DeepSeek changes rules of AI’s great game - FT中文网
登录×
电子邮件/用户名
密码
记住我
请输入邮箱和密码进行绑定操作:
请输入手机号码,通过短信验证(目前仅支持中国大陆地区的手机号):
请您阅读我们的用户注册协议隐私权保护政策,点击下方按钮即视为您接受。
FT商学院

DeepSeek changes rules of AI’s great game

Chinese start-up’s breakthrough may be bad news for US tech giants but could be a windfall for everyone else

It’s hard to talk about 21st-century economic history without discussing the “China shock”. That is the term often used to describe China’s entrance into the global market, a change that brought rich countries an abundance of cheap goods, but left entire industries and workforces mothballed.

DeepSeek may provide a sequel. A little-known Chinese hedge fund has thrown a grenade into the world of artificial intelligence with a large language model that, in effect, matches the market leader, Sam Altman’s OpenAI, at a fraction of the cost. And while OpenAI treats its models’ workings as proprietary, DeepSeek’s R1 wears its technical innards on the outside, making it attractive for developers to use and build on.

Things move faster in the AI age; terrifyingly so. Five of the biggest technology stocks geared to AI — chipmaker Nvidia and so-called hyperscalers Alphabet, Amazon, Microsoft and Meta Platforms — collectively shed almost $750bn of market value before US markets opened on Monday. It could be particularly grim for Nvidia if it proves true that DeepSeek won without the use of its shiniest chips.

Investors in tech companies — including Europeans such as chipmaker ASML, and energy companies that investors hoped would get a boost from fuelling data centres — are left wondering whether their investments will go up in smoke. The hyperscalers were due to plough almost $300bn into capital expenditure this year, according to Visible Alpha estimates. Analysts expect that on Wednesday, when they report earnings, Meta and Microsoft will report investment for 2024 totalling $94bn.

In truth, the game isn’t over. DeepSeek’s actual potential is still unclear, and it has yet to achieve “artificial general intelligence”, the humanlike state that Meta and OpenAI are pursuing. But the rules might have changed. At the very least, DeepSeek may take some of the US giants’ customers. At worst, it has challenged the core belief that more hardware is the key to better AI. That principle has underpinned the market value of Silicon Valley companies as they invest hand over fist.

What’s bad for the hyperscalers could still be a windfall for everyone else. For most business users, having the absolute best model is less important than having one that’s reliable and good enough. Not every driver needs a Ferrari. Advances in reasoning such as R1 could be a big step for “agents” that deal with customers and perform tasks in the workplace. If those are available more cheaply, corporate profitability should rise.

In that sense, this second China shock could resemble the first. It could bring not just destruction but a reshuffling — albeit a painful one for many. Researchers have estimated that for every job lost to the China shock, US households’ purchasing power rose by more than $400,000. The race for AI supremacy is on pause; the great giveaway has begun.

john.foley@ft.com

版权声明:本文版权归FT中文网所有,未经允许任何单位或个人不得转载,复制或以任何其他方式使用本文全部或部分,侵权必究。

印度的俄罗斯石油难题

面对特朗普的施压,莫迪要么接受美国的关税,要么从俄罗斯转向其他供应国,要么尝试与特朗普达成某种妥协。

梅德韦杰夫:从自由派希望到核威胁的漫长转变

这位俄罗斯前总统在社交媒体上公开挑衅特朗普。他已从俄罗斯自由派和白宫决策者眼中的白衣骑士,变成了莫斯科的“核战争狂人”。

数千公司董事在工党税改后离开英国

FT的分析发现,从去年10月到上月,有3790名公司董事报告离开英国,阿联酋成为首选目的地。

局长遭解雇前,美国劳工统计局就已然陷入困境

经费和人员削减阻碍了美国劳工统计局汇总重要报告的能力。

FT社评:特朗普对经济数据的攻击产生寒蝉效应

特朗普解雇美国劳工统计局局长会破坏关键国家统计数据的公信力。

特朗普的阿拉斯加州液化天然气项目未能打动亚洲盟友

日本和韩国抵制美国压力,拒绝将对管道项目的承诺纳入贸易协议。
设置字号×
最小
较小
默认
较大
最大
分享×