# Liang Wenfeng

> Source: https://aiwiki.ai/wiki/liang_wenfeng
> Updated: 2026-09-23
> Fact-checked: 2026-09-23
> Categories: Chinese AI, People
> License: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/) - attribute to "AI Wiki (aiwiki.ai)"
> Cite as: AI Wiki. "Liang Wenfeng." aiwiki.ai, 23 Sept 2026. https://aiwiki.ai/wiki/liang_wenfeng
> From AI Wiki (https://aiwiki.ai), the free encyclopedia of artificial intelligence. Reuse freely with attribution.

**Liang Wenfeng** (Chinese: 梁文锋; born 1985) is a Chinese entrepreneur and engineer who founded the AI company [DeepSeek](https://aiwiki.ai/wiki/deepseek) in 2023 and co-founded the quantitative hedge fund [High-Flyer](https://aiwiki.ai/wiki/high_flyer) (Huanfang Quantitative) in 2015. Under his leadership, DeepSeek released the [DeepSeek-V3](https://aiwiki.ai/wiki/deepseek_v3) and [DeepSeek-R1](https://aiwiki.ai/wiki/deepseek_r1) models in late 2024 and early 2025. R1's release in January 2025 was followed by a technology sell-off in which [Nvidia](https://aiwiki.ai/wiki/nvidia) lost close to $600 billion of market value in a single day, the largest one-day loss for any company in U.S. history. [7][12]

In the family-name-first convention of Chinese names, his surname is Liang and his given name is Wenfeng. He is known for his low public profile, for arguing that [artificial general intelligence](https://aiwiki.ai/wiki/agi) (AGI) requires long-term foundational research rather than short-term commercial pursuits, and for releasing DeepSeek's models with [open weights](https://aiwiki.ai/wiki/open_source_ai). Forbes put his fortune at $11.5 billion when he debuted on its list of China's 100 richest people in November 2025. After DeepSeek closed its first outside funding round in mid-2026, the Bloomberg Billionaires Index valued him at $36 billion, making him the richest founder of an AI-model company. [28][29]

## Who is Liang Wenfeng and where did he grow up?

Liang Wenfeng was born in 1985 and grew up in the village of Mililing (米历岭村) in Wuchuan (吴川市), a small city administered by the prefecture-level city of Zhanjiang in Guangdong province, China. [7][19][49] Both of his parents were teachers at the village primary school. [19] He spent six years at Wuchuan No. 1 Middle School. [7] Residents of his hometown told the Financial Times that he had been a top student who read comic books and excelled at mathematics, and his junior-high homeroom teacher told reporters that he finished high-school mathematics while still in junior high and began studying university-level mathematics. [18][49]

In 2002, at the age of 17, Liang was admitted to [Zhejiang University](https://aiwiki.ai/wiki/zhejiang_university) in Hangzhou, where he studied electronic information engineering; the South China Morning Post reported that he had achieved the highest score in the Zhanjiang region on the college entrance examination (gaokao). [1][7] In 2007 he entered the university's master's program in information and communication engineering, supervised by Xiang Zhiyu and focused on machine vision, and he graduated in June 2010. [1][7] Phoenix Tech reported the title of his master's thesis as "基于低成本PTZ摄像机的目标跟踪算法研究" (Research on target-tracking algorithms based on low-cost PTZ cameras); it dealt with object tracking for pan-tilt-zoom cameras, an application of [computer vision](https://aiwiki.ai/wiki/computer_vision). [1][7] With Xiang he also published a paper on robust object tracking with PTZ cameras in the Journal of Zhejiang University (Engineering Science) in 2011, whose author note describes Liang, born in 1985 and from Zhanjiang, as a master's student working on machine vision. [3]

## How did Liang Wenfeng start in quantitative finance?

During the 2008 global financial crisis, while still a graduate student, Liang formed a team with classmates to collect financial market data and explore fully automated quantitative trading using [machine learning](https://aiwiki.ai/wiki/machine_learning). [1][6] High-Flyer's own history dates its founding team's exploration of fully automated trading to 2008-2014. [2] After graduating he moved to Chengdu, where, according to the 2023 Waves article, he tried applying AI to various fields without success until he turned to finance; the South China Morning Post reported that he worked there on a project using algorithms for stock trading. [7][16] According to the work histories filed with the Asset Management Association of China (AMAC), he was general manager of Hangzhou Yakebi Investment Management (杭州雅克比投资管理有限公司) from September 2013 to June 2015, and Xu Jin, a fellow Zhejiang University alumnus, was a partner there from April 2014 to May 2015. [42][45] In 2015 the two set up Hangzhou Huanfang Technology Co., Ltd. (杭州幻方科技有限公司), since renamed Zhejiang Jiuzhang Asset Management; the AMAC register gives its establishment date as 11 June 2015. [42][44]

## What is High-Flyer (Huanfang Quantitative)?

High-Flyer is the AI-driven quantitative hedge fund that Liang co-founded with two college classmates and that later incubated DeepSeek. [27] Its origins trace to 2015, when the founders established Hangzhou Huanfang Technology Co., Ltd., and in February 2016 they formally incorporated Ningbo High-Flyer Quantitative Investment Management Partnership (Limited Partnership), doing business as High-Flyer. [2][42][43][44] The company, also known by its Chinese name Huanfang (幻方), is headquartered in Hangzhou, Zhejiang. It built its trading around machine learning, and by 2021, about six years after its founding, it had passed 100 billion yuan in assets under management and was counted among the "Four Kings" of Chinese quantitative investing. [16]

| Year | Milestone |
|------|----------|
| 2015 | Hangzhou Huanfang Technology Co., Ltd. (now Zhejiang Jiuzhang Asset Management) established in June |
| 2016 | Ningbo High-Flyer Quantitative Investment Management Partnership incorporated (February) |
| 2019 | Assets under management surpass 10 billion yuan; High-Flyer AI research unit registered; Fire-Flyer 1 computing cluster built with 500 GPUs |
| 2020 | Fire-Flyer 1 put into use with 1,100 accelerator cards (total investment nearly 200 million yuan) |
| 2021 | Assets under management pass 100 billion yuan; Fire-Flyer 2 built with about 10,000 [Nvidia A100](https://aiwiki.ai/wiki/nvidia_a100) GPUs (about 1 billion yuan) |
| 2023 | High-Flyer launches [DeepSeek](https://aiwiki.ai/wiki/deepseek) as a separate organization focused on AGI research |
| 2025 | High-Flyer's funds return 56.6% for the year, ranking second among China's top-performing large hedge-fund firms |

By 2019, High-Flyer had accumulated over 10 billion yuan in assets under management and set up a dedicated AI research unit, High-Flyer AI, focused on AI algorithms and their basic applications. [2][6][16] The firm also began investing heavily in computing infrastructure. Liang later described the build-up as gradual: from a single graphics card in the early days to 100 cards in 2015, 1,000 in 2019, and then 10,000. [16]

## How did Liang Wenfeng build a private GPU supercomputer?

One of Liang's most consequential decisions was his early investment in GPU computing hardware. High-Flyer's first deep-learning training platform, Fire-Flyer 1 (萤火一号), cost nearly 200 million yuan. According to High-Flyer's own timeline, its AI team built the cluster in 2019 with 500 GPUs linked by a 200 Gbps network, and it went into use in 2020 equipped with 1,100 accelerator cards. [2] The 2023 Waves article dated the investment to 2019 and gave the 1,100-GPU figure. [16] This cluster was used for AI research for the firm's quantitative trading operations.

In 2021, Fire-Flyer 1 was retired and replaced by the much larger Fire-Flyer 2, which cost about 1 billion yuan. [2][16] A paper accepted at the SC24 supercomputing conference describes the system as about 10,000 PCIe A100 GPUs in roughly 1,250 compute nodes, with nearly 200 storage servers and a network built on 200 Gbps InfiniBand switches. It reports performance approximating Nvidia's DGX-A100 design at about half the cost and with 40% lower energy consumption. [11] The paper details a hardware-software co-design, including a custom all-reduce library called HFReduce and the 3FS file system, that let High-Flyer get high performance from PCIe-connected A100 GPUs rather than the more expensive SXM form factor. [11]

According to the Financial Times, Liang began buying thousands of Nvidia GPUs in 2021 as an AI side project that acquaintances regarded as a quirky hobby. [4] The timing of these purchases proved important. In October 2022, the United States imposed export controls that barred the sale of advanced AI chips, including the Nvidia A100 and [H100](https://aiwiki.ai/wiki/nvidia_h100), to China. [14] The 2023 Waves interview noted that High-Flyer had stockpiled 10,000 GPUs as early as 2021, and cited a Finance Eleven report that fewer than five companies in China owned more than 10,000 GPUs, High-Flyer being the only one that was not a major technology group. [16] This computing infrastructure became the foundation for DeepSeek's early research.

## How did Liang Wenfeng build DeepSeek?

In April 2023, High-Flyer announced that it would concentrate its resources on AI and set up a new, independent research organization to explore the nature of AGI. Yicai dated the announcement to the evening of 14 April, while the May 2023 Waves article gave 11 April. [46][16] In May the organization was named DeepSeek (深度求索), and its operating company was set up in July 2023. [16][47] It was registered as Hangzhou DeepSeek Artificial Intelligence Basic Technology Research Co., Ltd., is headquartered in Hangzhou, Zhejiang, and has Liang as its CEO. [49]

DeepSeek's founding reflected a strategic shift. While High-Flyer had used AI primarily as a tool for generating investment returns, DeepSeek was set up to do fundamental AI research aimed at AGI. Liang funded DeepSeek through High-Flyer. He said in 2023 that venture capital firms were hesitant to back research because they had exit requirements and wanted rapid commercialization. [16][30] The company did not take outside investment until 2026 (see below). [28][30]

The company recruited a team of researchers, many of them recent graduates from top Chinese universities. [8] The V3 technical report, published in December 2024, lists 150 contributors under research and engineering, 31 under data annotation, and 18 under business and compliance: 199 entries covering 197 people, since two names appear under two roles. [10]

## What AI models has DeepSeek released under Liang Wenfeng?

Under Liang's leadership, DeepSeek released a series of increasingly capable AI models, most of them with downloadable weights.

| Model | Release Date | Key Details |
|-------|-------------|-------------|
| [DeepSeek Coder](https://aiwiki.ai/wiki/deepseek_coder) | November 2023 | First model release; code-focused; repository code under the MIT License, models under the DeepSeek Model License, which permits commercial use [37] |
| [DeepSeek LLM](https://aiwiki.ai/wiki/deepseek_llm) (67B) | November 29, 2023 | General-purpose [large language model](https://aiwiki.ai/wiki/large_language_model); 67 billion parameters |
| [DeepSeek-V2](https://aiwiki.ai/wiki/deepseek_v2) | May 2024 | [Mixture of Experts](https://aiwiki.ai/wiki/mixture_of_experts) architecture with multi-head latent attention; 236B total, 21B active parameters [39] |
| DeepSeek-Coder-V2 | June 2024 | 236 billion parameters (21B active); 128K context window [38] |
| [DeepSeek-V3](https://aiwiki.ai/wiki/deepseek_v3) | December 26, 2024 | 671B parameters (37B active per token); trained on 14.8 trillion tokens; 2,048 H800 GPUs; about $5.6 million in rented GPU time for the final run |
| [DeepSeek-R1](https://aiwiki.ai/wiki/deepseek_r1) | January 20, 2025 | Reasoning model reported as comparable to [OpenAI o1](https://aiwiki.ai/wiki/openai_o-series); MIT License; open weights |
| DeepSeek-V3-0324 | March 2025 | Improved post-training with [reinforcement learning](https://aiwiki.ai/wiki/reinforcement_learning) techniques from R1 |
| DeepSeek-R1-0528 | May 2025 | Upgraded reasoning capabilities with more compute and advanced post-training |
| [DeepSeek V3.1](https://aiwiki.ai/wiki/deepseek_v3_1) / V3.2 | August to December 2025 | Hybrid thinking and non-thinking modes; successive API upgrades [33] |
| DeepSeek-V4-Flash / V4-Pro | April 24, 2026 (preview) | 284B (13B active) and 1.6T (49B active) parameter MoE models; 1-million-token context; MIT License [24][34] |
| DeepSeek-V4-Flash-0731 / V4-Pro GA | July 31 and August 13, 2026 | Official releases of the V4 models, with stronger agent capabilities and low, high, and max thinking-effort levels [33] |
| [DeepSeek V4.1-Flash](https://aiwiki.ai/wiki/deepseek_v4_1_flash) | September 10, 2026 | Natively multimodal model in a new architecture family; served as `deepseek-flash` [33] |

### What is DeepSeek-V3 and why was it cheap to train?

[DeepSeek-V3](https://aiwiki.ai/wiki/deepseek_v3), released on December 26, 2024, drew worldwide attention. It is a large [Mixture of Experts](https://aiwiki.ai/wiki/mixture_of_experts) (MoE) model containing 671 billion total parameters, of which 37 billion are activated for any given input token. The model was pre-trained on 14.8 trillion tokens. [10]

What made DeepSeek-V3 remarkable was its training efficiency. According to the technical report, the model was trained on a cluster of 2,048 [Nvidia H800](https://aiwiki.ai/wiki/nvidia_h800) GPUs (a variant of the H100 with reduced interconnect bandwidth, made for the Chinese market to comply with U.S. export restrictions). [10] The pre-training stage was completed in less than two months, consuming 2,664,000 GPU hours. Including context length extension (119,000 GPU hours) and post-training (5,000 GPU hours), the full training run came to 2,788,000 GPU hours. At an assumed rental price of $2 per GPU hour, this translates to $5.576 million. [10] The report attributes the efficiency to techniques including FP8 mixed-precision training and its MoE design. [10]

The $5.6 million figure covers only the official training run of DeepSeek-V3. The report states that it excludes the costs of prior research and ablation experiments on architectures, algorithms, and data. Because it is priced as rented GPU time, it also does not reflect hardware purchases or staff costs. [10]

### What is DeepSeek-R1?

On January 20, 2025, DeepSeek released [DeepSeek-R1](https://aiwiki.ai/wiki/deepseek_r1), a reasoning model built on the V3 base. DeepSeek reported that R1 achieved performance comparable to OpenAI-o1-1217 on reasoning tasks: 79.8% pass@1 on the American Invitational Mathematics Examination (AIME) 2024, 97.3% on [MATH](https://aiwiki.ai/wiki/math)-500, and a 2,029 rating on Codeforces. [35] These are DeepSeek's own measurements.

R1 was released under the permissive MIT License, meaning that anyone could download, modify, and deploy the model freely. [41] The DeepSeek chatbot app, powered by R1, reached No. 1 among free apps in the U.S. App Store and in 51 other countries by January 27, 2025, according to Appfigures data reported by TechCrunch. The same report said downloads had passed 2.6 million across Apple's and Google's stores. [20]

### What did DeepSeek release during Open Source Week?

In late February 2025, DeepSeek held an Open Source Week, releasing five code repositories over five days: FlashMLA, DeepEP, DeepGEMM, a set of optimized parallelism strategies, and the Fire-Flyer File System (3FS). [21] [FlashMLA](https://aiwiki.ai/wiki/flashmla) is a decoding kernel for [multi-head latent attention](https://aiwiki.ai/wiki/multi-head_latent_attention), [DeepEP](https://aiwiki.ai/wiki/deepep) a communication library for Mixture of Experts models, and [DeepGEMM](https://aiwiki.ai/wiki/deepgemm) an FP8 matrix-multiplication library. The releases showed that DeepSeek's work extended beyond model architecture into the AI infrastructure stack, from GPU kernels to distributed file systems.

## Why did DeepSeek cause a market crash in January 2025?

The combined impact of DeepSeek-V3 (December 2024) and DeepSeek-R1 (January 2025) triggered one of the sharpest market disruptions in the history of the technology sector. The episode is covered in more detail at [DeepSeek market crash (Jan 2025)](https://aiwiki.ai/wiki/deepseek_market_crash).

On Monday, January 27, 2025, one week after R1's release, a sell-off swept through U.S. technology stocks. [Nvidia](https://aiwiki.ai/wiki/nvidia) shares fell 17% to close at $118.58, erasing close to $600 billion in market capitalization in a single trading day, the biggest drop for any company on a single day in U.S. history. [12] [Broadcom](https://aiwiki.ai/wiki/broadcom) also fell 17%, and Oracle fell 14%. [12]

The sell-off was driven by concerns about increased competition and about future demand for AI computing. DeepSeek had said V3 took about two months and less than $6 million to build using reduced-capability Nvidia H800 chips, and analysts at Cantor described "great angst as to the impact for compute demand, and therefore, fears of peak spending on GPUs." [12] Venture capitalist Marc Andreessen wrote on social media that "DeepSeek-R1 is AI's Sputnik moment," comparing it to the Soviet Union's 1957 satellite launch. [13]

## How did U.S. chip export controls shape DeepSeek?

DeepSeek's achievements are closely tied to the broader context of U.S.-China technology competition and semiconductor export controls.

In October 2022, the U.S. Department of Commerce imposed export controls on advanced semiconductors destined for China, barring the export of Nvidia's A100 and H100 GPUs. [14] Nvidia responded with reduced-interconnect versions of its chips that could legally be sold in China, including the H800, which DeepSeek used to train V3. [10][15] In October 2023 the U.S. government tightened the performance thresholds to block the A800 and H800 as well. [15]

The semiconductor consulting firm [SemiAnalysis](https://aiwiki.ai/wiki/semianalysis), citing anonymous industry sources, wrote that DeepSeek had a total of 50,000 Hopper-generation GPUs (a category that includes H100s, H800s, and H20s). It said High-Flyer's and DeepSeek's computing stack included at least 10,000 H100s, 10,000 H800s, 30,000 H20s, and 10,000 A100s. [15] The reported H100s drew scrutiny because the H100 was never legally available in China. No public evidence has established how or whether DeepSeek obtained them. [15]

Whatever the exact hardware, DeepSeek's results under chip constraints led analysts to argue that efficient algorithms and software could partly offset limited access to the most advanced chips. That raised questions about whether chip restrictions alone could preserve a decisive U.S. lead. [14]

## What is Liang Wenfeng's philosophy and leadership style?

### Research over commercialization

Liang Wenfeng's approach to AI development differs sharply from the venture-capital-driven model prevalent in Silicon Valley. He has repeatedly said that DeepSeek's goal is not short-term profitability but the long-term pursuit of AGI. His most detailed public statements come from two long interviews with Waves (暗涌), a sub-brand of the Chinese technology outlet 36Kr, published in May 2023 and July 2024. [5][8][9][16] Asked in 2023 why a commercial company would pursue open-ended research, he said: [9][16]

> "It might be hard if we must find a commercial justification, because it's not cost-effective. From a business perspective, fundamental research has a very low return on investment."

He has argued that Chinese technology companies had grown accustomed to monetizing innovations developed elsewhere through applications, and that this was unsustainable. In 2024 he said China must "gradually transition from being a beneficiary to a contributor, rather than continuing to ride on the coattails of others," adding that over 30 years of the IT revolution, Chinese companies "barely participated in core tech innovation." [8]

Liang has also been candid about the gap between Chinese and American AI capabilities. In the 2024 interview he estimated that China's best models "likely require twice the compute power to match top global models," attributing this to gaps in model structure and training dynamics, with data efficiency also lagging. [8] He has argued that "the most enduringly profitable U.S. companies are tech giants built on long-term R&D." [8]

### Why does Liang Wenfeng open-source DeepSeek's models?

A defining feature of Liang's leadership is his commitment to releasing DeepSeek's models openly. "In disruptive tech, closed-source moats are fleeting," he told Waves in 2024, arguing that DeepSeek's real moat was its team's growth and accumulated know-how. [5][8] In the same interview he said DeepSeek would not go closed-source. [8] Bloomberg reported in July 2026 that he had pledged in at least one meeting with prospective investors to keep developing open-source models while pursuing AGI. [29]

Liang has said that "both AI and API services should be affordable and accessible to everyone," and that DeepSeek's pricing principle is "neither to sell at a loss nor to seek excessive profits." [8] The company's price cuts for DeepSeek-V2 in 2024 were followed by cuts from larger Chinese rivals. [8]

### What is DeepSeek's organizational culture?

DeepSeek's internal culture reflects Liang's research-oriented mindset. He has described the organization as "entirely bottom-up," saying that the company does not preassign roles and lets a natural division of labor emerge. [8] In 2023 he said that "innovation requires as little intervention and management as possible, giving everyone the space to explore and the freedom to make mistakes. Innovation often arises naturally, it's not something that can be deliberately planned or taught." [16]

Asked in 2024 about the "inscrutable wizards" behind DeepSeek's models, Liang said the team was made up of "fresh graduates from top universities, PhD candidates (even fourth- or fifth-year interns), and young talents with a few years of experience." [8] The flat structure and research-first approach have been credited as factors in DeepSeek's ability to produce influential work with a small team.

### Low-profile personality

Despite leading one of the most closely watched AI companies in the world, Liang has kept a low public profile. His most detailed public statements from DeepSeek's early years are the two Waves interviews. [8][16] This reticence contrasts with the high visibility of Western AI leaders such as [Sam Altman](https://aiwiki.ai/wiki/sam_altman) of OpenAI or [Elon Musk](https://aiwiki.ai/wiki/elon_musk) of [xAI](https://aiwiki.ai/wiki/xai). In July 2026, Bloomberg reported that Liang was frustrated by online reports about comments he had allegedly made to investors. The posts, which Bloomberg had not verified, cited a supposed meeting transcript in which he discussed a reliance on Nvidia chips for AI development and China's lag behind the United States. People familiar with the matter said the frustration contributed to DeepSeek pausing its second funding round. [29]

## What is Liang Wenfeng's relationship with the Chinese government?

### Meeting with Premier Li Qiang

On January 20, 2025, the same day that DeepSeek released its R1 model, Liang attended a symposium in Beijing hosted by Chinese Premier Li Qiang. [6] At the meeting, a select group of experts from technology, education, science, culture, health, and sports offered opinions and suggestions on a draft of the government work report, according to the state news agency Xinhua. [6]

The South China Morning Post described Liang's participation as putting a spotlight on "China's new face of AI" and as reflecting DeepSeek's growing prominence in the industry. [6] A year later, MiniMax founder [Yan Junjie](https://aiwiki.ai/wiki/yan_junjie) became the second founder of a foundation-model developer to attend such a meeting. [40]

### Xi Jinping private-sector symposium

Less than a month later, on February 17, 2025, Liang attended a symposium on private enterprise hosted by Chinese leader Xi Jinping in Beijing. [7][22] Other attendees included Alibaba founder Jack Ma, Huawei founder Ren Zhengfei, Tencent CEO Pony Ma, Xiaomi founder Lei Jun, BYD chairman Wang Chuanfu, and CATL founder Robin Zeng. [22] Liang was not among the six executives that Xinhua reported as speaking at the meeting, but video footage and other media reports showed him in attendance, and analysts called the symposium a "symbolic" end to China's crackdown on its technology sector. [22]

## What recognition has Liang Wenfeng received?

Following the global attention generated by DeepSeek's R1 release, Liang's home village of Mililing became a tourist destination, and local authorities paid for a makeover that included tree planting, road widening, and demolitions. [19] When Liang returned to his hometown for the Lunar New Year in early 2025, he was given a hero's welcome, and the village hung a banner congratulating him on "becoming the pride of his hometown." [18]

His story, rising from a rural village to global AI prominence, resonated widely across China. [7] In 2025, *Time* included him in its list of the 100 most influential people in AI. [36] In 2026, the trade publication AI Magazine ranked Liang No. 5 on its list of the world's Top 100 AI Leaders, citing the global impact of his work at DeepSeek. [23]

## What is Liang Wenfeng's net worth?

Estimates of Liang's net worth have risen sharply along with DeepSeek's valuation. In November 2025, Forbes put his fortune at $11.5 billion when he debuted on its list of China's 100 richest people, based largely on his stake in DeepSeek, which Forbes then valued at about $15 billion. [28] The Financial Times reported in 2026 that Liang controlled nearly 90% of DeepSeek. [25]

DeepSeek's first outside funding round in 2026 pushed those estimates higher. In July 2026, the Bloomberg Billionaires Index put his net worth at $36 billion, up from about $16.7 billion, ranking him above Anthropic co-founder Dario Amodei and OpenAI's Greg Brockman as the richest founder of an AI-model company. [29] Forbes' real-time estimate stood at about $40 billion on September 23, 2026. [27]

## What are DeepSeek and Liang Wenfeng doing in 2026?

DeepSeek continued to scale rapidly under Liang's leadership in 2026. On April 24, 2026, the company released a preview of its V4 series: DeepSeek-V4-Pro, a Mixture-of-Experts model with 1.6 trillion total parameters (49 billion active per token), and the smaller DeepSeek-V4-Flash, with 284 billion total parameters (13 billion active). [24] Both models support a context window of up to 1 million tokens, and their weights were published under the MIT License. [24][34] DeepSeek followed with the official release of V4-Flash on July 31 and general availability of V4-Pro on August 13, 2026. On September 10 it released DeepSeek-V4.1-Flash, a natively multimodal model that it describes as the smallest in a new architecture family. [33]

In parallel, DeepSeek opened itself to outside investors for the first time. Reports in early May 2026 said talks had begun at a valuation of about $20 billion and could reach $45 billion, with China's state-backed chip investment fund expected to lead and Tencent and Alibaba in talks to participate. According to the Financial Times, Liang decided to raise funds partly so that he could offer employees shares in the company. [25] On June 3, Reuters reported, citing people familiar with the matter, that the round of about 50 billion yuan ($7.4 billion) could value DeepSeek at 350 billion to 400 billion yuan after the investment. [48] In July, again citing unnamed sources, Reuters reported that DeepSeek had raised about $7.4 billion in June at a post-money valuation of about 450 billion yuan, and noted that filings by two Chinese investors suggested a valuation of 350.88 billion yuan (about $52 billion). Liang personally committed 20 billion yuan, and Tencent and battery maker CATL put in 10 billion and 5 billion yuan respectively to become the largest external shareholders. [30] Other reported investors included China's national AI fund, NetEase, JD.com, IDG Capital, Loyal Valley Capital, Monolith Management, and Shixiang Capital. [30] After the round, DeepSeek said it planned to double staff across departments. [30]

The company soon sought more capital. In mid-July, Reuters reported that DeepSeek was planning a second round at a valuation of about 500 billion yuan ($74 billion) and had begun early deliberations on an IPO on Shanghai's STAR Market. [30] Bloomberg reported on July 25 that DeepSeek had paused the second round. [29] By August 6, according to a Bloomberg News report relayed by Reuters, DeepSeek had resumed it and was seeking close to $8 billion at a valuation close to 500 billion yuan (about $74 billion). [31] On September 9, 2026, the South China Morning Post reported, citing two unnamed people, that DeepSeek had hired four underwriters including CITIC Securities and aimed to begin the IPO process that year. [32] All of these figures come from unnamed-source reports rather than company disclosures.

Liang's hedge fund also had a strong year. High-Flyer's funds returned 56.6% in 2025, ranking second among China's 10 top-performing large hedge-fund firms in a ranking published by Shenzhen PaiPaiWang Investment & Management, behind only Lingjun Investment. [17][26]

## Personal Life

Liang is known for being private about his personal life, and little public information is available about his family or his interests outside AI research and quantitative finance. Most of what is known about his views comes from the interviews he gave to 36Kr's Waves. [8][16]

## ELI5: Who is Liang Wenfeng?

Liang Wenfeng is a Chinese man, born in 1985, who grew up in a small village and was very good at math. As an adult he started a company that uses computers to invest money (a hedge fund called High-Flyer), and he bought thousands of powerful computer chips. He then used those chips to start an AI company called DeepSeek. In early 2025, DeepSeek released a chatbot that performed close to the best American ones but cost far less to train. That surprised many people, and chipmaker Nvidia's stock fell sharply for a day. In 2026, investors put billions of dollars into DeepSeek, which made Liang one of the richest people in AI, even though he rarely gives interviews.

## See Also

- [DeepSeek](https://aiwiki.ai/wiki/deepseek)
- [DeepSeek-V3](https://aiwiki.ai/wiki/deepseek_v3)
- [DeepSeek-R1](https://aiwiki.ai/wiki/deepseek_r1)
- [DeepSeek V4](https://aiwiki.ai/wiki/deepseek_v4)
- [DeepSeek market crash (Jan 2025)](https://aiwiki.ai/wiki/deepseek_market_crash)
- [Artificial General Intelligence](https://aiwiki.ai/wiki/agi)
- [GPU Computing](https://aiwiki.ai/wiki/gpu_computing)
- [Nvidia](https://aiwiki.ai/wiki/nvidia)
- [OpenAI](https://aiwiki.ai/wiki/openai)
- [Mixture of Experts](https://aiwiki.ai/wiki/mixture_of_experts)
- [ChatGPT](https://aiwiki.ai/wiki/chatgpt)

## References

1. Danny and Dong Yuqing. "只招1%的天才，这家中国公司让硅谷难安" (Hiring only the top 1%: the Chinese company unsettling Silicon Valley). Phoenix Tech (凤凰网科技), republished by Sina Finance, 26 January 2025. https://finance.sina.com.cn/roll/2025-01-26/doc-inehhkse0668392.shtml
2. High-Flyer. "History" (幻方历程). High-Flyer website, English and Chinese versions (accessed September 2026). https://www.high-flyer.cn/en/history
3. Liang Wenfeng and Xiang Zhiyu. "鲁棒的PTZ摄像机目标跟踪算法" (Algorithm of robust object tracking using PTZ camera). Journal of Zhejiang University (Engineering Science) (浙江大学学报(工学版)), 2011, 45(1): 59-63. DOI: 10.3785/j.issn.1008-973X.2011.01.009. https://www.zjujournals.com/eng/CN/10.3785/j.issn.1008-973X.2011.01.009
4. "Meet DeepSeek founder Liang Wenfeng, a hedge fund manager." Fortune, January 27, 2025. https://fortune.com/2025/01/27/deepseek-founder-liang-wenfeng-hedge-fund-manager-high-flyer-quant-trading/
5. Yu Lili (于丽丽), 36Kr Waves (暗涌Waves). "揭秘DeepSeek:一个更极致的中国技术理想主义故事" (Unveiling DeepSeek: a more extreme story of Chinese technological idealism). Via Sina Finance, 22 July 2024. https://finance.sina.com.cn/roll/2024-07-22/doc-inceypaf4744158.shtml
6. "Beijing meeting puts spotlight on China's new face of AI, DeepSeek founder Liang Wenfeng." South China Morning Post, January 21, 2025. https://www.scmp.com/tech/policy/article/3295662/beijing-meeting-puts-spotlight-chinas-new-face-ai-deepseek-founder-liang-wenfeng
7. "From gifted student to AI hero, DeepSeek's founder Liang Wenfeng inspires a nation." South China Morning Post, 2025. https://www.scmp.com/tech/tech-trends/article/3307072/gifted-student-ai-hero-deepseeks-founder-liang-wenfeng-inspires-nation
8. "Interview with Deepseek Founder: We're Done Following. It's Time to Lead." The China Academy (translation of a July 2024 Waves interview). https://thechinaacademy.org/interview-with-deepseek-founder-were-done-following-its-time-to-lead/
9. Yu Lili (于丽丽), 36Kr Waves (暗涌Waves). "疯狂的幻方：一家隐形AI巨头的大模型之路" (Crazy High-Flyer: a hidden AI giant's road to large models). Via Sina Finance, 24 May 2023. https://finance.sina.cn/fund/jjyj/2023-05-24/detail-imyuwfwe3048541.d.html
10. "DeepSeek-V3 Technical Report." arXiv:2412.19437, December 2024. https://arxiv.org/abs/2412.19437
11. "Fire-Flyer AI-HPC: A Cost-Effective Software-Hardware Co-Design for Deep Learning." SC24 Conference, 2024. https://arxiv.org/abs/2408.14158
12. "Nvidia sheds almost $600 billion in market cap, biggest one-day loss in U.S. history." CNBC, January 27, 2025. https://www.cnbc.com/2025/01/27/nvidia-sheds-almost-600-billion-in-market-cap-biggest-drop-ever.html
13. "Marc Andreessen warns DeepSeek is 'AI's Sputnik moment.'" Fortune, January 27, 2025. https://fortune.com/2025/01/27/marc-andreessen-deepseek-sputnik-ai-markets/
14. "DeepSeek, Huawei, Export Controls, and the Future of the U.S.-China AI Race." CSIS, 2025. https://www.csis.org/analysis/deepseek-huawei-export-controls-and-future-us-china-ai-race
15. Gregory C. Allen, "DeepSeek: A Deep Dive." CSIS (congressional testimony), April 8, 2025. https://www.csis.org/analysis/deepseek-deep-dive
16. "Deepseek: From Hedge Fund to Frontier Model Maker." ChinaTalk (translation of a May 2023 Waves interview). https://www.chinatalk.media/p/deepseek-from-hedge-fund-to-frontier
17. "DeepSeek founder's High-Flyer ranks among China's top hedge-fund firms in 2025." South China Morning Post, January 2026. https://www.scmp.com/tech/tech-trends/article/3339633/deepseek-founders-high-flyer-ranks-among-chinas-top-hedge-fund-firms-2025
18. "'Pride of His Hometown': Who Is DeepSeek Founder Liang Wenfeng?" Entrepreneur, 2025. https://www.entrepreneur.com/business-news/who-is-liang-wenfeng-the-founder-of-ai-disruptor-deepseek/486564
19. "Hometown of DeepSeek founder becomes tourist hotspot, renovations paid by government." South China Morning Post, 2025. https://www.scmp.com/news/people-culture/china-personalities/article/3301282/hometown-deepseek-founder-becomes-tourist-hotspot-renovations-paid-government
20. "DeepSeek displaces ChatGPT as the App Store's top app." TechCrunch, January 27, 2025. https://techcrunch.com/2025/01/27/deepseek-displaces-chatgpt-as-the-app-stores-top-app/
21. "DeepSeek's Open Source Week showcases commitment to open, collaborative AI ecosystem." Xinhua, February 28, 2025. https://english.news.cn/20250228/1ca33895cfe0495ab62bbce6d12195d2/c.html
22. "Not just Jack Ma: Who's who at Xi Jinping's Monday summit." Fortune, February 18, 2025. https://fortune.com/asia/2025/02/18/who-met-xi-jinping-china-tech-meeting-jack-ma-huawei-byd-deepseek-unitree-xiaomi/
23. "Liang Wenfeng: The Visionary Behind DeepSeek's AI Rise." AI Magazine, 2026. https://aimagazine.com/news/liang-wenfeng-top-100-ai-leaders-2026
24. "DeepSeek V4 Preview Release." DeepSeek API Docs, April 24, 2026. https://api-docs.deepseek.com/news/news260424
25. "DeepSeek could hit $45B valuation from its first investment round." TechCrunch, May 6, 2026. https://techcrunch.com/2026/05/06/deepseek-could-hit-45b-valuation-from-its-first-investment-round/
26. "DeepSeek Founder Liang's Funds Surge 57% as China Quants Boom." Bloomberg, January 12, 2026. https://www.bloomberg.com/news/articles/2026-01-12/deepseek-founder-liang-s-funds-surge-57-as-china-quants-boom
27. "Liang Wenfeng." Forbes profile (real-time net worth as of September 23, 2026). https://www.forbes.com/profile/liang-wenfeng/
28. Yue Wang, "AI Wunderkind DeepSeek's Founder Debuts On China's 100 Richest List." Forbes, November 5, 2025. https://www.forbes.com/sites/ywang/2025/11/05/ai-wunderkind-deepseeks-founder-debuts-on-chinas-100-richest-list/
29. "DeepSeek said to tell backers of funding pause after viral posts." Fortune (Bloomberg), July 25, 2026. https://fortune.com/2026/07/25/deepseek-liang-wenfeng-backers-fundraising-pause-viral-posts-investors/
30. "China's DeepSeek to raise fresh capital at $74 billion valuation ahead of onshore IPO, sources say." Reuters, July 15, 2026 (via Yahoo Finance). https://finance.yahoo.com/technology/ai/articles/chinas-deepseek-raise-fresh-capital-095807183.html
31. Reuters. "DeepSeek resumes funding round seeking nearly $8 billion, Bloomberg News reports." 6 August 2026 (via Zawya). https://www.zawya.com/en/world/ai-shaping-digital-future/deepseek-resumes-funding-round-seeking-nearly-8-billion-bloomberg-news-reports-423914
32. "Chinese AI firm DeepSeek taps underwriters including Citic Securities for IPO: sources." South China Morning Post, September 9, 2026. https://www.scmp.com/tech/tech-trends/article/3366948/chinese-ai-firm-deepseek-taps-underwriters-including-citic-securities-ipo-sources
33. DeepSeek, "Change Log" (API updates). DeepSeek API Docs, accessed September 2026. https://api-docs.deepseek.com/updates/
34. DeepSeek, "DeepSeek-V4-Pro" model card. Hugging Face. https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro
35. DeepSeek-AI, "DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning." arXiv:2501.12948, January 2025. https://arxiv.org/abs/2501.12948
36. "The 100 Most Influential People in AI 2025." Time, August 2025. https://time.com/collections/time100-ai-2025/
37. DeepSeek, "DeepSeek-Coder" repository README (license section). GitHub. https://github.com/deepseek-ai/DeepSeek-Coder
38. DeepSeek, "DeepSeek-Coder-V2" repository README. GitHub. https://github.com/deepseek-ai/DeepSeek-Coder-V2
39. DeepSeek-AI, "DeepSeek-V2: A Strong, Economical, and Efficient Mixture-of-Experts Language Model." arXiv:2405.04434, May 2024. https://arxiv.org/abs/2405.04434
40. "China signals global AI push as MiniMax CEO joins meeting with Premier Li Qiang." South China Morning Post, January 2026. https://www.scmp.com/tech/policy/article/3340515/china-signals-global-ai-push-minimax-ceo-joins-meeting-premier-li-qiang
41. DeepSeek, "DeepSeek-R1" model card (MIT License). Hugging Face. https://huggingface.co/deepseek-ai/DeepSeek-R1
42. Asset Management Association of China. "私募基金管理人公示信息: 浙江九章资产管理有限公司" (Private fund manager disclosure: High-Flyer Asset Management (Zhejiang) Co., Ltd., P1020148). Information last updated 27 November 2025. https://gs.amac.org.cn/amac-infodisc/res/pof/manager/101000028665.html
43. Asset Management Association of China. "私募基金管理人公示信息: 宁波幻方量化投资管理合伙企业（有限合伙）" (Private fund manager disclosure: High-Flyer Quant Investment Management (Ningbo) L.P., P1032021). Information last updated 10 November 2025. https://gs.amac.org.cn/amac-infodisc/res/pof/manager/101000052186.html
44. Zhang Zhao. "谁在爆炒每日互动？与DeepSeek无股权关联，2024年或亏损近亿元" (Who is pumping Getui? No equity link to DeepSeek). Time Weekly (时代财经), 9 February 2025. https://time-weekly.com/post/318463
45. Cailian Press. "幻方量化深夜处置婚外情事件：涉事创始人停职，量化圈再被带到风口浪尖" (High-Flyer suspends founder; quant sector back in the spotlight). 27 October 2023. https://www.cls.cn/detail/1498506
46. Yicai. "独家|幻方量化回应市场关注：AGI不是用来炒股的，'和金融没关系'" (Exclusive: High-Flyer responds, AGI is not for trading stocks). 16 April 2023. https://www.yicai.com/news/101732215.html
47. Li Yu, 21st Century Business Herald, via Securities Times (stcn.com). "量化巨头幻方搅局AI大模型：首期投入自有资金30亿元" (Quant giant High-Flyer shakes up AI large models: 3 billion yuan of own funds in first phase). 19 July 2024. https://www.stcn.com/article/detail/1263664.html
48. Reuters, via CNBC. "DeepSeek slated to draw $7 billion in maiden fundraising, sources say." 3 June 2026. https://www.cnbc.com/2026/06/03/deepseek-slated-to-draw-7-billion-in-maiden-fundraising-sources-say.html
49. Jimu News (极目新闻). "DeepSeek创始人梁文锋回乡过年：家乡拉横幅，还有游客打卡" (DeepSeek founder Liang Wenfeng returns home for the New Year: banner in his hometown, visitors check in). 31 January 2025. https://www.ctdsb.net/c1476_202501/2363629.html