Profile - To The Moon

帅的被罚款

Posts · 6Posts · 6
Following · 0Following · 0
Followers · 0Followers · 0

- 帅的被罚款帅的被罚款
  ·01-19
  Tree New Bee
  No business model-DeepSeek's strongest "moat"
  在全球AI巨头都被资本裹挟着必须赚钱的时候，DeepSeek是唯一一个可以不计成本、不看脸色、只为AGI梦想狂奔的“自由人”。幻方量化（去年回报率53%）超级“印钞机”兜底之下，DeepSeek是唯一的“零融资”异类。
  No business model-DeepSeek's strongest "moat"
  518 Comment
  1
  Share
  Report
- 帅的被罚款帅的被罚款
  ·2024-12-24
  $纳指100ETF(QQQ)$
  858 Comment
  1
  Share
  Report

Most Discussed

{"i18n":{"language":"en_US"},"isCurrentUser":false,"userPageInfo":{"id":"4109725158442240","uuid":"4109725158442240","gmtCreate":1646622089147,"gmtModify":1775557237746,"name":"帅的被罚款","pinyin":"sdbfkshuaidebeifakuan","introduction":"","introductionEn":"","signature":"","avatar":"https://community-static.tradeup.com/news/d682becea1117760ced228894c368faf","hat":"https://static.tigerbbs.com/f87c5cf2575fbb4ebd1b95bcedd4e74a","hatId":"shopping-0963876b8d03045537f0da210f83634a","hatName":"永涨头像框","vip":1,"status":2,"fanSize":0,"headSize":0,"tweetSize":6,"questionSize":0,"limitLevel":999,"accountStatus":2,"level":{"id":0,"name":"","nameTw":"","represent":"","factor":"","iconColor":"","bgColor":""},"themeCounts":0,"badgeCounts":0,"badges":[],"moderator":false,"superModerator":false,"manageSymbols":null,"badgeLevel":null,"boolIsFan":false,"boolIsHead":false,"favoriteSize":0,"symbols":null,"coverImage":null,"realNameVerified":"init","userBadges":[{"badgeId":"972123088c9646f7b6091ae0662215be-1","templateUuid":"972123088c9646f7b6091ae0662215be","name":"Elite Trader","description":"Total number of securities or futures transactions reached 30","bigImgUrl":"https://static.tigerbbs.com/ab0f87127c854ce3191a752d57b46edc","smallImgUrl":"https://static.tigerbbs.com/c9835ce48b8c8743566d344ac7a7ba8c","grayImgUrl":"https://static.tigerbbs.com/76754b53ce7a90019f132c1d2fbc698f","redirectLinkEnabled":0,"redirectLink":null,"hasAllocated":1,"isWearing":1,"stamp":null,"stampPosition":0,"hasStamp":0,"allocationCount":1,"allocatedDate":"2025.10.24","exceedPercentage":"60.70%","individualDisplayEnabled":0,"backgroundColor":null,"fontColor":null,"individualDisplaySort":0,"categoryType":1100},{"badgeId":"1026c425416b44e0aac28c11a0848493-2","templateUuid":"1026c425416b44e0aac28c11a0848493","name":"Senior Tiger","description":"Join the tiger community for 1000 days","bigImgUrl":"https://static.tigerbbs.com/0063fb68ea29c9ae6858c58630e182d5","smallImgUrl":"https://static.tigerbbs.com/96c699a93be4214d4b49aea6a5a5d1a4","grayImgUrl":"https://static.tigerbbs.com/35b0e542a9ff77046ed69ef602bc105d","redirectLinkEnabled":0,"redirectLink":null,"hasAllocated":1,"isWearing":0,"stamp":null,"stampPosition":0,"hasStamp":0,"allocationCount":1,"allocatedDate":"2024.12.02","exceedPercentage":null,"individualDisplayEnabled":0,"backgroundColor":null,"fontColor":null,"individualDisplaySort":0,"categoryType":1001},{"badgeId":"a83d7582f45846ffbccbce770ce65d84-1","templateUuid":"a83d7582f45846ffbccbce770ce65d84","name":"Real Trader","description":"Completed a transaction","bigImgUrl":"https://static.tigerbbs.com/2e08a1cc2087a1de93402c2c290fa65b","smallImgUrl":"https://static.tigerbbs.com/4504a6397ce1137932d56e5f4ce27166","grayImgUrl":"https://static.tigerbbs.com/4b22c79415b4cd6e3d8ebc4a0fa32604","redirectLinkEnabled":0,"redirectLink":null,"hasAllocated":1,"isWearing":0,"stamp":null,"stampPosition":0,"hasStamp":0,"allocationCount":1,"allocatedDate":"2024.10.17","exceedPercentage":null,"individualDisplayEnabled":0,"backgroundColor":null,"fontColor":null,"individualDisplaySort":0,"categoryType":1100}],"userBadgeCount":3,"currentWearingBadge":{"badgeId":"972123088c9646f7b6091ae0662215be-1","templateUuid":"972123088c9646f7b6091ae0662215be","name":"Elite Trader","description":"Total number of securities or futures transactions reached 30","bigImgUrl":"https://static.tigerbbs.com/ab0f87127c854ce3191a752d57b46edc","smallImgUrl":"https://static.tigerbbs.com/c9835ce48b8c8743566d344ac7a7ba8c","grayImgUrl":"https://static.tigerbbs.com/76754b53ce7a90019f132c1d2fbc698f","redirectLinkEnabled":0,"redirectLink":null,"hasAllocated":1,"isWearing":1,"stamp":null,"stampPosition":0,"hasStamp":0,"allocationCount":1,"allocatedDate":"2025.10.24","exceedPercentage":"60.87%","individualDisplayEnabled":0,"backgroundColor":null,"fontColor":null,"individualDisplaySort":0,"categoryType":1100},"individualDisplayBadges":null,"crmLevel":11,"crmLevelSwitch":0,"location":null,"starInvestorFollowerNum":0,"starInvestorFlag":false,"starInvestorOrderShareNum":0,"subscribeStarInvestorNum":0,"ror":null,"winRationPercentage":null,"showRor":false,"investmentPhilosophy":null,"starInvestorSubscribeFlag":false},"page":1,"watchlist":null,"tweetList":[{"id":523368408249008,"gmtCreate":1768810617365,"gmtModify":1768810621153,"author":{"id":"4109725158442240","authorId":"4109725158442240","name":"帅的被罚款","avatar":"https://community-static.tradeup.com/news/d682becea1117760ced228894c368faf","crmLevel":11,"crmLevelSwitch":0,"followedFlag":false,"idStr":"4109725158442240","authorIdStr":"4109725158442240"},"themes":[],"title":"","htmlText":"Tree New Bee","listText":"Tree New Bee","text":"Tree New Bee","images":[],"top":1,"highlighted":1,"essential":1,"paper":1,"likeSize":1,"commentSize":0,"repostSize":0,"link":"https://ttm.financial/post/523368408249008","repostId":"2604481920","repostType":4,"repost":{"id":"2604481920","kind":"highlight","weMediaInfo":{"introduction":"追踪全球财经热点，精选影响您财富的资讯，投资理财必备神器！","home_visible":1,"media_name":"华尔街见闻","id":"1084101182","head_image":"https://static.tigerbbs.com/66809d1f5c2e43e2bdf15820c6d6897e"},"pubTimestamp":1768726706,"share":"https://ttm.financial/m/news/2604481920?lang=en_US&edition=fundamental","pubTime":"2026-01-18 16:58","market":"sh","language":"zh","title":"No business model-DeepSeek's strongest \"moat\"","url":"https://stock-news.laohu8.com/highlight/detail?id=2604481920","media":"华尔街见闻","summary":"在全球AI巨头都被资本裹挟着必须赚钱的时候，DeepSeek是唯一一个可以不计成本、不看脸色、只为AGI梦想狂奔的“自由人”。幻方量化（去年回报率53%）超级“印钞机”兜底之下，DeepSeek是唯一的“零融资”异类。","content":"<p><html><head></head><body>When the global AI giants are wrapped up in capital and have to make money, DeepSeek is the only \"free man\" who can run wild for AGI dreams regardless of cost, face and face. Under the cover of High-Flyer (53% return last year) super \"money printing machine\", DeepSeek is the only \"zero financing\" outlier.</p><p>On the upcoming January 27th-the first anniversary of \"DeepSeek Moment\", the global AI community is looking forward to another big move from DeepSeek.</p><p>When the market was noisy and expecting, Kevin Xu, a well-known science and technology review author, published a long article analyzing DeepSeek's business model and moat, which caused great repercussions in the AI circle.</p><p><strong>He believes that DeepSeek's strongest moat is its unique model of zero external financing and no commercialization pressure. When the global AI giants are wrapped up in capital and have to make money, DeepSeek is the only \"free man\" who can run wild for AGI dreams regardless of cost, face and face.</strong></p><p><strong>Kevin S. Xu ( Xu Kaiwen)</strong>He is an independent observer and a well-known science and technology commentator who focuses on the intersection of technology and capital between China and the United States. He founded ChinaTalk podcast and Newsletter, which is quite influential in the professional circle, and is good at deconstructing the logic of technological evolution from the multi-dimensional dimensions of capital flow, organizational behavior and geopolitics.</p><p>We have sorted out the main points of the article for you:</p><p><ul style=\"\"><li><strong>Pouring cold water on market expectations</strong>: While everyone is waiting for DeepSeek's new model, the author bluntly says \"don't expect it to shock the world again like it did last year\". Because the current market has been fed by the \"open source model\", although DeepSeek fired the first shot, it is not the only or most open source player now (for example, it has not opened source data sets so far).</p><p></li><li><strong>The Only 'Zero Financing' Outlier</strong>: The current AI circle is a \"golden cave\", and even Musk has not held it. Recently, he has raised a huge sum of $20 billion for xAI. But Liang Wenfeng, the boss of DeepSeek, has so far insisted on \"zero external financing\", which is unique among all top laboratories. He values control over money.</p><p></li><li><strong>Super \"money printing machine\" pocket bottom</strong>: Liang Wenfeng can be so hard because the quantitative fund \"High-Flyer\" behind him is too profitable. Last year, Magic Square made more than 700 million dollars (about 5 billion yuan) in profit with a return of 53%. Liang Wenfeng directly used the money to buy graphics cards and recruit people. This mode of raising \"new dreams\" with \"old money\" made him not need to look at the faces of investors at all.</p><p></li><li><strong>Being rejected by VC is a blessing</strong>: In 2023, when DeepSeek was first established, it actually looked for financing, but at that time, domestic VCs were generally short-sighted and afraid to invest. As a result, this became a \"blessing in disguise\". The author believes that once you take the money of VC, you must carry the commercial KPI on your back, and your actions will be deformed. However, now DeepSeek only needs to be responsible for technology, not financial report.</p><p></li><li><strong>\"More Money\" More Trouble</strong>: This is a very deep insight – too much money breeds \"big company disease\". The author gives an ironic example: Some rich labs (such as Thinking Machines) even use them in office gyms<strong>Custom barbell pieces printed with their own Logo</strong>。 Behind this exaggerated pomp, employees stare at the \"paper wealth\" brought by options, and departments engage in \"palace fights\" in order to compete for computing resources. Because DeepSeek has no external valuation, the team is extremely flat and there is no internal consumption of resources.</p><p></li><li><strong>Computing power is not a panacea</strong>: The article quotes former OpenAI great god Ilya Sutskever —<strong>Truly disruptive innovation often does not require unlimited computing power</strong>。 The Transformer architecture ran out with only 8 to 64 cards back in the day. Too much money will make the team lazy, thinking that piling computing power can solve everything, thus losing its \"scientific research taste\".</p><p></li><li><strong>The ultimate paradox of investors</strong>Finally, the author let out a sigh from the perspective of an investor-as an investor, he dreams of investing in DeepSeek; But he knew in his heart,<strong>Once DeepSeek accepts outside investment, that pure quality will disappear</strong>。</p><p></li></ul>The author's original text is as follows (translated by AI):</p><p><strong>No Business Model: DeepSeek's Long-Term Advantage</strong></p><p>Authors : Kevin S. Xu</p><p>As the first anniversary of the \"DeepSeek Moment\" approaches on January 27, anticipation is growing for DeepSeek to release a more powerful new model ahead of the Lunar New Year (February 17).</p><p>However, excessive expectations are almost bound to bring disappointment. While I'm also eagerly awaiting this new model to see exactly what DeepSeek is up to, I don't expect it to shock the market again as it did a year ago. The whole AI industry and even the wider market are used to having a new \"open weight\" model released every month or two, especially models from Chinese laboratories, which tend to perform solidly, if not the most cutting-edge. With their open and free nature, these models are gaining traction everywhere in a way that is untraceable and untraceable — a virtue (or flaw) of open source technology as a whole. just as<a href=\"https://laohu8.com/S/NVDA\">NVIDIA</a>As CEO Huang Renxun likes to say, in terms of Token generation, the largest AI in the world is still OpenAI, but the second largest is open source AI.</p><p>DeepSeek may have started last year with the release of the V3 and R1<a href=\"https://laohu8.com/S/ME8U.SI\">MIT</a>License licensed the first shot of the open-source model, but other labs quickly followed up from competitors also from China<a href=\"https://laohu8.com/S/BABA\">Alibaba</a>And its Qwen (Tongyi Qianwen) model series, to the gpt-oss model launched by OpenAI. (Both use the Apache 2.0 license, another common open source permissive license.) As I pointed out earlier,<strong>If it weren't for DeepSeek, OpenAI probably never would have felt the pressure to deliver on the promises of its open source roots, if only lip service, such as releasing gpt-oss.</strong>While DeepSeek continues to open source the weights of its models and various tools and libraries, it is no longer the most open lab on the market, as it has never opened its dataset or master code base for training its models.</p><p>According to Artificial Analysis, the top three models in terms of global openness are currently from NVIDIA (Nemotron 3), Allen Institute (Olmo 3), and Mohammed Bin Zayed University of Artificial Intelligence (MBZUAI, K2-V2).</p><p><p class=\"t-img-caption\"><img src=\"https://static.tigerbbs.com/9db0d303a04711cb0df0229809a8f3ab\" tg-width=\"680\" tg-height=\"472\"/></p><p>As a result, DeepSeek's models are no longer the most capable of open models, the cheapest, or even the most open. So is there one more good reason for us to continue giving this laboratory that once shocked the world the unusual attention?</p><p>Yes, I think there is. But not because of its models or technological advancements, but because of its internal incentives and business models. Rather,<strong>We should look at DeepSeek because it has no business model</strong>— This is a unique and counter-intuitive feature, not a flaw, among all the cutting-edge laboratories in China, the United States and beyond.</p><p><h2 id=\"id_4024624920\">1. The AGI dream supported by self-financing</h2>This \"no business model advantage\" was mentioned when I first wrote about DeepSeek (also discussed on the ChinaTalk podcast at the time) weeks before DeepSeek had its \"highlight moment\" and became a household name. It is worth noting that when the whole AI world is inundated with capital, this advantage is so stubborn and lasting.</p><p>Liang Wenfeng did not raise any external funds. The lab does generate some revenue from its API services, but it is still constantly lowering its pricing. Liang Wenfeng was content to use the proceeds of his quantitative fund, High-Flyer, to fund the research and development of DeepSeek, which was originally incubated by High-Flyer. To be clear, Liang Wenfeng did try to raise venture capital from Chinese investors when he wanted to start DeepSeek in 2023, months after ChatGPT launched. But his \"AGI-pilled\" idealism, coupled with a lack of business plans, and the famous myopia and risk aversion of Chinese VCs, led to the failure of that fundraising effort — but it was a blessing in disguise.</p><p>As 2025 flies by, more money is pouring in to fund more AI advancements in order not to miss out on more potential advancements. Every lab with even a tiny bit of traction raises more money, and with it more expectations for future commercialization. A batch of so-called \"new AI Labs\" – pure research Labs led by star researchers and backed by VCs – have sprung up, such as Thinking Machines Lab (Mira Murati), SSI (Ilya Sutskever) and AMI Labs (Yann LeCun). Even xAI, which I used to draw the closest analogy to DeepSeek — because they similarly lacked the urgency of commercialization — succumbed to the temptation of external capital. A few days ago, Elon Musk's lab closed a $20 billion Series E funding round with a funding structure that includes equity and debt.</p><p>Even one of the richest people in the world can't say \"no\" to more money! Meanwhile, Liang Wenfeng continues to self-fund to maintain his mysterious shop and his AGI dream. Of course, Liang Wenfeng's quantitative fund performance last year<a href=\"https://laohu8.com/S/002296\">brilliant</a>, generating over $700 million in profits at a 53% return, which certainly helped tremendously. Presumably, the bulk of those profits have been invested in buying more GPUs (doing what they can despite facing U.S. export controls) and hiring talent to continue driving DeepSeek's research roadmap. Still, after DeepSeek gained global visibility and achieved 100% brand recognition in the AI industry, raising a billion or two dollars to accelerate its pace should have been a breeze. Whether you love it or hate it, everyone has heard of it.</p><p>However, saying \"no\" to external funding allows you to take control of your own destiny. If your self-determined destiny is \"committing to making AGI a reality…solving AGI mysteries with curiosity…answering essential questions with long-termism\" — that's DeepSeek's tagline on the HuggingFace page — then trading less money for total control is a worthwhile trade-off. Of course, you can also get there with some \"corporate governance innovations\". This is the case with Thinking Machines Lab, where a founder's 1 vote is worth 1 more than all other board members combined, thereby effectively having full control.<strong>But no matter how \"innovative\" you are at the board level, once you embrace VC, large-scale commercialization in order to generate VC-scale returns will be expected sooner or later.</strong></p><p>DeepSeek doesn't have any of these expectations. Therefore, it has no business model and does not need a business model. The path to AGI requires computing power, talent, and a fair dose of good research taste. Nobody says it has to require a business model.</p><p><h2 id=\"id_2633551634\">2. More money is more trouble, no money is no trouble</h2>Of course, the most common reason to raise more money is to buy more computing power to support research. However, what is not obvious is that more computing power is always necessary to produce good research results.</p><p>This insight is not unique to DeepSeek's unique self-funding architecture. Ilya Sutskever, the AI researcher with arguably the best research taste, agrees. He said in his own words on the Dwarkesh podcast:</p><p>\"The scale of computing power is already large enough that it is not obvious that you need that much extra computing power to prove an idea. Let me give you an analogy. AlexNet is built on two GPUs. That's the total computing power used for it. Transformer is built on eight to 64 GPUs. No Transformer paper in 2017 had experiments that used more than 64 GPUs. What is that equivalent to, two GPUs today? The same goes for ResNet, right? You can argue that o1 reasoning is not the most computing-powerful thing in the world.</p><p>So for research, you absolutely need a certain amount of computing power, but it's far from obvious that you need the absolute maximum amount of computing power to do the research. \"</p><p>On the other hand, by eliminating the strings attached to external funding, and the necessity of \"having to buy more computing power just because you have money\", two organizational-level advantages emerge. While DeepSeek has both.</p><p><strong>First, even if resources are more limited, there is no internal competition for resources.</strong>There is no bureaucracy, infighting, power struggle, or exchange of benefits to decide whether GPUs should support new product launches, expand the inferential needs of existing popular services, or be assigned to a new research idea. In a small lab with no external funding and no business model, good research taste and new ideas can be supported and embraced to the greatest extent, even if total computing power is limited on an absolute level.</p><p><strong>Secondly, there is less jealousy and hierarchical order based on compensation, perks, or how much computing power the team gets, which are often toxic but embedded in human nature.</strong>Your tissue is instantly flattened and can stay flat. With external funding comes not only the expectations of the business model, but also valuations, stock options, and all the glitz of a highly valued AI lab, no matter how meaningless that valuation was in the early days. Paper wealth from fantasy valuations tends to give way to inflated senses of self-worth, jealousy, and being poached by more well-funded competitors. These flashy decorations aren't just about personal wealth creation, they are tied to the lab's brand, prestige, and charm to attract more talent. And these decorations can become rather ridiculous, such as those barbell pieces in the lab's office gym with the \"Thinking Machines\" brand:</p><p><p class=\"t-img-caption\"><img src=\"https://static.tigerbbs.com/157333b07bbb73ecd453a22984cebf35\" tg-width=\"675\" tg-height=\"900\"/></p><p>Does DeepSeek have its own gym? I don't know. There might be. But I'm pretty sure it doesn't have the barbell piece of the brand printed on it. Does Ilya Sutskever wish he could self-fund to run SSI (which has raised $3 billion so far) like Liang Wenfeng? Probably. He may have the best instincts and research taste, but he can't stop poachers — even his co-founder CEO has been<a href=\"https://laohu8.com/S/META\">Meta</a>Zuckerberg, who has deeper pockets, poached. That's not to say that DeepSeek is immune to poaching. One of its star researchers, Luo Fuli, is now leading the more financially-powerful Lei Jun<a href=\"https://laohu8.com/S/XIACY\">Millet</a>The AI work.</p><p>In an industry inundated with money, and the intrigue and drama that money always brings, DeepSeek's absence of business model due to zero external funding is the source of its only lasting advantage:<strong>Internal maximization alignment achieved to serve AGI research, nothing else.</strong></p><p>I'm not even sure I believe in AGI. But as an investor and capitalist, I would love to be on DeepSeek's shareholder register. Having said that, if DeepSeek ever let me or any outside investor in, the traits that made DeepSeek DeepSeek would be completely ruined and no longer exist.</p><p></body></html></p>","collect":0,"html":"<!DOCTYPE html>\n<html>\n<head>\n<meta http-equiv=\"Content-Type\" content=\"text/html; charset=utf-8\" />\n<meta name=\"viewport\" content=\"width=device-width,initial-scale=1.0,minimum-scale=1.0,maximum-scale=1.0,user-scalable=no\"/>\n<meta name=\"format-detection\" content=\"telephone=no,email=no,address=no\" />\n<title>No business model-DeepSeek's strongest \"moat\"</title>\n<style type=\"text/css\">\na,abbr,acronym,address,applet,article,aside,audio,b,big,blockquote,body,canvas,caption,center,cite,code,dd,del,details,dfn,div,dl,dt,\nem,embed,fieldset,figcaption,figure,footer,form,h1,h2,h3,h4,h5,h6,header,hgroup,html,i,iframe,img,ins,kbd,label,legend,li,mark,menu,nav,\nobject,ol,output,p,pre,q,ruby,s,samp,section,small,span,strike,strong,sub,summary,sup,table,tbody,td,tfoot,th,thead,time,tr,tt,u,ul,var,video{ font:inherit;margin:0;padding:0;vertical-align:baseline;border:0 }\nbody{ font-size:16px; line-height:1.5; color:#999; background:transparent; }\n.wrapper{ overflow:hidden;word-break:break-all;padding:10px; }\nh1,h2{ font-weight:normal; line-height:1.35; margin-bottom:.6em; }\nh3,h4,h5,h6{ line-height:1.35; margin-bottom:1em; }\nh1{ font-size:24px; }\nh2{ font-size:20px; }\nh3{ font-size:18px; }\nh4{ font-size:16px; }\nh5{ font-size:14px; }\nh6{ font-size:12px; }\np,ul,ol,blockquote,dl,table{ margin:1.2em 0; }\nul,ol{ margin-left:2em; }\nul{ list-style:disc; }\nol{ list-style:decimal; }\nli,li p{ margin:10px 0;}\nimg{ max-width:100%;display:block;margin:0 auto 1em; }\nblockquote{ color:#B5B2B1; border-left:3px solid #aaa; padding:1em; }\nstrong,b{font-weight:bold;}\nem,i{font-style:italic;}\ntable{ width:100%;border-collapse:collapse;border-spacing:1px;margin:1em 0;font-size:.9em; }\nth,td{ padding:5px;text-align:left;border:1px solid #aaa; }\nth{ font-weight:bold;background:#5d5d5d; }\n.symbol-link{font-weight:bold;}\n/* header{ border-bottom:1px solid #494756; } */\n.title{ margin:0 0 8px;line-height:1.3;color:#ddd; }\n.meta {color:#5e5c6d;font-size:13px;margin:0 0 .5em; }\na{text-decoration:none; color:#2a4b87;}\n.meta .head { display: inline-block; overflow: hidden}\n.head .h-thumb { width: 30px; height: 30px; margin: 0; padding: 0; border-radius: 50%; float: left;}\n.head .h-content { margin: 0; padding: 0 0 0 9px; float: left;}\n.head .h-name {font-size: 13px; color: #eee; margin: 0;}\n.head .h-time {font-size: 12.5px; color: #7E829C; margin: 0;}\n.small {font-size: 12.5px; display: inline-block; transform: scale(0.9); -webkit-transform: scale(0.9); transform-origin: left; -webkit-transform-origin: left;}\n.smaller {font-size: 12.5px; display: inline-block; transform: scale(0.8); -webkit-transform: scale(0.8); transform-origin: left; -webkit-transform-origin: left;}\n.bt-text {font-size: 12px;margin: 1.5em 0 0 0}\n.bt-text p {margin: 0}\n</style>\n</head>\n<body>\n<div class=\"wrapper\">\n<header>\n<h2 class=\"title\">\nNo business model-DeepSeek's strongest \"moat\"\n</h2>\n<h4 class=\"meta\">\n<a class=\"head\" href=\"https://laohu8.com/wemedia/1084101182\">\n\n<div class=\"h-thumb\" style=\"background-image:url(https://static.tigerbbs.com/66809d1f5c2e43e2bdf15820c6d6897e);background-size:cover;\"></div>\n\n<div class=\"h-content\">\n<p class=\"h-name\">华尔街见闻 </p>\n<p class=\"h-time smaller\">2026-01-18 16:58</p>\n</div>\n</a>\n</h4>\n</header>\n<article>\n<p><html><head></head><body>When the global AI giants are wrapped up in capital and have to make money, DeepSeek is the only \"free man\" who can run wild for AGI dreams regardless of cost, face and face. Under the cover of High-Flyer (53% return last year) super \"money printing machine\", DeepSeek is the only \"zero financing\" outlier.</p><p>On the upcoming January 27th-the first anniversary of \"DeepSeek Moment\", the global AI community is looking forward to another big move from DeepSeek.</p><p>When the market was noisy and expecting, Kevin Xu, a well-known science and technology review author, published a long article analyzing DeepSeek's business model and moat, which caused great repercussions in the AI circle.</p><p><strong>He believes that DeepSeek's strongest moat is its unique model of zero external financing and no commercialization pressure. When the global AI giants are wrapped up in capital and have to make money, DeepSeek is the only \"free man\" who can run wild for AGI dreams regardless of cost, face and face.</strong></p><p><strong>Kevin S. Xu ( Xu Kaiwen)</strong>He is an independent observer and a well-known science and technology commentator who focuses on the intersection of technology and capital between China and the United States. He founded ChinaTalk podcast and Newsletter, which is quite influential in the professional circle, and is good at deconstructing the logic of technological evolution from the multi-dimensional dimensions of capital flow, organizational behavior and geopolitics.</p><p>We have sorted out the main points of the article for you:</p><p><ul style=\"\"><li><strong>Pouring cold water on market expectations</strong>: While everyone is waiting for DeepSeek's new model, the author bluntly says \"don't expect it to shock the world again like it did last year\". Because the current market has been fed by the \"open source model\", although DeepSeek fired the first shot, it is not the only or most open source player now (for example, it has not opened source data sets so far).</p><p></li><li><strong>The Only 'Zero Financing' Outlier</strong>: The current AI circle is a \"golden cave\", and even Musk has not held it. Recently, he has raised a huge sum of $20 billion for xAI. But Liang Wenfeng, the boss of DeepSeek, has so far insisted on \"zero external financing\", which is unique among all top laboratories. He values control over money.</p><p></li><li><strong>Super \"money printing machine\" pocket bottom</strong>: Liang Wenfeng can be so hard because the quantitative fund \"High-Flyer\" behind him is too profitable. Last year, Magic Square made more than 700 million dollars (about 5 billion yuan) in profit with a return of 53%. Liang Wenfeng directly used the money to buy graphics cards and recruit people. This mode of raising \"new dreams\" with \"old money\" made him not need to look at the faces of investors at all.</p><p></li><li><strong>Being rejected by VC is a blessing</strong>: In 2023, when DeepSeek was first established, it actually looked for financing, but at that time, domestic VCs were generally short-sighted and afraid to invest. As a result, this became a \"blessing in disguise\". The author believes that once you take the money of VC, you must carry the commercial KPI on your back, and your actions will be deformed. However, now DeepSeek only needs to be responsible for technology, not financial report.</p><p></li><li><strong>\"More Money\" More Trouble</strong>: This is a very deep insight – too much money breeds \"big company disease\". The author gives an ironic example: Some rich labs (such as Thinking Machines) even use them in office gyms<strong>Custom barbell pieces printed with their own Logo</strong>。 Behind this exaggerated pomp, employees stare at the \"paper wealth\" brought by options, and departments engage in \"palace fights\" in order to compete for computing resources. Because DeepSeek has no external valuation, the team is extremely flat and there is no internal consumption of resources.</p><p></li><li><strong>Computing power is not a panacea</strong>: The article quotes former OpenAI great god Ilya Sutskever —<strong>Truly disruptive innovation often does not require unlimited computing power</strong>。 The Transformer architecture ran out with only 8 to 64 cards back in the day. Too much money will make the team lazy, thinking that piling computing power can solve everything, thus losing its \"scientific research taste\".</p><p></li><li><strong>The ultimate paradox of investors</strong>Finally, the author let out a sigh from the perspective of an investor-as an investor, he dreams of investing in DeepSeek; But he knew in his heart,<strong>Once DeepSeek accepts outside investment, that pure quality will disappear</strong>。</p><p></li></ul>The author's original text is as follows (translated by AI):</p><p><strong>No Business Model: DeepSeek's Long-Term Advantage</strong></p><p>Authors : Kevin S. Xu</p><p>As the first anniversary of the \"DeepSeek Moment\" approaches on January 27, anticipation is growing for DeepSeek to release a more powerful new model ahead of the Lunar New Year (February 17).</p><p>However, excessive expectations are almost bound to bring disappointment. While I'm also eagerly awaiting this new model to see exactly what DeepSeek is up to, I don't expect it to shock the market again as it did a year ago. The whole AI industry and even the wider market are used to having a new \"open weight\" model released every month or two, especially models from Chinese laboratories, which tend to perform solidly, if not the most cutting-edge. With their open and free nature, these models are gaining traction everywhere in a way that is untraceable and untraceable — a virtue (or flaw) of open source technology as a whole. just as<a href=\"https://laohu8.com/S/NVDA\">NVIDIA</a>As CEO Huang Renxun likes to say, in terms of Token generation, the largest AI in the world is still OpenAI, but the second largest is open source AI.</p><p>DeepSeek may have started last year with the release of the V3 and R1<a href=\"https://laohu8.com/S/ME8U.SI\">MIT</a>License licensed the first shot of the open-source model, but other labs quickly followed up from competitors also from China<a href=\"https://laohu8.com/S/BABA\">Alibaba</a>And its Qwen (Tongyi Qianwen) model series, to the gpt-oss model launched by OpenAI. (Both use the Apache 2.0 license, another common open source permissive license.) As I pointed out earlier,<strong>If it weren't for DeepSeek, OpenAI probably never would have felt the pressure to deliver on the promises of its open source roots, if only lip service, such as releasing gpt-oss.</strong>While DeepSeek continues to open source the weights of its models and various tools and libraries, it is no longer the most open lab on the market, as it has never opened its dataset or master code base for training its models.</p><p>According to Artificial Analysis, the top three models in terms of global openness are currently from NVIDIA (Nemotron 3), Allen Institute (Olmo 3), and Mohammed Bin Zayed University of Artificial Intelligence (MBZUAI, K2-V2).</p><p><p class=\"t-img-caption\"><img src=\"https://static.tigerbbs.com/9db0d303a04711cb0df0229809a8f3ab\" tg-width=\"680\" tg-height=\"472\"/></p><p>As a result, DeepSeek's models are no longer the most capable of open models, the cheapest, or even the most open. So is there one more good reason for us to continue giving this laboratory that once shocked the world the unusual attention?</p><p>Yes, I think there is. But not because of its models or technological advancements, but because of its internal incentives and business models. Rather,<strong>We should look at DeepSeek because it has no business model</strong>— This is a unique and counter-intuitive feature, not a flaw, among all the cutting-edge laboratories in China, the United States and beyond.</p><p><h2 id=\"id_4024624920\">1. The AGI dream supported by self-financing</h2>This \"no business model advantage\" was mentioned when I first wrote about DeepSeek (also discussed on the ChinaTalk podcast at the time) weeks before DeepSeek had its \"highlight moment\" and became a household name. It is worth noting that when the whole AI world is inundated with capital, this advantage is so stubborn and lasting.</p><p>Liang Wenfeng did not raise any external funds. The lab does generate some revenue from its API services, but it is still constantly lowering its pricing. Liang Wenfeng was content to use the proceeds of his quantitative fund, High-Flyer, to fund the research and development of DeepSeek, which was originally incubated by High-Flyer. To be clear, Liang Wenfeng did try to raise venture capital from Chinese investors when he wanted to start DeepSeek in 2023, months after ChatGPT launched. But his \"AGI-pilled\" idealism, coupled with a lack of business plans, and the famous myopia and risk aversion of Chinese VCs, led to the failure of that fundraising effort — but it was a blessing in disguise.</p><p>As 2025 flies by, more money is pouring in to fund more AI advancements in order not to miss out on more potential advancements. Every lab with even a tiny bit of traction raises more money, and with it more expectations for future commercialization. A batch of so-called \"new AI Labs\" – pure research Labs led by star researchers and backed by VCs – have sprung up, such as Thinking Machines Lab (Mira Murati), SSI (Ilya Sutskever) and AMI Labs (Yann LeCun). Even xAI, which I used to draw the closest analogy to DeepSeek — because they similarly lacked the urgency of commercialization — succumbed to the temptation of external capital. A few days ago, Elon Musk's lab closed a $20 billion Series E funding round with a funding structure that includes equity and debt.</p><p>Even one of the richest people in the world can't say \"no\" to more money! Meanwhile, Liang Wenfeng continues to self-fund to maintain his mysterious shop and his AGI dream. Of course, Liang Wenfeng's quantitative fund performance last year<a href=\"https://laohu8.com/S/002296\">brilliant</a>, generating over $700 million in profits at a 53% return, which certainly helped tremendously. Presumably, the bulk of those profits have been invested in buying more GPUs (doing what they can despite facing U.S. export controls) and hiring talent to continue driving DeepSeek's research roadmap. Still, after DeepSeek gained global visibility and achieved 100% brand recognition in the AI industry, raising a billion or two dollars to accelerate its pace should have been a breeze. Whether you love it or hate it, everyone has heard of it.</p><p>However, saying \"no\" to external funding allows you to take control of your own destiny. If your self-determined destiny is \"committing to making AGI a reality…solving AGI mysteries with curiosity…answering essential questions with long-termism\" — that's DeepSeek's tagline on the HuggingFace page — then trading less money for total control is a worthwhile trade-off. Of course, you can also get there with some \"corporate governance innovations\". This is the case with Thinking Machines Lab, where a founder's 1 vote is worth 1 more than all other board members combined, thereby effectively having full control.<strong>But no matter how \"innovative\" you are at the board level, once you embrace VC, large-scale commercialization in order to generate VC-scale returns will be expected sooner or later.</strong></p><p>DeepSeek doesn't have any of these expectations. Therefore, it has no business model and does not need a business model. The path to AGI requires computing power, talent, and a fair dose of good research taste. Nobody says it has to require a business model.</p><p><h2 id=\"id_2633551634\">2. More money is more trouble, no money is no trouble</h2>Of course, the most common reason to raise more money is to buy more computing power to support research. However, what is not obvious is that more computing power is always necessary to produce good research results.</p><p>This insight is not unique to DeepSeek's unique self-funding architecture. Ilya Sutskever, the AI researcher with arguably the best research taste, agrees. He said in his own words on the Dwarkesh podcast:</p><p>\"The scale of computing power is already large enough that it is not obvious that you need that much extra computing power to prove an idea. Let me give you an analogy. AlexNet is built on two GPUs. That's the total computing power used for it. Transformer is built on eight to 64 GPUs. No Transformer paper in 2017 had experiments that used more than 64 GPUs. What is that equivalent to, two GPUs today? The same goes for ResNet, right? You can argue that o1 reasoning is not the most computing-powerful thing in the world.</p><p>So for research, you absolutely need a certain amount of computing power, but it's far from obvious that you need the absolute maximum amount of computing power to do the research. \"</p><p>On the other hand, by eliminating the strings attached to external funding, and the necessity of \"having to buy more computing power just because you have money\", two organizational-level advantages emerge. While DeepSeek has both.</p><p><strong>First, even if resources are more limited, there is no internal competition for resources.</strong>There is no bureaucracy, infighting, power struggle, or exchange of benefits to decide whether GPUs should support new product launches, expand the inferential needs of existing popular services, or be assigned to a new research idea. In a small lab with no external funding and no business model, good research taste and new ideas can be supported and embraced to the greatest extent, even if total computing power is limited on an absolute level.</p><p><strong>Secondly, there is less jealousy and hierarchical order based on compensation, perks, or how much computing power the team gets, which are often toxic but embedded in human nature.</strong>Your tissue is instantly flattened and can stay flat. With external funding comes not only the expectations of the business model, but also valuations, stock options, and all the glitz of a highly valued AI lab, no matter how meaningless that valuation was in the early days. Paper wealth from fantasy valuations tends to give way to inflated senses of self-worth, jealousy, and being poached by more well-funded competitors. These flashy decorations aren't just about personal wealth creation, they are tied to the lab's brand, prestige, and charm to attract more talent. And these decorations can become rather ridiculous, such as those barbell pieces in the lab's office gym with the \"Thinking Machines\" brand:</p><p><p class=\"t-img-caption\"><img src=\"https://static.tigerbbs.com/157333b07bbb73ecd453a22984cebf35\" tg-width=\"675\" tg-height=\"900\"/></p><p>Does DeepSeek have its own gym? I don't know. There might be. But I'm pretty sure it doesn't have the barbell piece of the brand printed on it. Does Ilya Sutskever wish he could self-fund to run SSI (which has raised $3 billion so far) like Liang Wenfeng? Probably. He may have the best instincts and research taste, but he can't stop poachers — even his co-founder CEO has been<a href=\"https://laohu8.com/S/META\">Meta</a>Zuckerberg, who has deeper pockets, poached. That's not to say that DeepSeek is immune to poaching. One of its star researchers, Luo Fuli, is now leading the more financially-powerful Lei Jun<a href=\"https://laohu8.com/S/XIACY\">Millet</a>The AI work.</p><p>In an industry inundated with money, and the intrigue and drama that money always brings, DeepSeek's absence of business model due to zero external funding is the source of its only lasting advantage:<strong>Internal maximization alignment achieved to serve AGI research, nothing else.</strong></p><p>I'm not even sure I believe in AGI. But as an investor and capitalist, I would love to be on DeepSeek's shareholder register. Having said that, if DeepSeek ever let me or any outside investor in, the traits that made DeepSeek DeepSeek would be completely ruined and no longer exist.</p><p></body></html></p>\n</article>\n</div>\n</body>\n</html>\n","type":0,"thumbnail":"https://static.tigerbbs.com/a2d10971e5f3c03dc2f148badda342d3","relate_stocks":{"BK4023":"应用软件","BK4585":"ETF&股票定投概念","BK4528":"SaaS概念","AI":"C3.ai, Inc.","BK4543":"AI","BK4551":"寇图资本持仓","BK4587":"ChatGPT概念","BK4588":"碎股"},"source_url":"https://wallstreetcn.com/articles/3763534","is_english":false,"share_image_url":"https://static.laohu8.com/e9f99090a1c2ed51c021029395664489","article_id":"2604481920","content_text":"在全球AI巨头都被资本裹挟着必须赚钱的时候，DeepSeek是唯一一个可以不计成本、不看脸色、只为AGI梦想狂奔的“自由人”。幻方量化（去年回报率53%）超级“印钞机”兜底之下，DeepSeek是唯一的“零融资”异类。在即将到来的1月27日——也就是“DeepSeek时刻”一周年之际，全球AI社区都期待DeepSeek再来个大招。市场喧嚣期待之时，知名科技评论作者Kevin Xu发布长文对DeepSeek的商业模式、护城河进行了分析，在AI圈内引起了很大反响。他认为，DeepSeek最坚固的护城河，是它零外部融资、无商业化压力的独特模式。在全球AI巨头都被资本裹挟着必须赚钱的时候，DeepSeek是唯一一个可以不计成本、不看脸色、只为AGI梦想狂奔的“自由人”。Kevin S. Xu（徐凯文）是专注中美科技与资本交叉领域的独立观察者，知名科技评论人。他创办的ChinaTalk播客及Newsletter在专业圈层颇具影响力，擅长从资本流向、组织行为与地缘政治多维度解构技术演进逻辑。我们给大家梳理了一下文章要点：给市场预期泼盆冷水：虽然大家都在等DeepSeek的新模型，但作者直言“不要指望它能像去年那样再次震惊世界”。因为现在的市场已经被“开源模型”喂饱了，DeepSeek虽然打响了第一枪，但现在并不是唯一、也不是最开源的玩家了（比如它至今没开源数据集）。唯一的“零融资”异类：现在的AI圈就是“销金窟”，连马斯克都没扛住，最近也给xAI融了一笔高达200亿美元的巨资。但DeepSeek的老板梁文锋至今坚持“零外部融资”，这在所有顶尖实验室里是独一份的，他把控制权看得比钱重。超级“印钞机”兜底：梁文锋能这么硬气，是因为他背后的量化基金“幻方量化”太赚钱了。去年，幻方以53%的回报率赚了超7亿美元（约50亿人民币）的利润。梁文锋直接把这些钱拿来买显卡、招人，这种用“老钱”养“新梦”的模式，让他完全不需要看投资人的脸色。被VC拒绝反而是福报：2023年DeepSeek刚成立时其实找过融资，但当时国内VC普遍短视、不敢投，结果这反而成了“塞翁失马”。作者认为，一旦拿了VC的钱，就必须背上商业化KPI，动作就会变形，而现在的DeepSeek只需要对技术负责，不需要对财报负责。“钱多”麻烦多：这是一个非常深刻的洞察——钱太多会滋生“大公司病”。作者举了个很讽刺的例子：有些富得流油的实验室（比如Thinking Machines），甚至在办公室健身房里用上了印着自家Logo的定制杠铃片。这种浮夸的排场背后，是员工盯着期权带来的“纸面富贵”，部门之间为了争夺算力资源搞“宫斗”。而DeepSeek因为没有外部估值，团队极度扁平，没有资源内耗。算力不是万能药：文章引用了前OpenAI大神Ilya Sutskever的观点——真正的颠覆性创新往往不需要无限堆算力。Transformer架构当年只用了8到64张卡就跑出来了。钱太多反而会让团队变懒，以为堆算力就能解决一切，从而失去了“科研品味”。投资人的终极悖论：最后，作者站在一位投资人的角度发出了一声叹息——作为一个投资人，他做梦都想投DeepSeek；但他心里也清楚，一旦DeepSeek接受了外部投资，那种纯粹的特质就会消失了。作者原文如下（由AI翻译）：没有商业模式：DeepSeek的长期优势作者：Kevin S. Xu随着1月27日“DeepSeek时刻（DeepSeek Moment）”一周年纪念日的临近，市场对于DeepSeek在农历新年（2月17日）前发布一款更强大新模型的期待正日益高涨。然而，过高的期望几乎注定会带来失望。虽然我也急切地等待着这款新模型，想看看DeepSeek究竟在忙些什么，但我并不预期它会像一年前那样再次震惊市场。整个AI行业乃至更广泛的市场，已经习惯了每隔一两个月就有一款新的“开放权重”模型发布，尤其是来自中国实验室的模型，它们即便算不上最尖端，也往往表现扎实。凭借着开放和免费的特性，这些模型正以一种无法追踪和溯源的方式在各地获得牵引力——这也是开源技术整体上的一种美德（或者说是缺陷）。正如英伟达CEO黄仁勋喜欢说的那样，以Token生成量计算，世界上最大的AI仍然是OpenAI，但第二大的，是开源AI。DeepSeek或许在去年通过发布V3和R1打响了MIT License许可的开源模型的第一枪，但其他实验室迅速跟进，从同样来自中国的竞争对手阿里巴巴及其Qwen（通义千问）模型系列，到OpenAI推出的gpt-oss模型。（两者均使用Apache 2.0许可，这是另一种常见的开源宽松许可。）正如我之前指出的，如果不是因为DeepSeek，OpenAI可能永远不会感受到压力，从而去兑现其开源根源的承诺，哪怕只是口头上的，比如发布gpt-oss。虽然DeepSeek继续开源其模型的权重以及各种工具和库，但它已不再是市场上最开放的实验室，因为它从未开放过其用于训练模型的数据集或主代码库。根据Artificial Analysis的数据，目前全球开放程度排名前三的模型分别来自NVIDIA（Nemotron 3）、Allen Institute（Olmo 3）和穆罕默德·本·扎耶德人工智能大学（MBZUAI，K2-V2）。因此，DeepSeek的模型不再是开放模型中能力最强的，不再是最便宜的，甚至也不再是最开放的。那么，是否还有一个充分的理由，让我们继续给予这个曾经震惊世界的实验室超乎寻常的关注？是的，我认为有。但不是因为它的模型或技术进步，而是因为其内部的激励机制和商业模式。更确切地说，我们应该关注DeepSeek，因为它没有商业模式——在中国、美国及其他地区的所有前沿实验室中，这是一个独特且反直觉的特性，而非缺陷。1. 自筹资金支撑的 AGI 梦想早在DeepSeek迎来其“高光时刻”并成为家喻户晓的名字之前的几周，我首次撰写关于DeepSeek的文章时（当时也在ChinaTalk播客上讨论过），就提到了这种“无商业模式优势”。值得注意的是，当整个AI世界都被资本淹没时，这种优势竟表现得如此顽固且持久。梁文锋没有筹集任何外部资金。该实验室确实通过其API服务产生一些收入，但仍在不断降低定价。梁文锋满足于用他的量化基金——幻方量化的收益来资助DeepSeek的研发，正是幻方量化最初孵化了DeepSeek。需要明确的是，梁文锋在2023年ChatGPT推出几个月后想要创办DeepSeek时，确实曾尝试从中国投资者那里筹集风险投资。但他那“笃信AGI（AGI-pilled）”的理想主义，加上缺乏商业计划书，以及中国VC著名的短视和风险厌恶，导致了那次融资努力的失败——但这反而塞翁失马，焉知非福。随着2025年的飞逝，为了不错过更多潜在的进步，越来越多的资金涌入，资助了更多的AI进展。每一个哪怕只有一点点牵引力的实验室都筹集了更多的钱，随之而来的是对未来商业化的更多期望。一批所谓的“AI新实验室”——由明星研究员领导的、风投支持的纯研究实验室——如雨后春笋般涌现，例如Thinking Machines Lab（Mira Murati）、SSI（Ilya Sutskever）和AMI Labs（Yann LeCun）。甚至连我过去常拿来与DeepSeek做最紧密类比的xAI——因为它们同样缺乏商业化的紧迫性——也屈服于外部资本的诱惑。几天前，埃隆·马斯克（Elon Musk）的实验室完成了一轮200亿美元的E轮融资，资金结构包含了股权和债权。即便是世界上最富有的人之一，也无法对更多的钱说“不”！与此同时，梁文锋继续自筹资金维持他那神秘的店铺和他的AGI梦想。当然，梁文锋的量化基金去年业绩辉煌，以53%的回报率创造了超过7亿美元的利润，这无疑提供了巨大帮助。据推测，这些利润的大部分都投入到了购买更多GPU（尽管面临美国的出口管制，但尽其所能）和招聘人才上，以继续推动DeepSeek的研究路线图。尽管如此，在DeepSeek获得全球知名度并在AI行业实现100%的品牌认知度之后，筹集一二十亿美元以加速其步伐本应是轻而易举的事。无论你爱它还是恨它，每个人都听说过它。然而，对外部资金说“不”，让你能够掌控自己的命运。如果你的自定命运是“致力于让AGI成为现实……用好奇心解开AGI之谜……用长期主义回答本质问题”——这是DeepSeek在HuggingFace页面上的标语——那么用较少的资金换取完全的控制权是一个值得的权衡。当然，你也可以通过一些“公司治理创新”来达到目的。Thinking Machines Lab就是这种情况，创始人的1票投票权价值比所有其他董事会成员的票数总和还多1票，从而实际上拥有完全控制权。但无论你在董事会层面如何“创新”，一旦你接受了风险投资，为了产生风投规模的回报而进行的大规模商业化，迟早是会被期待的。DeepSeek没有任何这些预期。因此，它没有商业模式，也不需要商业模式。通往AGI的道路需要算力、人才，以及相当剂量的良好研究品味。没人规定它必须需要一个商业模式。2. 钱多麻烦多，没钱没烦恼当然，筹集更多资金最常见的理由是为了购买更多的算力来支持研究。然而，并不显见的是，要产出好的研究成果，更多的算力总是必要的。这一洞察并非DeepSeek那独特的自筹资金架构所独有。Ilya Sutskever，这位可以说是拥有最佳研究品味的AI研究员，也表示同意。他在Dwarkesh播客中用自己的话说道：“算力规模已经足够大，以至于并不显见你需要那么多额外的算力来证明某个想法。我给你打个比方。AlexNet是建立在两块GPU上的。那是用于它的总算力。Transformer是建立在8到64块GPU上的。2017年没有哪篇Transformer论文的实验使用了超过64块GPU，这相当于什么，今天的两块GPU？ResNet也是如此，对吧？你可以争辩说，o1推理模型（o1 reasoning）并不是世界上最耗算力的东西。因此，对于研究来说，你绝对需要一定量的算力，但远非显而易见的是，你需要绝对最大量的算力来进行研究。”另一方面，通过消除外部资金的附带条件，以及“仅仅因为有钱就必须购买更多算力”的必要性，两种组织层面的优势显现出来。而DeepSeek两者兼备。首先，即使资源更加有限，内部也不存在资源竞争。 这里没有官僚主义、内斗、权力斗争，也没有为了决定GPU是应该支持新产品发布、扩展现有流行服务的推理需求，还是分配给一个新的研究想法而进行的利益交换。在一个没有外部资金、没有商业模式的小型实验室里，良好的研究品味和新想法可以得到最大程度的支持和拥抱，即使绝对层面上的总算力是有限的。其次，基于薪酬、津贴或团队获得的算力多少而产生的嫉妒和等级秩序更少，而这些往往是有毒的，却根植于人性之中。 你的组织瞬间变得扁平，并且可以保持扁平。伴随外部资金而来的不仅是商业模式的期望，还有估值、股票期权，以及一个高估值AI实验室所有的浮华装饰，无论这种估值在早期是多么没有意义。来自梦幻估值的纸面财富往往会让位于膨胀的自我价值感、嫉妒，以及被资金更雄厚的竞争对手挖角。这些浮华装饰不仅仅关乎个人财富创造，还与实验室的品牌、声望和吸引更多人才的魅力捆绑在一起。而这些装饰可能会变得相当荒谬，比如该实验室办公室健身房里那些印有“Thinking Machines”品牌的杠铃片：DeepSeek有自己的健身房吗？我不知道。可能有。但我很确定它没有印着品牌的杠铃片。Ilya Sutskever是否希望他能像梁文锋一样自筹资金来运营SSI（迄今已筹集30亿美元）？可能吧。他可能拥有最好的直觉和研究品味，但他无法阻止挖角者——甚至他的联合创始人CEO也被Meta财力更雄厚的扎克伯格挖走了。这并不是说DeepSeek对挖角免疫。其明星研究员之一罗福莉，现在正领导着财力更雄厚的雷军旗下小米的AI工作。在一个被金钱淹没的行业，以及金钱总是带来的阴谋和戏剧性中，DeepSeek因零外部资金而导致的商业模式缺失，是其唯一持久优势的源泉：为服务于AGI研究而实现的内部最大化对齐，别无其他。我甚至不确定我是否相信AGI。但作为一名投资者和资本家，我很乐意进入DeepSeek的股东名册。话虽如此，如果DeepSeek曾经让我或任何外部投资者进入，那么让DeepSeek之所以成为DeepSeek的特质，将彻底毁掉，不复存在。","news_type":1,"symbols_score_info":{"AI":0.9}},"isVote":1,"tweetType":1,"viewCount":518,"authorTweetTopStatus":1,"verified":2,"comments":[],"imageCount":0,"langContent":"EN","totalScore":0},{"id":384889695179016,"gmtCreate":1734986719883,"gmtModify":1734989728439,"author":{"id":"4109725158442240","authorId":"4109725158442240","name":"帅的被罚款","avatar":"https://community-static.tradeup.com/news/d682becea1117760ced228894c368faf","crmLevel":11,"crmLevelSwitch":0,"followedFlag":false,"idStr":"4109725158442240","authorIdStr":"4109725158442240"},"themes":[],"htmlText":"<a href=\"https://ttm.financial/S/QQQ\">$纳指100ETF(QQQ)$</a>&nbsp;","listText":"<a href=\"https://ttm.financial/S/QQQ\">$纳指100ETF(QQQ)$</a>&nbsp;","text":"$纳指100ETF(QQQ)$","images":[{"img":"https://community-static.tradeup.com/news/7db092aa6bc329cb034446aa34b44a05","width":"972","height":"947"}],"top":1,"highlighted":1,"essential":1,"paper":1,"likeSize":1,"commentSize":0,"repostSize":0,"link":"https://ttm.financial/post/384889695179016","isVote":1,"tweetType":1,"viewCount":858,"authorTweetTopStatus":1,"verified":2,"comments":[],"imageCount":1,"langContent":"EN","totalScore":0}],"defaultTab":"posts","isTTM":true}