DeepSeek Anxiety: Why Nvidia's Selloff Looks Like A Big Overreaction

Seeking Alpha2025-01-28

SummaryTech stocks are plummeting due to concerns over DeepSeek, a new AI model from China, potentially disrupting Nvidia Corporation's GPU demand.DeepSeek's low-budget development and performance ...

网页链接

免责声明：本文观点仅代表作者个人观点，不构成本平台的投资建议，本平台不对文章信息准确性、完整性和及时性做出任何保证，亦不对因使用或信赖文章信息引发的任何损失承担责任。

精彩评论

我们需要你的真知灼见来填补这片空白

发表看法

{"i18n":{"language":"zh_CN"},"isChannel":false,"data":{"share":"https://www.laohu8.com/m/news/1175190263?lang=zh_CN&edition=full","thumbnail":"","is_english":true,"pubTime":"2025-01-28 13:52","share_image_url":"https://static.laohu8.com/e9f99090a1c2ed51c021029395664489","id":"1175190263","market":"us","top_or_hot":-1,"title":"DeepSeek Anxiety: Why Nvidia's Selloff Looks Like A Big Overreaction","media":"Seeking Alpha","content":"<div>\n<p>SummaryTech stocks are plummeting due to concerns over DeepSeek, a new AI model from China, potentially disrupting Nvidia Corporation's GPU demand.DeepSeek's low-budget development and performance ...</p>\n\n<a href=\"https://seekingalpha.com/article/4752241-deepseek-anxiety-why-nvidias-selloff-looks-like-a-big-overreaction\">网页链接</a>\n\n</div>\n","source":"lsy1728464409321","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>DeepSeek Anxiety: Why Nvidia's Selloff Looks Like A Big Overreaction</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: 11px; color: #7E829C; margin: 0;line-height: 11px;}\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\">\nDeepSeek Anxiety: Why Nvidia's Selloff Looks Like A Big Overreaction\n</h2>\n\n<h4 class=\"meta\">\n\n\n2025-01-28 13:52 GMT+8&nbsp;&nbsp;&nbsp;<a href=https://seekingalpha.com/article/4752241-deepseek-anxiety-why-nvidias-selloff-looks-like-a-big-overreaction><strong>Seeking Alpha</strong></a>\n\n\n</h4>\n\n</header>\n<article>\n<div>\n<p>SummaryTech stocks are plummeting due to concerns over DeepSeek, a new AI model from China, potentially disrupting Nvidia Corporation's GPU demand.DeepSeek's low-budget development and performance ...</p>\n\n<a href=\"https://seekingalpha.com/article/4752241-deepseek-anxiety-why-nvidias-selloff-looks-like-a-big-overreaction\">网页链接</a>\n\n</div>\n\n\n</article>\n</div>\n</body>\n</html>\n","isBrief":false,"type":0,"news_type":1,"symbol":"AMD","symbol_name":"美国超微公司","start_time":0,"source_url":"https://seekingalpha.com/article/4752241-deepseek-anxiety-why-nvidias-selloff-looks-like-a-big-overreaction","article_id":"1175190263","we_media_id":null,"thumbnails":[],"rights":null,"url":"https://stock-news.laohu8.com/highlight/detail?id=1175190263","pubTimestamp":1738043538,"columns":[],"sourceInfo":{"source_id":"lsy1728464409321","name":"Seeking Alpha"},"weMediaInfo":null,"summary":"I’ve seen a handful of new AI models emerge over the years, but the DeepSeek release is as mysterious as it gets. Despite being totally free and even currently sitting at the top spot in the free apps category on the App Store, “everyone” says it offers performance comparable to ChatGPT’s paid tiers. You can throw intricate coding questions at it, pepper it with advanced math problems, or ask for in-depth analyses, and the early feedback suggests it handles all of that without breaking a sweat. ","collect":0,"end_time":0,"defaultTopTitle":"seekingalpha.com","property":[],"viewcount":null,"language":"en","relate_stocks":{"AMD":"美国超微公司","MRVL":"迈威尔科技","TSM":"台积电","UMC":"联电","INTC":"英特尔","SMCI":"超微电脑","NXPI":"恩智浦","UCTT":"超科林半导体","ARM":"ARM Holdings","SOX":"费城半导体指数","STM":"意法半导体","AMAT":"应用材料","QCOM":"高通","MU":"美光科技","TXN":"德州仪器","GFS":"GLOBALFOUNDRIES Inc.","MCHP":"微芯科技","ON":"安森美半导体","AVGO":"博通","NVDA":"英伟达","ASML":"阿斯麦"},"translate_title":"DeepSeek焦虑：为什么英伟达的抛售看起来像是过度反应","themeId":"7c0849b3e4b9e2a90d64fbdac85e001a","isJumpTheme":false,"ttsUrl":null,"symbols_score_info":{"AMD":1.1,"NXPI":1.1,"AVGO":1.1,"STM":1.1,"ARM":1.1,"UCTT":1.1,"SOX":1.1,"ON":1.1,"ASML":1.1,"GFS":1.1,"AMAT":1.1,"SMCI":1.1,"NVDA":1.1,"TXN":1.1,"MCHP":1.1,"QCOM":1.1,"TSM":1.1,"INTC":1.1,"MU":1.1,"UMC":1.1,"MRVL":1.1},"content_text":"SummaryTech stocks are plummeting due to concerns over DeepSeek, a new AI model from China, potentially disrupting Nvidia Corporation's GPU demand.DeepSeek's low-budget development and performance comparable to top-tier AI models have spooked investors, questioning the necessity of high-end GPUs.Despite the panic, DeepSeek's real-world impact and scalability remain unproven, and regulatory issues could hinder its adoption in the West.Nvidia's diverse business and ongoing hardware innovations suggest the sell-off is an overreaction, with long-term bullish prospects intact.Indices are taking a beating as I am writing this on Monday morning, with tech stocks bearing the brunt of the selloff in a dramatic fashion. The sentiment on Wall Street seems to have fallen off a cliff due to rising concerns surrounding DeepSeek. It has a new AI large language model, or LLM, reportedly built by this small, relatively unknown team in China on a shoestring training budget of under $6 million.NVDA Stock Price (Google Finance)The group behind DeepSeek, rumored to be operating out of a startup in Shenzhen, claims they’ve managed to replicate the capabilities of leading large-scale models at a fraction of the usual cost. I gave the platform a try, and I have to say it really does behave much like what we’re used to from ChatGPT, Gemini, and other advanced language models.In that sense, it’s no shock to see investors selling off tech stocks across the board. Still, Nvidia Corporation seems to be in a particularly tight spot. Its stock is currently down by double digits, as it appears to be one of the biggest potential losers from DeepSeek’s reveal. Naturally, this concern stems from the fact that Nvidia’s GPUs power the training and operation of the bulk of the leading AI systems.The fear is that DeepSeek’s approach could disrupt the lucrative hardware demand that’s been critical to NVIDIA’s growth, making Wall Street wonder whether demand for high-end GPUs might shrink if startups can do more with less. Yet, it’s still very early. DeepSeek only launched a few days ago—and while the frenzy surrounding it has been intense, I am staying open-minded and leaning on the idea that the market has jumped the gun in hammering NVDA’s stock. So, let's address one key question at a time as they arise.Why Is The Tech Community Buzzing About DeepSeek?I’ve seen a handful of new AI models emerge over the years, but the DeepSeek release is as mysterious as it gets. Despite being totally free and even currently sitting at the top spot in the free apps category on the App Store, “everyone” says it offers performance comparable to ChatGPT’s paid tiers. You can throw intricate coding questions at it, pepper it with advanced math problems, or ask for in-depth analyses, and the early feedback suggests it handles all of that without breaking a sweat. Having tried it myself, I can confirm the same thoughts.Part of this frenzy is powered by the story behind DeepSeek’s development budget. The team claims they built and trained the entire system for less than $6 million, which is a fraction of the tens (if not hundreds) of millions that big names like OpenAI/Microsoft (MSFT), Google/Alphabet (GOOGL)(GOOG), and Meta (META) sink into AI training each quarter. Add to that the fairly maverick backstory: a small group of engineers quietly hacking away at memory optimization on unremarkable GPUs, ignoring the old adage that you absolutely need top-of-the-line hardware to compete. It’s got a David and Goliath narrative, and the tech community on X is eating it up.What Is DeepSeek's Approach?Now, at first glance, DeepSeek seems almost too good to be true. When you peek under the hood, though, you’ll find a couple of interesting tactics that, in hindsight, appear more clever than they do supernatural.One big question is how DeepSeek managed to get around the U.S. export restrictions limiting the power of GPUs shipped to China. For a while, it was theorized that maybe DeepSeek discovered a mysterious loophole in those regulations or used some off-the-radar black-market hardware. Well, it turns out they just squeezed every drop of efficiency out of the less capable hardware they could get their hands on.They systematically tweaked memory management so they never bumped into that dreaded performance ceiling. No, the H800 (a restricted version of Nvidia’s H100) didn't suddenly become as strong as the original. Instead, DeepSeek’s core algorithms were so optimized that limited chip-to-chip bandwidth didn’t end up throttling them as much as one might expect.Then there’s the matter of how DeepSeek pulled off training at a fraction of what other tech giants typically spend. The way I understand it is that they basically changed the rules of training itself. Instead of training every possible parameter for every token, they used an internal formula that predicted which tokens in the dataset would be activated by the model.They basically focused their training only on those slices, cutting out a massive chunk of wasted GPU cycles. By their estimates, each token update engaged only about 5% of the model’s total parameters, meaning they needed 95% fewer GPUs than a typical big-league setup. That’s an enormous cost saver, since GPU time is one of the most expensive line items in AI R&D.But there's more! The team reportedly developed a new method to compress the KV (key-value) cache, which is an essential part of how large language models “remember” and build context during inference. If you’re running an AI model on a shoestring, any trick to make inference less hungry in terms of GPU cycles can open the door to more experiments and faster iteration. Those extra experiments presumably let them reverse-engineer advanced model behaviors from competitors like ChatGPT Pro (regarding model o1) with reinforcement learning. They tested math and code questions that had easy-to-verify answers, and then updated their model if it spat out the correct solution. Over time, they shaped it into a direct competitor for highly specialized tasks.Worries on the Horizon for NvidiaWith everything I have covered so far, it’s no surprise that Nvidia has found itself in the crosshairs. For years, Nvidia has been riding a wave of soaring demand for GPUs as AI hype swept through every corner of tech. Whether it’s training large LLMs or powering cutting-edge generative AI workloads, you can bet that Nvidia’s hardware is front and center in the data centers of major players like OpenAI, Microsoft, Google, and Meta. The prevailing assumption has been that these large enterprises need the best chips to stay competitive, a trend that, as time goes by, gets repeatedly confirmed (evident in the ever-rising CAPEX on data centers/NVDA chips).Now DeepSeek arrives, chirping from the sidelines that they built a ChatGPT-class model for $6 million using hardware that skirts U.S. export restrictions. If you take that at face value, the immediate question is: “Wait, does that mean big companies don’t actually need all those expensive GPUs from Nvidia anymore?” Just last Friday, Meta CEO Mark Zuckerberg announced the company plans to invest $60 billion to $65 billion in CAPEX this year to build its AI infrastructure. Why would our companies in the West spend such insane amounts of money on infrastructure (which will need billions of dollars on NVDA chips) if AI is becoming a commodity and/or AI can be trained with just a few million dollars, and that's all it takes? Wall Street sees that scenario playing out, and it spooks everyone.That leads to another angle that’s unsettling for Nvidia. That is, export restrictions might not be as ironclad as originally thought. If the Chinese market can keep pushing forward with H800s or other restricted hardware but still match the performance of U.S.-based teams, that hurts the assumption that Nvidia’s latest and greatest chips are indispensable. Sure, in theory, H800s can’t scale up as smoothly as H100s due to reduced bandwidth. But if the software is cunning enough, it might not matter. For investors, that’s a big question mark hanging over the near-term outlook of NVDA stock.Reasons to Question DeepSeek’s Real ImpactHowever, before running off to dump your NVDA shares in a panic, I think it’s worth taking a deep breath and asking if DeepSeek is really all it’s cracked up to be. In my opinion, there’s nothing genuinely magical here. What we’re seeing are a couple of big cost-cutting “tricks”: a more selective training process for tokens and a slick compression approach for inference.Those breakthroughs allowed DeepSeek’s team to run more experiments in a short timespan, which sped up the reverse-engineering of advanced model behaviors. But that doesn’t automatically translate to mainstream, production-ready solutions that big companies can rely on. I spoke with people who build enterprise software, and they are especially cautious. Raw model performance is one thing. However, an AI model needs reliable support, guaranteed uptime, and compliance in areas like data privacy regulations. It’s questionable whether DeepSeek can offer those assurances at scale.There’s also the issue of censorship. DeepSeek reportedly has a fair amount of built-in content filters to align with Chinese regulations, and those constraints might make it less appealing for users outside that ecosystem. One user reported that DeepSeek censors its response in real-time as soon as Xi Jinping is mentioned! For this reason, I don't see companies in the West adopting an AI tool that’s subject to a foreign government’s speech guideline. In a world where data sovereignty and national security are hot-button issues, the prospect of leaning on a Chinese-born AI solution is absurd.The U.S. government probably isn’t going to sit back and watch if a Chinese AI model truly leaps ahead of its American counterparts in a way that threatens national security or the technological status quo. So it’s entirely possible that we’ll see an expanded set of chip restrictions or licensing hurdles if DeepSeek or similar projects start eating the U.S. giants’ lunch. Whether that’s good or bad policy is up for debate. However, the bottom line is that there are plenty of regulatory levers to pull that could slow or block the adoption of a foreign AI model in the U.S. That is, assuming DeepSeek can live up to American products.Now, from Nvidia’s perspective, the threat of a newly emergent AI competitor overshadowing them might be overblown for still another reason: the pace of AI hardware innovation is relentless. Even if DeepSeek’s approach seems revolutionary today, it won't be tomorrow. Big tech, along with Nvidia, are all investing massive sums in next-gen architectures, advanced interconnects, and specialized software stacks that make the most of every watt of power. Over time, those improvements might well outpace the gains DeepSeek achieved by going all-in on memory optimization and token pruning.Let’s not forget, also, that Nvidia’s business extends far beyond just training large language models. They’re involved in areas like autonomous vehicles, gaming, data center acceleration for enterprise applications, and other domains that continue to rely heavily on premium GPUs. So again, this whole sell-off seems like a major overreaction to me against the long-term bullish case.Final ThoughtsUltimately, there’s a lot we don’t know about DeepSeek. The model just came out, and real-world performance under rigorous enterprise conditions might tell a different story. If it were truly as polished as top-tier solutions from established firms, we’d expect to see big partnerships and immediate commercial adoption. That hasn’t happened yet. And due to regulatory and national securities issues, it shouldn't/won't happen.Meanwhile, the panic fueling the dramatic today's sell-off of NVDA stock feels more like knee-jerk fear than a measured response to the real data. There’s a good argument that all the excitement around DeepSeek will eventually cool off, leaving it as just one more interesting case study in AI model development—impressive in some respects, but hardly the GPU supply killer some have made it out to be.","kind":"news","is_publish_news":true,"is_publish_highlight":true,"is_publish_live":false,"is_publish_wemedia":null,"editions":null,"column":"","sentiment":"0","news_tag":"","news_rank":0,"isVideo":false,"video":null,"symbols":[],"gpt_button":0,"need_auth":false,"need_login_tip":false,"code":"91000000","status":"200"},"commentList":[],"isCommentEnd":true,"newsSizeData":{"likeSize":0,"commentSize":0,"repostSize":0,"favoriteSize":1,"likeStatus":false,"favoriteStatus":false},"APP":{"userAgent":"Mozilla/5.0 AppleWebKit/537.36 (KHTML, like Gecko; compatible; ClaudeBot/1.0; +claudebot@anthropic.com)","isDev":false,"isTTM":false,"tenantId":"TBCN","deviceId":"web-server-community-laohu8-v3","version":"4.42.2","shortVersion":"4.42.2","platform":"web","vendor":"web","appName":"laohu8","isIOS":false,"isAndroid":false,"isTiger":false,"isTHS":false,"isWeiXin":false,"isWeiXinMini":false,"isWeiBo":false,"isQQ":false,"isBaiduSwan":false,"isBaiduBox":false,"isDingTalk":false,"isToutiao":false,"isOnePlus":false,"isHuaWei":false,"isXiaomi":false,"isXiaomiWebView":false,"isOppo":false,"isVivo":false,"isSamsung":false,"isMobile":false},"href":"/m/news/1175190263","isCrawlerRequest":true}