梁文锋留住97%员工

marsbitPublicado a 2026-05-04Actualizado a 2026-05-04

“确实在接触DeepSeek融资”,一位FA机构朋友告诉我们。

过去半个月,DeepSeek终于开启融资的消息发酵。而外界归结这次“反常”时几乎都不可避免提到:梁文锋要给内部核心员工一个确定的估值了。

毕竟这一年,大模型竞争日趋焦灼,DeepSeek核心人才流动沸沸扬扬,诸如罗福莉、王炳宣、郭达雅等人陆续跳至小米、腾讯和字节。

喧嚣之外,一组数据映射出更为平和的情况——DeepSeek V4在4月下旬终于发布,技术报告里一份长长的作者致谢名单显示,研究工程团队约270人中10人在研发期间离去。对应下来,技术研发人员离职率不到4%。

其实梁文锋留住了绝大多数人。

掀开DeepSeek跳槽风波,10名员工离开

2023年起,一种强烈的推背感袭来。

ChatGPT席卷之后,月之暗面、阶跃星辰、MiniMax等明星公司纷纷在这年或前后成立,豆包、通义千问、文心一言等大厂大模型产品密集涌现。

不早不晚地,梁文锋也在这一年将DeepSeek落地北京和杭州。

此时他罕见的一次对外分享中,聊到了人才观:DeepSeek大部分开发人员都是应届毕业生或AI从业时间不长,如果追求短期目标,招聘有经验的人当然没错,但从长远来看,基本技能、创造力和热情更为重要。

确实如此。2025年初,DeepSeek R1实力爆发,人们才开始真正关注到这个约150人规模的团队,许多都是国内TOP高校刚毕业或还没毕业的年轻人,清北含量极高。

几乎不可避免,此后一年里,人才流动的话题开始袭向DeepSeek。

2025年开始,DeepSeek传出罗福莉、王炳宣、魏浩然、阮翀等核心骨干离职,其中不少人跳槽他处成为核心业务负责人。坊间为此做了个通俗易懂的打趣:“当DeepSeek内部成员发现段位差不多的人跳槽出去能拿到那么多,那我为什么不可以?”

直到2026年初,随着郭达雅跳槽去字节seed团队,关于DeepSeek人才流失的讨论被推至高点,而当时配上迟迟不发布的DeepSeek V4,难免令人生出几分青黄不接隐忧。

但现实并没有这么沮丧。如今DeepSeek V4终于亮相,在同步发布的技术报告中披露了一份作者致谢名单。细细看下来,其Research & Engineering也就是研究工程团队约270人,这部分也被认为是一家AI公司最核心的研发团队,另有Business& Compliance即商业合规成员48人。

在DeepSeek V4研发期间,只有10名研究工程团队的成员离开。

也就是说,270人的研发团队10人选择离开,核心部门离职率仅不到4%——这已经足够低。一组数据显示,OpenAI 前两年流失了超过25%的关键研究人才,他们大多跳槽去了Meta等竞争对手或自行创业。

首次打开融资大门,稳定军心

眼下创投圈尤其期待:谁能参与DeepSeek的首次融资?

4月开始,DeepSeek最先被爆正以超100亿美元的估值启动首轮外部融资。随后不过一周,消息称DeepSeek与腾讯阿里就投资展开洽谈。后来业内流传,DeepSeek投前估值3000亿人民币。

截至目前,DeepSeek未对融资消息做出任何回应。

一位FA告诉我们,近日在和投资机构接触关于DeepSeek融资的合作方式,本轮融资中财务投资机构极少。另一点也得到证实:腾讯与DeepSeek在日常业务上有沟通,但并无融资的实质性接洽。

一切仍扑朔迷离。

4月27日,DeepSeek注册资本由1000万元增加至1500万元,其中梁文锋认缴的注册资本由10万元增加到510万元,直接持股比例由1%升至34%,同步地,梁文锋控制的宁波程恩企业管理咨询合伙企业持股比例由99%下降至66%。此次变化后,梁文锋以间接、直接方式持有DeepSeek约84.29%股权。

值得注意的是,此前梁文锋通过宁波程恩持有DeepSeek绝大多数股权,直接持股极少,而这次变化后,梁文锋直接持股比例上升到34%。如此一来,梁文锋的控股权摆在了更容易被看见的位置——如果开展融资尽调,DeepSeek股权结构会显得更加清晰。

“不是绝大多数人能参与的”,投资人们由衷感叹。诚然,中国大模型江湖经历一番鏖战后,DeepSeek依旧很吸引人。

正如DeepSeek V4预览版终于亮相,Pro版和Flash版百万上下文标配,Pro版高至1.6万亿参数,价格感人:Pro每百万token输入1元(缓存命中)或 12元(缓存未命中),输出24元,Flash分别为0.2元、1元、2元。

与此同时,传闻的国产芯片适配证实,DeepSeek V4技术报告中,虽然能看出模型训练部分依然大概率用的英伟达芯片,但华为昇腾和英伟达并列写在验证平台,“预计下半年昇腾950超节点批量上市并部署之后,Pro版本的价格也会大幅度下调。”

这一举,意味着DeepSeek在英伟达坚固的CUDA生态敲开了一条裂缝。背后的想象力不言而喻。

梁文锋的笃定,国产AI时代真正开始

梁文锋和DeepSeek走在一条反共识的路上。

通常,一家明星科技公司的时钟是这样的:在崭露锋芒之际把握融资机会,伴随人才扩张和产品迭代加速,尽快占领市场并谋求上市。这个动作一旦串联起来,就很难停下来。

但DeepSeek的每个环节都出乎意料慢一些。

2025年初DeepSeek R1发布之际,梁文锋几乎没有对手,但锋芒毕露时他拒绝了所有前来叩门的投资人。偏偏是在竞争白热化、对手林立的今天,DeepSeek首次放出融资消息,外界讨论归因大多绕不开两点:研发需要资金,更深一层,DeepSeek需要给内部人才一个确定的估值。

产品迭代同样姗姗而来。DeepSeek V4发布距离上一版重大更新已经过去15个月,千呼万唤始出来,DeepSeek只先放出了V4预览版,一直视为缺憾的多模态也未同步更新。直到4月29日,DeepSeek才上线灰测识图模式,释放出多模态能力信号。

“不诱于誉,不恐于诽”,这是DeepSeek的姿态。而市场给出的反馈,似乎验证着某种事缓则圆。

DeepSeek V4发布当日,华为昇腾、寒武纪、海光信息、摩尔线程、沐曦股份、昆仑芯、平头哥真武、天数智芯等国产AI芯片就完成了适配。一时间,激起二级市场国产芯片上涨的“一池春水”。

与此同时,消息显示华为昇腾950系列AI芯片的市场需求大幅飙升,字节、腾讯、阿里三大国内头部互联网企业,已就新增芯片订单与华为展开接洽。

于是,一个反共识者,用慢节奏触发了产业共振——当底层芯片商与头部大厂开始围绕DeepSeek的标尺去咬合进化时,DeepSeek或许已经跃出了原本的竞争牌桌。

一如那句:慢就是快。只是,真正敢信的人并不多。

本文来自微信公众号 “投资界”(ID:pedaily2012),作者:冯雨晨

Lecturas Relacionadas

Two Legends Lost in Three Days: Is Google's AI Talent Dam Cracking?

In three days, Google lost two AI legends. On June 18, Noam Shazeer, co-author of the seminal "Attention is All You Need" paper and Gemini co-lead, left for OpenAI. Just 48 hours later, John Jumper, 2024 Nobel laureate and AlphaFold lead, departed DeepMind for Anthropic. This follows Andrej Karpathy joining Anthropic in May. These moves highlight a structural trend: top AI talent is concentrating at mission-driven, pre-IPO firms like OpenAI and Anthropic, while Google becomes a primary source. The exodus stems from a core mission mismatch. Google's ad-centric model often subordinates AI research to product and revenue goals, creating friction for pioneers like Shazeer, who returned in 2024 only to leave again. In contrast, OpenAI and Anthropic offer singular focus on pushing AI boundaries, whether towards AGI or safety-aligned models, which deeply appeals to top researchers like Jumper. Financial incentives amplify the pull. With both OpenAI and Anthropic nearing IPO, employees stand to gain immensely from equity, an upside Google's mature stock cannot match. Furthermore, the 2023 merger of Google Brain and DeepMind, intended to consolidate strength, has instead created cultural tension and slowed the path from research to product, as evidenced by Gemini's pace. This talent redistribution is reshaping the AI landscape. While Google retains vast data and compute resources, its true crisis is the quiet, continuous loss of the people who define the field's future. The real moat in AI is not infrastructure, but the concentration of brilliant minds—a battle Google is currently losing.

marsbitHace 52 min(s)

Two Legends Lost in Three Days: Is Google's AI Talent Dam Cracking?

marsbitHace 52 min(s)

Behind the AI Report Card, Lies a Chinese 'Exam Setter'

Beyond the familiar performance charts like MMLU-Pro and MMMU, which major AI models strive to ace, stands a key "examiner": Chinese-Canadian researcher Wenhu Chen. An assistant professor at the University of Waterloo and founder of TIGERLab, Chen addresses the crucial need for more rigorous AI evaluation. As models like GPT-4 began scoring near-perfect results on older benchmarks like MMLU, it became difficult to distinguish their true capabilities. In response, Chen introduced MMLU-Pro in 2024, featuring harder, more reasoning-focused questions with more answer choices, successfully reintroducing meaningful performance gaps. His work extends to multi-modal evaluation with MMMU and its enhanced version, MMMU-Pro. These benchmarks test a model's ability to understand and reason with complex information from images, charts, and text across diverse academic subjects, exposing the significant challenges even top models face in genuine comprehension. Chen's background in complex QA, table reasoning, and his experience at Google DeepMind on projects like Gemini inform his approach. He understands that effective benchmarks must anticipate how models might "cheat" by memorizing data or avoiding visual analysis. His lab also actively researches video understanding and generation models (e.g., UniVideo, Vamba), ensuring his evaluation work is grounded in practical model-building challenges. Now at Meta's Super Intelligence Lab, Chen continues his focus on multi-modal data and evaluation, representing the deep yet often unseen contributions of Chinese talent in shaping the fundamental tools of the AI industry.

marsbitHace 1 hora(s)

Behind the AI Report Card, Lies a Chinese 'Exam Setter'

marsbitHace 1 hora(s)

Alliance Co-founder's Letter to Entrepreneurs: Written at the Moment Cursor Sold for $600 Billion

Alliance Co-founder's Letter to Entrepreneurs: On Cursor's $60 Billion Sale Many aspiring founders see massive exits like Cursor's $60B sale and wonder why they can't achieve the same, often concluding opportunities are exhausted. But great companies aren't built in obvious, crowded spaces. Cursor, like Stripe, Figma, and Shopify before it, started with a non-consensus belief about the future. Before ChatGPT, they believed AI would transform knowledge work. They focused on a genuinely exciting domain, became their own customer, and obsessed over power users. Their journey involved years of "glass-chewing" effort before the market was ready. The pattern is consistent: identify a long-term technological shift, find a missed entry point, and execute for years before the trend becomes obvious. First-generation products (PayPal, Adobe, Amazon) prove a market exists. Second-generation winners (Stripe, Figma, Shopify) rebuild that market around new insights, technology, or changing customer behaviors. Founders must identify their phase in the cycle. Early entrants like Coinbase or Cursor focus on making new technology usable for power users. Later entrants find the "yin" to the established "yang"—the blind spots incumbents miss as they grow distant from individual users. The key is deep market immersion. Use every product in your space. Talk to users. Build an audience. Stop looking for ideas and start *seeing* them everywhere. Then, choose one. The idea must offer a 10x improvement or solve a "hair-on-fire" pain point—something severe enough that users are already crafting workarounds. When building, avoid feature bloat. Ask: why would someone switch? Great startups rarely force new behaviors; they improve familiar workflows with drastically lower friction (e.g., Cursor forked VS Code instead of creating a new editor). Distribution is the underestimated moat. Before product-market fit, achieve distribution-market fit. How do customers discover new tools? Founders like those at Airbnb, Stripe, and Cursor did unscalable, manual work to recruit early users. The final, unteachable ingredient is resilience. Cursor built for years pre-market, faced rejection, and persisted. So did Airbnb, Nvidia, and Rain (which launched post-FTX collapse). The lesson isn't that these founders were smarter, but that they stayed in the game long enough for their insights to compound. Framework: Spot technological cycles. Cultivate unique insight. Obsess over your market. Talk to customers. Find a hair-on-fire problem. Build the simplest wedge. Win your distribution channel. Above all, don't quit when it gets hard. Most people won't do these things consistently. The few who do build the next generation of great companies. Go build.

marsbitHace 1 hora(s)

Alliance Co-founder's Letter to Entrepreneurs: Written at the Moment Cursor Sold for $600 Billion

marsbitHace 1 hora(s)

Weekly Editor's Picks (0613-0619)

Weekly Editor's Picks (0613-0619): Market Insights & Analysis This weekly digest curates in-depth analysis often lost in the information flow, focusing on key insights across macro trends, investment, and technology. **Macro & Geopolitics:** With the Strait of Hormuz reopening and military conflict shifting to negotiation, markets are pivoting from "war shock" to "supply restoration." Trades include shorting crude risk premiums, longing airlines/tourism, Asian energy importers, and bond duration, while shorting inflation expectations. LNG, fertilizer, and chemical chains are also being repriced. **Investment & VC:** Ray Dalio advises against betting on concentrated AI giants dominating indices, advocating for diversified portfolios of high-quality, low-correlation assets instead. Analysis covers the 4-year crypto cycle, predicting the core surviving product by 2029 will be asset trading markets. Current BTC metrics suggest a potential bottoming zone, presenting a patient accumulation window. SpaceX's high-profile IPO at a $2.1T valuation faces scrutiny over fundamentals, with key watchpoints being its likely inclusion in the Nasdaq index and Q2 earnings. Concerns are raised about potential "gamma squeeze" and systemic risks if its narrative-driven valuation gets amplified by passive index funds. Robinhood (HOOD) is noted for breaking its high correlation with crypto, bolstered by its stock trading and new underwriting business. **Web3 & AI:** A warning highlights ~$1.8T in off-balance-sheet AI infrastructure commitments (purchase commitments, leases) as a potential systemic risk if AI monetization lags. AI models are being used for World Cup predictions, adding a new layer for betting markets. A cost breakdown of a $20 AI subscription reveals the supply chain from model companies to cloud, GPUs, and power. **Prediction Markets:** The emergence of prediction market "concept stocks" is noted, with Robinhood developing its own platform, Rothera, signaling a shift from market competition to a "channel war" for user access. **CeFi & DeFi:** The SpaceX IPO tested perpetual contract mechanisms for pre-IPO assets, highlighting challenges in handling corporate actions like stock splits on-chain. The de-pegging of STRC (Strategy's preferred share) to ~$89 reflects market concerns over MicroStrategy's capital structure and BTC-backed leverage model. BlackRock's covered-call Bitcoin ETF (BITA) offers yield but caps upside, appealing to yield-seeking institutions. **Ethereum:** An opinion piece argues Ethereum's core strength is its vast developer community and composability, solidifying its role as the default operating system for the financial internet. **Weekly Hot Topics:** Include the US-Iran deal reopening the Strait of Hormuz, Fed's hawkish hold, Anthropic restricting model access, SpaceX acquiring Cursor, and a humorous stock surge for "Liuliumei" due to its "LLM" ticker.

marsbitHace 1 hora(s)

Weekly Editor's Picks (0613-0619)

marsbitHace 1 hora(s)

Alliance's Co-Founder's Letter to Entrepreneurs: Written on the Occasion of Cursor's $60 Billion Sale

In this letter to entrepreneurs, Alliance reflects on the success of Cursor's $60 billion sale to Elon Musk, using it as a case study to counter the misconception that opportunities in crowded fields like AI or crypto are exhausted. The piece argues that great companies like Cursor, Stripe, Figma, and Shopify are not built by geniuses with perfect ideas, but by founders who start with a non-consensus belief about the future and build for years before that future becomes obvious to everyone. They identify long-term shifts, find overlooked entry points, and execute relentlessly. The framework for success involves: 1. **Identifying your place in the technology cycle**: Early-stage opportunities focus on making new tech usable for power users (e.g., Coinbase, Cursor). Later-stage opportunities involve finding the "yin" to an existing "yang"—the blind spots of first-generation players (e.g., Stripe vs. PayPal, Figma vs. Adobe). 2. **Cultivating unique insights**: Immerse yourself deeply in the market. Use every product, talk to users, and build an audience. Insights will emerge naturally from deep engagement. 3. **Finding a "hair-on-fire" problem**: Look for a 10x improvement or a severe, urgent pain point. The strongest signal is people already building clumsy workarounds. 4. **Building a focused MVP**: Don't just add features because you can. Ask why users would abandon their current tool for yours. The best startups rarely force new behaviors; they improve familiar workflows with drastically lower friction. 5. **Winning a distribution channel**: Distribution is often the moat. Before product-market fit, achieve channel-market fit. Find where your customers are and build an engine to reach them, even through unscalable, manual efforts initially. 6. **Persistence**: The final, unteachable ingredient is resilience. Success stories like Cursor, Airbnb, and Nvidia involved years of grinding, rejection, and perseverance when the path forward seemed unclear. The conclusion is that there is no secret. Most people fail to consistently execute these steps over the long term. The few who do build the companies that define the next era. The world is yours to create.

链捕手Hace 1 hora(s)

Alliance's Co-Founder's Letter to Entrepreneurs: Written on the Occasion of Cursor's $60 Billion Sale

链捕手Hace 1 hora(s)

Trading

Spot
Futuros
活动图片