Workflow
How China's new AI model DeepSeek is threatening U.S. dominance
GOOGLAlphabet(GOOGL) CNBC·2025-01-24 13:00

Core Insights - A Chinese AI lab named DeepSeek has developed a large-language model that outperforms leading models from American companies at a significantly lower cost and with less powerful hardware [1][2][3] - The advancements from DeepSeek have raised concerns about the diminishing global lead of the U.S. in artificial intelligence and questioned the substantial investments made by big tech in AI development [2][3] Company Developments - DeepSeek released a free, open-source large-language model in December, built in two months for under 6millionusingNvidia′sH800chips[1]−Thelab′snewreasoningmodel,r1,hasshownsuperiorperformancecomparedtoOpenAI′slatestmodelinvariousbenchmarktests[3]−DeepSeek′sabilitytonavigateU.S.semiconductorrestrictionssuggestseitherinnovativeworkaroundsorthattheexportcontrolsarelesseffectivethanintended[3]IndustryContext−OtherChinesecompanies,suchas01.aiandByteDance,arealsomakingsignificantstridesinAI,with01.aireportedlytrainedforonly6 million using Nvidia's H800 chips [1] - The lab's new reasoning model, r1, has shown superior performance compared to OpenAI's latest model in various benchmark tests [3] - DeepSeek's ability to navigate U.S. semiconductor restrictions suggests either innovative workarounds or that the export controls are less effective than intended [3] Industry Context - Other Chinese companies, such as 01.ai and ByteDance, are also making significant strides in AI, with 01.ai reportedly trained for only 3 million [5] - The competitive landscape indicates that necessity has driven Chinese firms to develop more efficient AI models despite restrictions [5]