Liang Wenfeng’s Vision Meets Reality

Jiang Ji, The Investors Network

Last weekend, flights to meet Liang Wenfeng were fully booked across the global AI investment community, all with a singular purpose: to secure a meeting with him.

DeepSeek (Deep Exploration), a company that once firmly declared it could not “agree with venture capitalists (VCs) on fundamental issues,” relied solely on profits from its quant trading subsidiary, High-Flyer, for funding and steadfastly refused external capital. However, it has now reportedly opened its first external funding round since its inception in July 2023. With a target valuation of no less than US$10 billion, the company plans to raise at least US$300 million. The technical idealist who once refused to even answer calls from investors has finally opened the door to capital.

Is this a sign of compromise or part of a grander strategy? The answer may lie in the impending launch of the V4 model.

US$10 Billion Valuation: A Benchmark for Primary Market Pricing

What does a US$10 billion valuation signify? For context, Zhipu AI, the “first large model stock” listed on the Hong Kong Stock Exchange in January 2026, saw its market capitalization soar past HK$52.8 billion (approximately US$6.8 billion) on its debut day. MiniMax, which followed shortly after, now boasts a market cap exceeding HK$230 billion (approximately US$29.5 billion). Despite being unprofitable, having yet to release its V4 model, and lacking any commercialization data, DeepSeek has set a private market valuation that falls between these two benchmarks.

This pricing reflects a shift in capital logic within the AI infrastructure race. OpenAI secured massive funding at a valuation exceeding US$850 billion, while Anthropic stabilized at US$380 billion. In this fiercely competitive landscape, capital has moved beyond traditional price-to-earnings (P/E) ratio assessments, focusing instead on “computing power reserves + technological generational gaps + ecosystem positioning.” The US$10 billion valuation represents investors’ confidence in DeepSeek’s technological foundation and their assessment of its potential to advance to the next stage of competition.

However, a more critical question remains: With High-Flyer (managing over RMB 70 billion in assets and generating nearly RMB 4.9 billion in revenue in 2025) as a strong financial backer, why does Liang Wenfeng need external funding now?

Behind the V4 Delay: From Efficiency Myth to Computing Power Gap

In 2025, DeepSeek stunned the industry with its R1 model, developed at extremely low training costs, establishing an “efficiency myth.” However, by 2026, the rules of the game had completely changed.

The V4 model, originally scheduled for release around the Spring Festival, is now confirmed for a late-April launch. Its parameter specifications are expected to undergo a comprehensive upgrade: approximately 1 trillion parameters (using a Mixture of Experts, or MoE, architecture), 1 million token context, and native support for multimodality. If achieved, it could rival the world’s top large models.

The delay stems not from technical weakness but from an exponential surge in engineering complexity and computing power costs. First, V4 represents a systemic overhaul rather than mere parameter stacking. Second, DeepSeek has chosen a challenging path: deeply adapting to Huawei’s Ascend domestic chip ecosystem, migrating core code from CUDA to the CANN architecture. This migration entails significant costs for precision alignment and cross-chip debugging. Finally, and most crucially—money. Training and deploying a trillion-parameter model has escalated computing power investments from the millions to the billions of dollars. Even with High-Flyer’s profitability, sustaining such bottomless spending independently is untenable. External capital has become essential to bridge this computing power gap.

Ideals Can’t Pay the Bills; Equity Needs Pricing

More urgent than computing power is the organizational “bleeding.”

Since the R1 model’s breakthrough success, DeepSeek’s core team has experienced significant turnover. Market rumors frequently cite Luo Fuli, a key contributor to the V3 architecture, joining Xiaomi; Wang Bingxuan, a core author of the initial large model, moving to Tencent; Ruan Chong, a multimodal core member, becoming chief scientist at Yuanrong Qixing; Wei Haoran, an OCR core member, departing; and Guo Daya, born in 1994, being poached by ByteDance to lead Agent development.

Why can’t they retain talent? At this stage, lofty ideals pale in comparison to hard cash. DeepSeek has never raised external funding, lacks market-based valuation benchmarks, and employees’ equity commitments suffer from poor liquidity and unclear value. When giants like ByteDance and Xiaomi offer highly attractive packages, DeepSeek, lacking capitalization channels, is at a disadvantage in the talent war.

The answer is clear: Initiating external financing establishes the company’s fair value and creates a market-driven equity incentive pool. Without funding, core talent will continue to depart; with funding, technical ideals can transform into organizational combat effectiveness.

From Efficiency Myth to Capital Warfare: The Endgame Logic of the AI Sector

DeepSeek’s maiden funding round marks a new phase in China’s large model competition.

Over the past two years, it disrupted industry blind faith in “mere computing power stacking” through open-source strategies and extreme cost-effectiveness. However, by 2026, the industry landscape has shifted toward “head clustering and polarized hot-and-cold dynamics.” Despite potential fluctuations in the overall funding environment, capital is frenziedly concentrating at the top. For example, Jieyue Xingchen secured over RMB 5 billion in a Series B+ round in January 2026, setting a domestic record. After Zhipu AI and MiniMax listed on the Hong Kong Stock Exchange, their combined market cap exceeded HK$480 billion.

Computing power demands burning money, talent commands high salaries, and ecosystem positioning is critical. Without sustained capital infusions, even the strongest technical models may remain trapped in the lab. For Liang Wenfeng, answering the capital deluge’s calls is not mere compromise but a necessary choice to proactively connect to the commercial world’s power grid and engage in all-out competition.

The US$10 billion valuation is just the starting point. Whether the V4 model, scheduled for late April, meets expectations, whether Huawei Ascend’s deep adaptation succeeds, and whether commercialization loops can be closed will determine DeepSeek’s ability to survive the next industry shakeout. The ultimate AI competition has never been a solo fight for technical geniuses but a comprehensive battalion battle integrating capital, computing power, talent, and ecosystems. The moment Liang Wenfeng pushed open the capital gate, DeepSeek graduated from its “efficiency myth” phase and embarked on a far crueler and real “commercial long march.” What do you think? Discuss in the comments (Produced by Siwei Finance)

Source: The Investors Network