文 | 高恒说
10月27日，41岁的DeepSeek创始人梁文锋首次登上《2026胡润百富榜》，便以2800亿元的财富跻身中国富豪榜前十，位列第七。
就在同一天，另一条关于他的消息引起关注。据凤凰网科技援引接近DeepSeek的人士透露，梁文锋长期在北京租房居住，房间布置比较简单。
一边是2800亿元的身家，一边是北京的出租房。这样的反差出现在近年来中国人工智能产业最受关注的创业者之一身上，多少有些出人意料。
从幻方量化到DeepSeek，从量化投资人到千亿富豪，梁文锋的财富随着人工智能产业的发展迅速增长。但相比不断攀升的财富数字，他的个人生活却鲜少出现在公众视野中。如今，DeepSeek正在加速扩张，这位新晋千亿富豪的事业也进入了新的阶段。
一、富豪榜之外的梁文锋
梁文锋在北京的生活，可能比外界想象的简单得多。
据凤凰网科技10月27日报道，自DeepSeek走红以来，梁文锋长期往返于北京、杭州两地，偶尔也会前往深圳。在北京，他一直租房居住，房间布置简单，没有在装修上花太多心思。
这样的居住安排，与DeepSeek的办公布局有关。北京和杭州是公司的两个主要办公地，其中多数算法团队在北京办公，部分基础设施团队位于杭州。随着公司业务发展，梁文锋需要在两地之间频繁往返。
与公司不断扩大的影响力相比，梁文锋本人的公众曝光度始终不高。
2023年5月和2024年7月，36氪旗下《暗涌Waves》曾两次对话梁文锋，成为外界了解其创业理念和工作方式的重要窗口。
在2024年的报道中，记者曾描述过梁文锋的日常：与其他研究人员一样，每天看论文、写代码、参与小组讨论。尽管当时DeepSeek已经凭借技术创新受到人工智能行业关注，他仍将大量精力放在技术研究和团队工作上。
这种低调，也体现在他的日常交往中。
2025年春节，DeepSeek凭借R1模型在全球走红，梁文锋也突然成为舆论焦点。
当年1月28日，梁文锋回到广东省湛江市吴川市覃巴镇米历岭村过年。当地村民挂起欢迎横幅，村口竖起拱门，不少游客专程前往拍照打卡。
据《羊城晚报》当时援引其同乡的消息，梁文锋在除夕返回家乡，大年初一上午便离开了村子。此前，他还曾与昔日同学、好友一起踢足球。
在接受媒体采访时，梁文锋的高中同学兼好友陈先生曾评价，他平时衣着打扮和为人处世都比较低调。家乡一位与梁文锋同辈的校友也回忆，他从小性格比较内向，过年过节回家通常不怎么出门。
2025年2月17日，梁文锋出席在北京召开的民营企业座谈会，与马云、任正非、马化腾、雷军等企业家一同出现在央视新闻画面中。这也是DeepSeek走红后，他为数不多的公开活动之一。
此后，DeepSeek持续推出新模型、发表技术论文，梁文锋本人却很少接受媒体采访，也没有像一些互联网企业创始人那样频繁出现在公众面前。
从2025年春节短暂返乡，到如今被披露长期在北京租房，外界能够了解到的梁文锋个人生活，实际上只有零散的几个片段。
在老家与朋友踢球，在公司参与技术研究，在北京和杭州之间奔波。这些公开信息勾勒出一个并不热衷于展示个人生活的科技创业者。
但在过去一年多时间里，梁文锋的个人财富已经发生了惊人的变化。
2025年，他尚未进入胡润百富榜。到了2026年，他不仅首次上榜，还直接跻身中国富豪榜前十。
这一变化，与他过去十余年的创业经历密切相关。
二、从幻方量化到2800亿身家
梁文锋的财富积累，始于DeepSeek诞生之前。
在进入人工智能大模型领域之前，他已经是国内量化投资行业的重要人物。
2008年前后，还在浙江大学学习的梁文锋便开始研究人工智能在金融市场中的应用。毕业后，他与团队尝试过多个技术方向，最终选择将机器学习应用于量化交易。
2015年，梁文锋与徐进等人共同创办幻方量化。与传统依靠基金经理经验进行投资决策的机构不同，幻方主要通过数学模型、人工智能算法和计算机程序分析市场数据、制定投资策略并执行交易。
凭借这一模式，幻方逐渐在国内量化私募行业站稳脚跟。
2016年，幻方开始将深度学习模型应用于交易策略。2019年，其资产管理规模突破百亿元。到2021年，幻方管理规模一度超过1000亿元，成为国内首批跨过千亿元门槛的量化私募机构之一。
不过，千亿元管理规模并不意味着梁文锋拥有千亿元财富。幻方管理的主要是客户资产，公司通过管理费、业绩报酬等方式获得收入。
量化投资真正为梁文锋后来创办DeepSeek提供的，除了资金积累，还有技术人才和算力基础。
早在大模型成为全球科技行业热点之前，幻方就已经开始大规模投入人工智能基础设施。
2019年，幻方投资近2亿元建设深度学习训练平台“萤火一号”，配置约1100块GPU。2021年，又投入约10亿元建设“萤火二号”，配置约1万张英伟达A100 GPU。
这些原本服务于量化投资的技术积累和计算资源，为梁文锋后来进入大模型领域创造了条件。
2023年，随着ChatGPT引发全球人工智能热潮，幻方宣布成立独立组织“深度求索”（DeepSeek），由梁文锋带领团队开展通用人工智能研究。
与当时不少通过应用产品探索商业化的大模型创业公司不同，DeepSeek将主要精力放在基础模型研究和技术创新上。
2024年5月，DeepSeek推出V2开源模型，凭借较低的推理成本引发国内大模型市场的价格竞争。此后，字节跳动、阿里巴巴、百度、腾讯等科技企业相继调整相关产品价格。
真正让DeepSeek获得全球关注的，是2025年1月20日发布的推理模型DeepSeek-R1。
这款模型通过强化学习等技术提升推理能力，在数学、代码和复杂推理等任务上表现突出。DeepSeek同步开放模型权重，使开发者和企业能够在其技术基础上开展研究和应用。
凭借模型性能、成本效率和开源策略，DeepSeek逐渐从一家中国大模型创业公司，成长为全球人工智能行业的重要竞争者。
技术突破所带来的影响，也开始反映在公司的市场估值上。
2026年，DeepSeek启动外部融资。据《财经》10月13日报道，公司当年4月开启首轮融资，6月底完成交割，彼时整体估值已超过3500亿元。进入第二轮融资阶段后，参与交易的人士给出的投前估值进一步达到约5000亿元。
公司估值的持续上升，也为理解梁文锋的个人财富提供了重要线索。
胡润研究院解释，2025年未将梁文锋列入百富榜，主要考虑到DeepSeek此前“不商业化”的定位。而到了2026年，随着公司估值快速攀升至500多亿美元，梁文锋以2800亿元的估算财富首次进入榜单前十。
不过，富豪榜上的2800亿元，并不意味着梁文锋实际拥有同等规模的现金。
对于尚未上市的企业，富豪榜通常需要结合企业估值、个人持股权益及其他资产情况估算财富。DeepSeek的估值反映的是投资者对其技术能力和未来商业价值的判断，梁文锋所持股权的估算价值也会随公司估值发生变化。
胡润研究院并未在公开榜单中披露这2800亿元财富的逐项计算过程，DeepSeek融资后的完整股权结构及梁文锋最终持有的权益比例，也尚无足够的公开资料可以精确还原。
从这一意义上说，2800亿元更多体现的是梁文锋所拥有的企业权益获得了怎样的市场定价，而非其能够随时支配的个人资金。
从幻方量化的千亿元管理规模，到DeepSeek的数千亿元估值，梁文锋用了十余年时间，完成了从量化投资到人工智能创业的跨越。
随着公司估值不断攀升，DeepSeek的发展方式也开始发生变化。曾经主要依靠自身积累支持技术研究的公司，正在吸引大规模外部资本进入。
对于梁文锋而言，技术突破依然重要，但融资、组织扩张和企业经营，也成为他必须面对的新课题。
三、DeepSeek扩张，梁文锋面临新考验
伴随个人财富迅速增长，梁文锋创办的DeepSeek也正在经历新一轮扩张。
今年6月25日，DeepSeek发布招聘公告，计划将所有部门的规模至少扩大一倍，首批开放33个岗位，涉及技术研发、算力基础设施、产品开发和企业运营等多个领域。
到了9月8日，DeepSeek又一次性释放约150个招聘名额，创下公司单次招聘规模的新高。
与此前相比，这次招聘更加集中于大模型研究平台、智能体（Agent）框架、数据工程、线上服务和底层计算系统等方向。
DeepSeek Harness团队负责人崔添翼当时解释，随着数据、机器和训练任务规模快速增长，原有后端系统已经难以满足新的需求，需要增加人手，对相关系统进行升级、维护和重写。
招聘方向的变化，反映出DeepSeek在组织建设上的新需求。
过去，DeepSeek的技术竞争力主要建立在精简团队、高效算法和持续研发之上。如今，随着模型调用量和业务复杂度增加，除了算法研究人员，公司还需要更多工程技术人才，支撑大规模模型训练、部署和运行。
与此同时，外部资本也在加速进入这家公司。
据《证券时报》10月23日报道，DeepSeek近日已完成约1000亿元的新一轮融资。参与投资的既有腾讯、宁德时代等老股东，也有君联资本、吉利汽车产业基金、中芯聚源等产业资本。
与一般的财务融资相比，这轮交易还有一个值得关注的细节。
《证券时报》援引投资人士的消息称，DeepSeek要求意向投资方说明能够提供的数据支持和业务协同资源。
这意味着，除了资金实力，投资方能够带来的芯片、算力和产业应用资源，也成为DeepSeek的重要考量。
对于大模型企业而言，训练和推理需要持续投入大量计算资源。随着模型规模扩大、技术迭代加快，企业不仅需要研发资金，还需要稳定的算力供应、工程基础设施和应用场景。
资本的进入，有助于DeepSeek获得进一步发展的资源，但也将使公司的经营管理面临更多要求。
上市计划也开始进入外界视野。
9月9日，路透社援引知情人士消息报道称，DeepSeek已聘请中信证券，为潜在的上海证券交易所科创板上市做准备。
《财经》10月13日进一步报道，DeepSeek正加快上市筹备，计划最快于2026年底递交申请，目标在2027年挂牌。
如果相关计划顺利推进，DeepSeek将进一步走向公开资本市场。届时，投资者关注的将不只是模型性能和技术突破，还包括公司的收入增长、研发成本、现金流和长期经营能力。
这也是DeepSeek从技术创业公司向大型人工智能企业发展过程中需要面对的变化。
在融资和组织扩张之外，DeepSeek仍在持续推进技术研发。
9月10日，公司发布V4.1 Flash模型，通过新的模型架构提升推理效率，并进一步降低API调用价格。10月17日，DeepSeek更新模型版本，开放自研搜索API。10月20日，公司又宣布开源内部推理框架DeepInfer，希望进一步降低大模型部署和运行成本。
这些动作延续了DeepSeek此前重视技术创新和成本优化的发展方向，也表明公司的竞争正在从单纯的模型性能，延伸至推理效率、部署成本和实际应用。
对于DeepSeek而言，如何让模型被更多开发者和企业使用，并在使用规模扩大时控制运营成本，将直接影响技术优势能否转化为长期经营能力。
与此同时，梁文锋也开始关注大模型之外的技术方向。
10月12日，凤凰网科技援引多个信源报道称，梁文锋近期密集接触具身智能创业者，探讨机器人、大模型与现实世界交互等方向。DeepSeek正在评估是否亲自组建团队进入这一领域，或者通过投资相关企业参与布局。
相比语言模型，具身智能涉及机器人本体、感知控制、软硬件协同及真实场景应用，技术路线更加复杂，也需要更长周期的研发和产业投入。
相关业务最终将如何推进，还有待DeepSeek进一步披露。但从已有报道来看，梁文锋正在关注大模型技术向更多实际场景延伸的可能性。
从量化投资到大模型研发，再到如今被报道的千亿元融资、团队扩张和潜在上市，梁文锋所面对的企业规模和经营环境，已经与创办DeepSeek之初明显不同。
过去，他需要证明的是，一支中国技术团队能否以相对有限的资源，训练出具有国际竞争力的大模型。如今，随着资本、人才和产业资源不断向DeepSeek聚集，如何将技术优势转化为持续的竞争力，成为更加现实的经营问题。
回到外界关注的北京出租房，这或许只是梁文锋个人生活中一个普通的选择。
过去一年多时间里，DeepSeek的技术成果不断获得市场认可，梁文锋的个人财富也随之迅速增长。相比财富数字的变化，外界能够了解到的他的生活，仍然是那些有限的片段。
胡润百富榜为梁文锋标出了2800亿元的身家，也让这位长期保持低曝光度的创业者跻身中国最富有的企业家之列。
但对DeepSeek而言，数千亿元估值所体现的市场期待，仍需要未来的技术进步、商业应用和企业经营来支撑。随着公司进入新的发展阶段，梁文锋面前的挑战，也已不同于当年创办DeepSeek之时。
本文由 AI图鉴自动抓取并翻译，版权归原作者所有；如需查看原文请点下方链接。
♪ Man, Man ♪
On 27 October, the 41-year-old founder of DeepSeek, Liang Wensai, first placed on the 2026 Hu Yun Bount list, and made China ' s top 10, seventh, with $28 billion in wealth.
On the same day, another piece of information about him raised concerns. Liang Wenfeng has long been renting a house in Beijing and making the room simpler, according to the phoenix technology citing people close to DeepSeek.
On the one hand, it's $280 billion and on the other hand, it's a rental house in Beijing. This contrast has occurred in recent years with one of China's leading entrepreneurs in the AI industry, which is somewhat unexpected.
From Fantasy quantization to DeepSeek, from quantization investors to hundreds of billions of wealthy people, Liang Wenbing's wealth grew rapidly with the development of the artificial intelligence industry. But his personal life is rarely visible in public view compared to the rising wealth figures. Today, DeepSeek is expanding at an accelerated pace, and the new billionaire has entered a new phase in his career.
I. Liang Wenzeng outside the list of rich people
Liang Wenfeng's life in Beijing may be much simpler than the outside world thinks.
On 27 October, Phoenix Network technology reported that since DeepSeek went red, Liang Wenbing had travelled to Beijing, Hangzhou and occasionally to Shenzhen. In Beijing, he had been renting a house, making it simple and not spending much time on renovations.
This arrangement relates to the office layout of DeepSeek. Beijing and Hangzhou are the two main offices of the company, with most of the algorithm teams located in Beijing and some infrastructure teams located in Hangzhou. As the company develops its business, Liang Wenbing needs to travel frequently between the two locations.
Liang's own public exposure remains low compared to the company's growing influence.
In May 2023 and July 2024, the Concealed Waves, under the banner of 36 krypton, engaged in two dialogues, which served as an important window of understanding of their entrepreneurial ideas and working methods.
In 2024, journalists described the daily life of Liang Wenbong: like other researchers, they read papers, write codes and participate in panel discussions every day. Despite the fact that DeepSeek had received attention from the artificial intelligence industry through technological innovation at that time, he had focused heavily on technology research and team work.
This low profile is also reflected in his daily interactions.
In the spring of 2025, DeepSeek gained global popularity with the R1 model, and Liang Wenfeng suddenly became a focus of opinion.
On 28 January of that year, Liang Wenfeng returned to the village of Michling, in the town of Wuchuan, in Changjiang, Guangdong Province, for a year. The local villagers had banners of welcome, arches at the village entrance had been erected, and many tourists had travelled to take photographs and cards.
Leung Wenfeng returned home on New Year's Eve and left the village one morning early in the year. He played football with former classmates and friends.
In interviews with the media, Liang Wenzai ' s high school classmate and friend, Mr. Chen, had assessed that he was in a low-profile fashion and in a low-profile fashion. An alumni from his hometown and his peers also recalled that he was introverted from a young age and did not usually go home very much during the New Year.
On 17 February 2025, Liang Wenfeng attended a private business symposium in Beijing, where he appeared with entrepreneurs such as Ma Yun, Zhengfei, Ma Zhengfei, Ma Zhenteng, and the Thunder Force in the news. This is also one of the few public events that DeepSeek did after he lost.
Since then, DeepSeek has continued to launch new models and publish technical papers, while Liang Wenfeng himself has rarely received media interviews and has not appeared to the public as frequently as some Internet business founders.
From a brief return in the spring of 2025 to the fact that it is now revealed that it has been renting houses in Beijing for a long time, the private life of Liang Wenbong is known to the outside world, and in fact there are only isolated segments.
Plays at home and with friends, engages in technical research at companies and runs between Beijing and Hangzhou. These public messages outline a technology entrepreneur who is not interested in showing personal life.
But for more than a year, Liang Wenbo's personal wealth has changed dramatically.
In 2025, he had not entered the Hu Yunbian list. By 2026, he was not only first on the list, but also directly on the top 10 of China's wealthier list.
This change is closely linked to his entrepreneurial experience over the past decade or so.
From Fantasia quantization to 280 billion
Leung Wenzor's accumulation of wealth began before DeepSeek was born.
Before entering the field of artificial intelligence LLM, he was already an important figure in the domestic investment industry of quantization.
Around 2008, Liang Wenbing, also studying at Zhejiang University, began studying the use of artificial intelligence in financial markets. After graduation, he and the team tried several technical directions and eventually chose to apply machine learning to quantization transactions.
In 2015, Liang Wensai and Suh-jin joined in creating the Fantasy quantization. Unlike institutions that traditionally rely on the experience of fund managers to make investment decisions, phantoms analyse market data, develop investment strategies and implement transactions mainly through mathematical models, artificial intelligence algorithms and computer programs.
With this model, the illusion has gradually stabilized in the domestic quantization private sector.
In 2016, the phantom began to apply deep learning models to trade strategies. In 2019, its assets were managed on a scale of over $10 billion. By 2021, the size of phantom management was once over $100 billion, making it one of the first quantization private institutions in the country to cross the $100 billion threshold.
However, the size of the hundreds of billions of dollars in management does not mean that Liang Wenfeng has hundreds of billions of dollars in wealth. Fantasy management is primarily a client ' s asset, and companies earn income through management fees, performance rewards, etc.
quantization investments were really provided for the subsequent establishment of DeepSeek by Liang Wenfeng, which, in addition to the accumulation of funds, also had a technological and statistical base.
Even before LLM became the hotspot of the global science and technology industry, the phantom had begun to invest massively in artificial intelligence infrastructure.
In 2019, phantom invested nearly $200 million in the construction of the In-depth Learning Training Platform, Fire One, with approximately 1100 GPUs. In 2021, an additional $1 billion was invested in the construction of Fireflies II, with an estimated 10,000 Yvda A100 GPUs.
These technology accumulation and computing resources, which were intended to serve the investment of quantization, created the conditions for Liang Wenbei to subsequently enter the field of LLM.
In 2023, with ChatGPT triggering a global wave of artificial intelligence, Fantasy announced the establishment of an independent organization, DeepSeek, led by Liang Wenbing, to conduct generic artificial intelligence research.
Unlike a number of LLM start-ups commercialized at the time by applying product exploration, DeepSeek focused on basic modelling research and technological innovation.
In May 2024, DeepSeek launched the V2 open source model, which generated price competition in the domestic market of LLM based on lower reasoning costs. Since then, technology enterprises such as byte, Alibaba, Baidu, and Tencent have adjusted their respective product prices.
What really brought DeepSeek to global attention was the DeepSeek-R1 reasoning model of publish of 20 January 2025.
This model enhances the ability to reason through techniques such as enhanced learning, and is prominent in tasks such as mathematics, code and complex reasoning. DeepSeek ' s synchronous open model weights enable developers and enterprises to conduct research and applications based on their technology.
With model performance, cost-efficiency and open-source strategies, DeepSeek has grown from a Chinese LLM start-up company to become a major competitor in the global artificial intelligence industry.
The impact of technological breakthroughs is also beginning to be reflected in corporate market valuations.
In 2026, DeepSeek initiated external financing. According to Financial Affairs on 13 October, the company had opened its first round of financing in April of that year and had completed delivery by the end of June, when the overall valuation had exceeded $350 billion. After entering the second financing phase, the pre-investment valuation given by those involved in the transaction reached a further value of approximately $500 billion.
The continued rise in corporate valuation also provides important clues to the understanding of Liang Wenbong ' s personal wealth.
The Hu Yun Institute explained that Liang Wenbing was not included in the list in 2025, mainly taking into account the previous position of DeepSeek as “non-commercial”. By 2026, as corporate valuations rose rapidly to over $50 billion, Liang Wenfeng entered the top 10 list for the first time with an estimated wealth of $280 billion.
However, the $280 billion on the list of wealthy people does not mean that Liang Wenfeng actually owns cash of the same size.
For enterprises that have not yet been listed, the list usually needs to estimate wealth in the context of corporate valuation, individual equity and other assets. DeepSeek ' s valuation reflects the investor ' s judgement as to its technical capability and future business value, and the estimated value of Liang Wenfeng ' s equity will change with the company ' s valuation.
The Hu Yun Institute did not disclose on its public list the item-by-item calculation of the $280 billion in wealth, nor did it have sufficient publicly available information to accurately restore the full stock structure after DeepSeek ' s financing and the share of the ultimate interest held by Liang Wenbing.
In this sense, the $280 billion is more a reflection of the market pricing of the business interests of Liang Wenfeng than of the personal funds at his disposal.
From the hundreds of billions of dollars in the management scale of the Fantasy quantization, to the hundreds of billions of dollars in the valuation of DeepSeek, Liang Wenbing spent more than a decade on investing in artificial intelligence entrepreneurship.
The way DeepSeek developed began to change as corporate valuations continued to climb. Companies that have largely relied on their own accumulation to support technological research are attracting large-scale external capital.
Technological breakthroughs remain important for Liang Wenfeng, but financing, organizational expansion and entrepreneurship have also become new topics for him.
iii. DeepSeek expands and Liang Wensai faces new tests
Along with the rapid growth of personal wealth, DeepSeek, founded by Liang Wenfeng, is undergoing a new wave of expansion.
On 25 June this year, the DeepSeekpublish job announcement plans to double the size of all sectors by opening the first 33 jobs in a number of fields, including technological research and development, computing infrastructure, product development and business operations.
By 8 September, DeepSeek had released another 150 recruitment places, creating a new level of single recruitment for the company.
This recruitment was more focused than before on the LLM research platform, the AI agent (Agent) framework, data engineering, online services and bottom computing systems.
The head of the DeepSeek Harness team explained at that time that, with the rapid growth in the size of data, machinery and training tasks, the original back-end system was no longer sufficient to meet new needs and that additional staff were needed to upgrade, maintain and rewrite related systems.
The change in recruitment orientation reflects the new organizational needs of DeepSeek.
In the past, the technological competitiveness of DeepSeek was based mainly on streamlined teams, efficient algorithms and continuous research and development. Today, as models increase in volume and operational complexity, companies need more engineering skills to support large-scale model training, deployment and operation, in addition to algorithm researchers.
At the same time, external capital is accelerating entry into the company.
According to the Securities Times, on 23 October, DeepSeek recently completed a new round of financing of approximately $100 billion. The investment involved both old shareholders such as Tencent and the Ningde era, and industrial capital such as the King ' s Union Capital, the Guilly Motor Industry Fund, and the Central Chip Pool.
Compared to general financial financing, this round of transactions is a matter of concern.
The Securities Times quoted information from investors stating that DeepSeek required interested investors to indicate data support and business synergy resources that could be provided.
This means that, in addition to financial strength, the chip, calculus and industrial applications that investors can bring are also important considerations for DeepSeek.
For LLM enterprises, training and reasoning require substantial and sustained computing resources. As models grow in size and technologies accelerate, enterprises need not only research and development finance, but also stable computing, engineering infrastructure and application scenarios.
The entry of capital will help DeepSeek to gain resources for further development, but it will also expose the management of companies to additional requirements.
The marketing project also began to gain access to the outside world.
On 9 September, Reuters, citing well-informed sources, reported that DeepSeek had hired Chinese mail securities to prepare for the potential Shanghai Stock Exchange Section to open the market.
On 13 October, it was further reported that DeepSeek was expediting listing preparations and was scheduled to submit its application by the end of 2026, with a target of 2027.
If the plans move forward smoothly, DeepSeek will move further towards open capital markets. At that time, investors will focus not only on model performance and technological breakthroughs, but also on corporate income growth, R & D costs, cash flows and long-term capacity.
This is also the change that DeepSeek needs to face in its move from a technology start-up company to a large artificial intelligence enterprise.
Beyond financing and organizational expansion, DeepSeek continues to advance technological research and development.
On 10 September, the company publishV4.1 Flash model improved reasoning efficiency through a new model structure and further reduced the API call price. On 17 October, DeepSeek updated the model, open self-study search for API. On 20 October, the company also announced a framework for open-source internal reasoning, DeepInfer, in the hope of further reducing the deployment and operation costs of LLM.
These actions continue the development orientation of DeepSeek, which previously focused on technological innovation and cost optimization, and indicate that competition by companies is extending from mere model performance to reasoning efficiency, deployment costs and practical applications.
For DeepSeek, how models can be used by more developers and firms, and how operational costs can be controlled when the scale of use increases, will directly affect whether technological advantages can be translated into long-term business capacity.
At the same time, Liang Wenfeng has also begun to focus on technical directions beyond LLM.
On 12 October, Phoenix Network technology quoted several sources that reported that Liang Wenfeng had recently made intensive contact with smart entrepreneurs in the direction of robotics, LLM interaction with the real world. DeepSeek is assessing whether to form teams in person to enter the field or to participate in the layout through investment-related enterprises.
Compared to language models, smartness involves robotics, sensory control, software and hardware synergy and real-world applications, which are more complex and require longer cycles of research and development and industrial inputs.
The manner in which the related business will eventually proceed is subject to further disclosure by DeepSeek. However, according to already reported reports, Liang Wenfeng is concerned about the possibility of LLM technology extending to more practical scenarios.
From the investment of Z83Z to Z43Z R & D, to the hundreds of billions of yuan of financing, team expansion and potential listings that are now reported, Liang Wenbai has faced a distinctly different size and operating environment from the start of DeepSeek.
In the past, he had to prove whether a Chinese technical team could train an internationally competitive LLM with relatively limited resources. Today, as capital, talent and industrial resources gather to DeepSeek, how to translate technological advantages into continuous competitiveness becomes a more realistic business issue.
Returning to the private rental accommodation in Beijing is perhaps just a common option in Liang Wenbing ' s personal life.
For more than a year, DeepSeek ' s technological achievements have continued to gain market recognition, followed by rapid growth in Liang ' s personal wealth. The changes in wealth figures make it possible for the outside world to understand that his life remains limited.
Hu Yunbian's brand of Liang's $280 billion worth of capital has also made this long-term low-profile entrepreneur among China's richest entrepreneurs.
For DeepSeek, however, the market expectations embodied in the hundreds of billions of dollars valuation still need to be underpinned by future technological advances, commercial applications and business operations. As the company enters a new stage of development, the challenges facing Liang Wenfeng are no longer the same as when DeepSeek was founded.
Auto-fetched and translated by AI Atlas. All rights belong to the original author.