China's DeepSeek that wiped billions from US stock market in January 2025, sends New Year's message to American companies

2025年1月重创美股千亿市值，中国深度求索向美国企业发布新年宣言

Chinese AI startup DeepSeek started the year 2026 with the publication of research that could vindicate its earlier claims of training world-class models on a shoestring budget. That same approach has wiped nearly $600 billion off the market value of Nvidia in a single day last January. Furthermore, the release of chatbot last year, had made ChatGPT-maker issue companywide "code red," while also worrying other AI giants like Google, Anthropic, Meta, and more.

The Hangzhou-based company published a technical paper on January 1st that introduces "Manifold-Constrained Hyper-Connections" (mHC), a training method that will allow AI models to scale without ballooning computational costs. Co-authored by founder Liang Wenfeng, the breakthrough directly takes on the industry-wide consensus that the smarter AI is, the more computing power and chips it requires.

2026年初，中国人工智能初创企业深度求索（DeepSeek）发布重磅研究成果，为其早前“以极低成本训练世界级大模型”的主张提供了有力佐证。正是这一技术路径，曾在2025年1月单日重创英伟达，使其市值蒸发近6000亿美元。此外，该公司去年推出的智能聊天机器人，不仅迫使ChatGPT研发公司拉响全公司“最高警报”，更让谷歌、Anthropic、元宇宙等一众人工智能巨头倍感压力。

这家总部位于杭州的企业于1月1日发表技术论文，正式提出“流形约束超连接”（mHC）训练方法。这一技术可实现人工智能大模型的规模化升级，同时避免计算成本的急剧攀升。该论文由深度求索创始人梁文锋联合执笔，其核心突破直指行业共识——人工智能模型越智能，所需算力与芯片就越多。

The innovation arrives nearly a year after DeepSeek's R1 model sent shockwaves through Silicon Valley by matching ChatGPT's performance at reportedly a fraction of development costs. Back then, Nvidia scrambled to contain investor panic, insisting DeepSeek's efficiency gains would actually boost chip demand during AI "inference"—when models serve users at scale.

"Inference requires significant numbers of Nvidia GPUs," the chipmaker stated in January 2025, attempting to reassure markets even as its stock plummeted 17%. Nvidia acknowledged DeepSeek had used export-compliant H800 chips to achieve competitive results, raising uncomfortable questions about whether US export controls were achieving their intended effect.

DeepSeek's latest mHC paper provides technical depth behind those earlier claims. Testing models up to 27 billion parameters, the company's 19 researchers demonstrated "superior scalability" with "negligible computational overhead"—essentially proving one can build powerful AI without matching the massive chip purchases of OpenAI or Google.

早在近一年前，深度求索研发的R1模型就曾轰动硅谷。据称，该模型仅以几分之一的研发成本，便实现了与ChatGPT相当的性能表现。彼时，英伟达为平息投资者恐慌，紧急表态称深度求索的技术突破带来的效率提升，反而会推动人工智能“推理阶段”的芯片需求——即模型大规模向用户提供服务的阶段。

“推理环节需要大量英伟达图形处理器（GPU）支持。”2025年1月，英伟达在股价暴跌17%的背景下，以此言论安抚市场情绪。该公司同时承认，深度求索正是使用了符合美国出口管制规定的H800芯片，取得了具备竞争力的成果。此事也引发外界质疑：美国的芯片出口管制措施，是否达到了预期效果？

深度求索此次发布的mHC技术论文，为其早前的技术主张提供了详实的理论支撑。该公司19名研究人员针对参数规模最高达270亿的模型开展测试，证实mHC方法具备“卓越的可扩展性”，且“计算额外开销极低”。这一成果实质上证明，企业无需像开放人工智能公司（OpenAI）或谷歌那样斥巨资采购芯片，同样能打造出高性能人工智能模型。

以下是印度时报读者的评论：

MAROOF AZAM
Meanwhile in India ...govt and it's supporters are obsessed with a fictional god king ...which they can't even prove its exstence but still do all nonsense in his name .

再看看印度这边，政府和它的支持者，都沉迷于一个虚构的神王。连祂存在的证据都拿不出来，却还打着祂的旗号胡作非为。

Dfoxman
Yeah…And you and your kind can “prove” the exstence of your non-exstent quarter-moon god??? Get a life.

呵呵，那你和你这类人，倒是证明下你们那个压根不存在的弯月神啊？别在这瞎折腾了。

Peete
So easily you forget that the biggest global Virtual King whom the maxmum pray and kill for belongs to your clan ??!!!! Who bring all scientific developments to grinding halt ?! And then find excuses and faults with all others... Just to pick up a fight once again , and destroy all good things that ever happen ? Brainwashed brats !!!!! Introspect man!

你怎么这么快就忘了？全球最大的那个“虚拟神明”，多少人对着祂祈祷、为祂厮杀，祂不就是你们这边的吗？是谁让所有科学发展都停滞不前？是谁总找借口挑别人的刺？就是为了再挑起点争斗，毁掉所有好不容易变好的东西。一群被洗脑的蠢货！好好反省反省吧！

Aswani Kumar
Missed s point, in xyz sura of Quran, AI is already mentioned by cryptic numbers.

漏了个重点，某一章里，早就用神秘数字预言了人工智能。

BHARAT Patil
Ja bangladesh, pakistan mein jaaa

滚去孟加拉，滚去巴基斯坦！

Thakur
Ok otherwise in Congress’s govt… YOU would have made Deepseek. 10 years back right??? Do you know India also has its own quantum computer and how much govt funded that?? If no then don’t blame govt in any forms!!

行啊，那照你这么说，要是国大党执政，你们十年前就能造出深度求索了是吧？你知道印度也有自己的量子计算机吗？知道政府为它投了多少钱吗？不知道就别瞎指责政府！

Dilinmood
So, you better move to China

那你干脆搬去中国

Satendra Singh
You are just dragging India down with your skewed outlook

就是你们这种偏执的想法，在拖印度的后腿。

Alok
Never trust China! As simple as that.

永远别信中国！就这么简单。

lakshmikantha sugatoor
Suffering from anti-China bias.

典型的反华偏见。

Ajit Advani
Good to see strong competition to big tech dominance in AI. Hope for a breakthrough in bringing down inference costs and energy consumption as well.

很高兴看到人工智能领域的巨头垄断终于迎来强劲对手。也希望能在降低推理成本和能耗上取得突破。

Jay Boro
It's more of a smart chips games than a models. With Blockchain guys working on decentralisation models that can train in users phone itself, only research and other common high profile models will be trained in energy hungry data centers. So these new nose makes no difference to rich dudes. It's more about who can ship better chips to phones.

这根本就是智能芯片的博弈，不是模型的比拼。区块链领域的人都在研发去中心化模型，能直接在用户手机上完成训练。以后也就只有科研模型和一些热门大模型，需要在高耗能的数据中心里训练。所以这些新技术对那些大佬来说根本没影响，关键还是看谁能把更好的芯片装进手机。

Haniraj Chulani
DeepSeek givinga New Year gift to tech world!

深度求索这是给科技圈送了份新年大礼啊！

lopinion49
Becoz we Indians forget to innovate and are experts in Serving

因为我们印度人早就忘了要搞创新，就只会给人打工。

agw124
india can learn a lot, i think we need our major ai

印度能从中学到很多，我觉得我们得有自己的王牌人工智能。

zingVideozs
Have u seen algorithm of deep seek.. 99.9995% of Tcs , Infy developer can't understand nut or bolt of such complex steps..
I am a dev with 20 yrs exp.. many of my colleagues back when I was in service companies can't write basic program to solve simple problems..
India lost edge here..
Same with semiconductor manufacturing...u can't catchup with countries ahead and have many patients, research done

你见过深度求索的算法吗？塔塔咨询、印孚瑟斯99.9995%的程序员，连这种复杂算法的皮毛都搞不懂。

我是个有20年经验的开发者。我以前在服务外包公司的时候，好多同事连解决简单问题的基础程序都写不出来。

印度在这方面已经彻底没优势了。

半导体制造业也是一个样。你根本追不上那些领先的国家，人家早就手握大量专利和研究成果了。

ong5857
Yet another good talking, not surprised.
US: half talkers and half doers
China: all doers, no talkers
India: all talkers, no doers

又是一项亮眼成就，一点都不意外。

美国：一半人光说不练，一半人埋头实干

中国：全员埋头实干，从不夸夸其谈

印度：全员光说不练，没人踏实做事

AstaPutraBhava
China may have sent it's lunar probe in 2007 but in 2008 India sent it's Chandrayaan. nobody remembers about Made in China probe but everybody knows Chandrayaan. and now it's the same situation before us. even though we struggle now of lacking funds and resources in the begng to even compete among the bullies and looters of the world. the joy of arrogance is only temporary. haste makes waste. we may even take a couple of years more but it may be market killer for China all over the world. it's already happening to China in other fields like companies choosing to invest in India over China and especially in the border areas everybody know who's talking and who is doing

中国2007年确实发射了月球探测器，但印度2008年就推出了“月船一号”。没人记得中国制造的探测器，可全世界都知道印度的月船。现在我们面临的情况也是一样的。虽然我们现在还很艰难，起步阶段资金和资源都匮乏，甚至难以和那些恃强凌弱的掠夺者竞争。但傲慢带来的快感终究是短暂的，欲速则不达。我们或许还需要多花几年时间，但我们的成果终将在全球范围内成为中国的市场杀手。这种趋势在其他领域已经显现了，比如很多企业宁愿选择投资印度，也不选中国。尤其是在边境问题上，谁在光说不练，谁在付诸行动，大家都心知肚明。

继续阅读：深度求索   人工智能

此文由 三泰虎 编辑，未经允许不得转载！：首页 > 印度看中国 » 曾经重创美股千亿市值的中国深度求索，向美国企业发布新年宣言