DeepSeek reportedly appoints 90s investor Yan Wentao as CFO, prepares for Shanghai IPO

DeepSeek is reportedly on the verge of appointing a new Chief Financial Officer — Yan Wentao, a post-90s generation investment veteran — according to recent market chatter. The news has sparked considerable discussion, given that founder Liang Wenfeng is already within striking distance of becoming China's richest person. The addition of a finance chief with deep industry and capital market expertise has fueled fresh speculation about the company's future trajectory.

From Quantitative Trading to AI: A Calculated Transition

Liang Wenfeng did not emerge as a founder with an obvious halo. Born into an ordinary family, his standout trait was always mathematical aptitude. At Wuchuan No.1 Middle School in Guangdong, he was a top student. By junior high, he had already completed high school mathematics and begun exploring university-level math. Yet he was no mere "study machine" — he balanced work and play with high efficiency.

In 2002, at age 17, Liang achieved the top理科 (science) score in the college entrance exam at his school and enrolled in Zhejiang University's Electronic Information Engineering program. His homeroom teacher had suggested Tsinghua University, but Liang chose to follow his own interests — a pattern that would define his career.

At Zhejiang University, Liang continued to deepen his technical expertise. In 2007, he entered the graduate program in Information and Communication Engineering, focusing on machine vision. A conventional path would have led to a tech giant and a career in R&D. Liang chose otherwise.

In 2008, while still a graduate student, he began researching quantitative trading with senior student Xu Jin and others, aiming to use machine learning and algorithms to automate investment decisions. At the time, quantitative investing was far from mainstream; most investors relied on experience, judgment, and market intuition. The idea of letting computers make investment decisions seemed almost fantastical to many.

Liang was willing to experiment.

After earning his master's degree, he moved to Chengdu and continued refining quantitative models in a rented apartment. With the introduction of stock index futures, quantitative investing gained traction. Liang's team leveraged algorithmic advantages to optimize strategies, growing their own capital from tens of thousands of yuan to hundreds of millions — securing his first fortune.

But that was not the endgame.

His experience in quantitative trading taught Liang an early lesson: the future competition would not just be about data, but about computing power.

Quantitative trading is inherently dependent on algorithms, data, and computational resources. The more complex the model, the more information it processes, and the greater the demand for computing power.

This explains why High-Flyer Quant's massive investment in computing power was not a sudden pivot to "playing with AI," but a continuation of over a decade of strategic thinking.

• 2019: High-Flyer invested nearly 200 million yuan to build "Firefly No.1," deploying 1,100 GPUs.
• 2021: It invested nearly 1 billion yuan to build "Firefly No.2," using approximately 10,000 A100 GPUs.

At the time, many questioned why an investment firm would spend so heavily on chips. For Liang, computing power was not merely hardware expenditure — it was an integral part of research capability.

When the large language model wave arrived in 2023, Liang's shift into AI was therefore far from abrupt. That year, ChatGPT exploded globally, and companies rushed to compete — raising funds, expanding teams, and scrambling for GPUs. DeepSeek, by contrast, maintained a relatively low profile.

Liang did not rush to scale the company. Instead, he focused on model development and technical optimization.

In 2025, DeepSeek-R1 garnered attention upon release. Observers noted that it did not simply follow the "big investment, massive computing power" route. Instead, it explored new approaches to algorithmic efficiency and training methods.

Traditional large model training typically involves providing the model with extensive demonstrations — teaching it human-summarized problem-solving steps. Liang took a different approach:

Instead of telling the model exactly what to do at each step, he let it explore solutions on its own. Correct answers earned rewards; wrong answers prompted adjustments. Through trial and error, the model gradually learned to check its own answers and spend more time reasoning through complex problems.

In simple terms, rather than relying entirely on humans to teach "problem-solving methods" step by step, the model learned to find solutions through training.

This approach also had a notable cost advantage. DeepSeek disclosed in its paper that the total cost for the post-training phase of the R1 series was only approximately $294,000 — a figure that remains a talking point in the AI industry.

For years, the large model industry had been escalating computing power investments — larger models, more GPUs, bigger data centers. Often, the competition boiled down to who could secure more chips, more data, and more funding.

Liang's message was different: the scale of computing power matters, but how you use it matters just as much.

From this perspective, his journey from quantitative trading to AI carries a certain inevitability. The difference is that he now faces a larger, faster-changing AI industry. The next challenge for DeepSeek is to continue validating its technical approach and bring its models to more real-world applications.

Why Liang Wenfeng, Who Didn't Need Money, Is Now Turning to Capital

DeepSeek's valuation has now reached hundreds of billions of yuan. Liang Wenfeng holds approximately 84.29% of the company's equity (direct and indirect combined). If the company's market cap reaches 1 trillion yuan, his paper wealth could surpass that of Zhang Yiming, making him China's richest person.

Of course, paper wealth is just that — paper. But the shift behind these numbers is noteworthy: Liang has moved from his "three no's" principle (no fundraising, no commercialization, no roadshows) toward embracing capital.

In its early days, DeepSeek had little need for capital markets. By the end of 2025, Liang had achieved an average annual return of 114.35% over five years at High-Flyer, giving him ample confidence to pursue research at his own pace. For a young tech company, this was a comfortable position.

But AI model competition requires sustained investment in computing power, talent, and infrastructure. Relying solely on Liang's personal funds was no longer sufficient. Fundraising became inevitable.

• June 2026: DeepSeek completed its first external fundraising round since inception, raising approximately 51 billion yuan at a post-investment valuation of about 338 billion yuan. Liang personally committed 20 billion yuan. Tencent, CATL, and "national team" investors also participated.
• Market rumors later suggested DeepSeek planned a new round at a valuation of around 500 billion yuan, though this round was reportedly delayed and has not yet been finalized.

In September, foreign media reported that DeepSeek had hired CITIC Securities to prepare for a STAR Market IPO, with plans to file within the year.

Typically, founders set the direction, investors provide capital, and both work together to grow the company. Liang's approach is different: external capital is welcome, but he continues to invest personally, and the company's technical roadmap and R&D pace will not fundamentally change due to fundraising.

In other words, Liang needs capital — but he does not intend to let capital dictate the company's direction.

However, the AI industry has an unavoidable problem: burn rate.

In the first seven months of this year, Liang spent approximately 11 billion yuan on AI infrastructure, including leasing AI chip servers and purchasing computing equipment. In all of 2025, that figure was only about 1.2 billion yuan. In just over a year, spending surged nearly tenfold. R&D, meanwhile, is not a matter of investing today and seeing results tomorrow; many projects may take years to bear fruit.

As the company grows, how to allocate funds has become an unavoidable question for Liang. Should he continue to invest in computing power, or expand the team? Should he shore up infrastructure, or reserve funds for future R&D? In the past, Liang focused on technology, R&D, and efficiency. But no one person can handle everything.

Liang can still tightly control technical direction and R&D pace. But financial management, capital planning, and other capital-market-facing tasks require professional expertise.

Against this backdrop, a previously low-profile position has become critical: CFO.

On the Path to Becoming China's Richest Man, Liang Needs a Capital Markets Expert

If Yan Wentao joins DeepSeek, it signals that the company is filling a gap in capital management and corporate governance.

Born in 1991, Yan graduated from Fudan University. He previously worked at Tencent Investment and H Capital before joining Hillhouse Capital's venture arm in 2020, becoming a representative figure among the firm's younger generation of investors. In 2025, he was named to the Nova New Investor TOP50 by ChinaVenture.

Foreign media have described Yan as a "dealmaker" — someone skilled at identifying opportunities and executing transactions. He has publicly stated that the core of tech venture capital is finding the era's biggest technological challenge and the people capable of solving it.

Yan's focus on AI is not recent. He began researching large language models in 2020 and shifted his attention to embodied intelligence in 2023. In his view, the window of opportunity for investors in each technological wave is very short — the key is identifying the companies and founders that can seize the moment.

Notably, two of his portfolio companies — MiniMax and ZhiPu AI — have both initiated STAR Market IPO processes, signing advisory agreements in May and February of this year, respectively. This means Yan is well-versed in the full lifecycle of AI large model companies, from early growth to capital market readiness.

It is worth noting that Hillhouse is not an investor in DeepSeek. Therefore, if Yan ultimately joins DeepSeek, it would not be a traditional case of an investor "betting on a project." Rather, the two sides likely share a common understanding of tech industry trends and capital market logic.

DeepSeek has been searching for a CFO for some time.

• February 2025: The company publicly posted a CFO job listing.
• June 2026: DeepSeek announced plans to at least double the size of all departments. The finance team was actively recruiting, with CFO remaining a core position.

However, DeepSeek's needs clearly go beyond a traditional financial controller.

According to the company's official job requirements, the CFO must be familiar with accounting standards and tax policies, hold a CPA qualification (preferred), and be capable of establishing a comprehensive compliance and risk control system. Yan's strength, however, lies not in financial accounting expertise, but in his long-term involvement in tech investing and his deep understanding of AI industry development and capital market operations.

In a sense, he would serve as a "translator" for DeepSeek — converting technical advantages into a business logic that capital markets can understand, and transforming a technology-driven company's potential into a clearer development roadmap.

Over the past year, as DeepSeek has rapidly advanced in AI, Liang Wenfeng's personal wealth has soared — from approximately $1 billion to $39.9 billion, ranking 51st globally according to Forbes.

But rising wealth figures are only the surface. Whether Liang can ultimately reach the heights of "China's richest person" depends on whether DeepSeek can truly convert its technical advantages into commercial value.

For now, at least, he appears to have found a key partner who can help bridge capital, management, and industry for the company's next stage of growth.

Disclaimer: The above content (including any images or videos) is uploaded and posted by a user of NetEase Hao, a social media platform that provides information storage services only.