Anthropic co‑founder Jack Clark says Silicon Valley can dial down the alarms over China’s latest AI darling, DeepSeek.
What Happened: Speaking on a Hill & Valley Forum panel hosted by Lightspeed’s Ravi Mhatre, Clark called the buzz “perhaps a bit overblown.”
He said DeepSeek’s new model looks “six to eight months behind” frontier systems from U.S. labs and is hobbled chiefly by scarce computing power.
Anthropic ran the same national‑security red‑team tests it uses internally and found the Chinese model “significantly behind” its own, Clark added. “Neither one is yet a concern for the Pentagon — they’re simply symptoms of progress we should track,” he said.
What To Know: DeepSeek, founded in 2023 in Hangzhou, grabbed headlines by open‑sourcing its R1 reasoning model, claiming it cost just $5.6 million to build — a figure Wedbush analyst Dan Ives has called “likely a fictional story” given the billions the West spends on comparable AI research.
Google DeepMind chief Demis Hassabis recently labeled R1 “the best work to come out of China,” but stressed it offered no real scientific breakthrough.
Even Nvidia Corp. NVDA, whose GPUs power most large models, framed DeepSeek as proof that compliant hardware can still yield competitive systems under U.S. export curbs. Yet performance gaps persist. Benzinga’s technical review last month concluded DeepSeek’s threat to U.S. dominance is “still overrated.”
Why It Matters: Technical benchmarks show DeepSeek still trails U.S. rivals, yet Mark Zuckerberg refuses to ignore China’s push to build AI infrastructure. He told the Dwarkesh Podcast last week that the United States must streamline “the ability to build data centers and produce energy” or risk falling behind.
Chinese tech giants, meanwhile, keep pouring money into AI and slashing prices to grab market share. Late last month Baidu Inc. BIDU rolled out its latest reasoning models, Ernie 4.5 Turbo and Ernie X1 Turbo, while Alibaba Group Holdings BABA and other peers unveiled developer‑friendly tools that cost individuals just $1 a year.
Image via Shutterstock