails leading closed-source American systems in certain evaluations. This positions V4 as the strongest open-weight frontier model available, while stopping short of claiming parity with GPT-4 class closed systems.

Parameter Count

1.6 Trillion

Largest disclosed model size in DeepSeek's lineup. Likely a Mixture-of-Experts architecture, consistent with V3 and R1 design patterns.

Release Format

Open Source Preview

V4 ships as an open-weight model, continuing DeepSeek's strategy of releasing weights publicly, which accelerates external adoption and benchmarking.

Hardware

Huawei Ascend Chips

V4 training ran on Huawei Ascend processors, per earlier reporting by ObjectWire. DeepSeek deliberately withheld access from Nvidia and AMD amid export controls.

Top Capabilities

Coding, Reasoning, Agentic Tasks

DeepSeek claims V4 rivals leading American models in these three domains, with stated superiority over open-source alternatives in world knowledge.

For the full technical breakdown of V4's hardware architecture and training timeline, see the earlier ObjectWire report: DeepSeek V4 Launches Late April on Huawei Ascend Chips.

5. Strategic Context | Why This Moment Matters for China AI

DeepSeek's decision to raise outside capital now, alongside a flagship model release, is a deliberate signal to the global AI market. The company has operated as a research lab within a hedge fund structure since 2023, which shielded it from external pressure but also left its valuation opaque. A structured fundraise with a disclosed $20B+ benchmark changes its standing in international AI rankings overnight.

The round is partly intended to establish a market benchmark for DeepSeek's valuation, giving employees greater clarity on the worth of their stock options.
South China Morning Post, citing investors with direct knowledge, April 2026

The participation of state-backed funds alongside private tech giants reflects Beijing's continued prioritization of domestic frontier AI development. Following US export controls that restricted DeepSeek's access to Nvidia H100 and A100 chips, the company's pivot to Huawei Ascend hardware — and its success in training V4 on domestic silicon — carries policy significance well beyond the commercial round.

For coverage of the chip supply dimension, see ObjectWire's Nvidia hub. For the broader AI race context, see ObjectWire's OpenAI hub.

Sources & References

[1] The Information: DeepSeek Raises Outside Money at $10B+ Valuation — First report of DeepSeek's outside fundraise, April 17 2026.
[2] South China Morning Post: DeepSeek Seeks $300M, Selling Under 3% Equity — Details on deal structure, yuan denomination, and Big Fund III participation.
[3] Bloomberg: DeepSeek Resists Tencent's 20% Stake Proposal — DeepSeek pushes back on ceding control to Tencent.
[4] The Wall Street Journal: DeepSeek Funding Expected in Chinese Yuan — Deal structure and broader investor base details.
[5] CNN: DeepSeek V4 Surpasses Open-Source Models in World Knowledge — V4 launch coverage including benchmark comparisons.