China’s new AI model is threatening America’s industry dominance
Image: Shutterstock
The US has the Miracle on Ice – and now, China has the Miracle on AI-ce.
Silicon Valley is buzzing over a pair of new AI models developed by Chinese researchers that match or exceed the performance of the most advanced systems from top American companies, despite using inferior chips and far fewer resources.
Some quick background: DeepSeek, a relatively unknown Chinese AI research lab, released its next-gen chatbot model called V3 late last month. In third-party benchmark tests, V3 matches the capabilities of OpenAI's GPT-4o and Anthropic's Claude Sonnet 3.5 – while outperforming other top competitors, such as Meta's Llama 3.1.
- Then last week, DeepSeek introduced R1, a reasoning-focused model that also performs on-par with or better than OpenAI’s top reasoning model, o1, in many of the same third-party tests.
But the hype is about more than impressive results. DeepSeek’s models have received praise from several top Silicon Valley executives – including a16z founder Marc Andreessen and Microsoft CEO Satya Nadella – for two main reasons: 1) they’re far cheaper to build and run than competitors, and 2) they’re completely free and open source.
- DeepSeek says its V3 chatbot was trained in two months for just $5.6 million using reduced-capability chips from Nvidia, compared to OpenAI’s $100+ million price tag for training GPT-4. DeepSeek’s R1 model also runs queries at ~$0.14 per million tokens, or 98% cheaper than OpenAI's o1 reasoning model ($7.50).
- Unlike most leading AI models, which are proprietary, both V3 and R1's code and training methods are completely open source under the MIT license. This means anyone can grab the model, use it, and modify it without restrictions.
Big picture: As of this weekend, DeepSeek’s V3 and R1 models were both ranked in the top 10 on Chatbot Arena, a chatbot performance rating platform hosted by UC Berkeley researchers. That’s ahead of competitors like Anthropic’s Claude and Meta’s Llama – but behind the two top-rated models, Google’s Gemini and OpenAI’s GPT-4o.
đź“Š Flash poll: Do you regularly use AI models like ChatGPT, Gemini, or Claude?
See a 360° view of what pundits are saying →
Sprinkles from the Left
- Some commentators argue that superior US control over critical AI hardware in the future may give rivals little chance to compete – but DeepSeek’s recent success suggests open-source AI can drive innovation through creativity, rather than brute processing power.
- Others contend that AI is vital for America to hold our own against China and other rivals – but before we can use the most advanced microchips and run our state-of-the-art AI models, we need to update our outdated electrical grids.
- “The Guardian view on a global AI race: geopolitics, innovation and the rise of chaos” –The Guardian Editorial Board
- “To compete with China on AI, we need a lot more power” –Daniela Rus and Nico Enriquez, WaPo
- “Trump’s go-it-alone approach to AI will leave everyone a loser” –South China Morning Post Editorial Board
Sprinkles from the Right
- Some commentators argue that legislators on both sides of the aisle should recognize that America must lead the world in AI to preserve national security – giving the Trump admin the opportunity to establish a historic US advantage.
- Others contend that the US must not lose its lead over the rest of the world in digital innovation, making it crucial to commit resources to sovereign digital infrastructure – especially ​​power generation to support the demands of AI.
- “Trump Can Keep America’s AI Advantage” –Dario Amodei and Matt Pottinger, WSJ
- “America must secure its AI future with better energy infrastructure” –Lucian Niemeyer and Erik Bethel, Washington Times
- “On the brink of an unimaginable AI future” –Megan McArdle, WaPo
Share this!
Recent Discussion stories
Discussion
  | January 24, 2025The Trump admin is working to reshape the federal workforce
🏛️💼 In his first few days after taking office on Monday, President Trump has signed a series of executive orders aimed at reshaping the federal government’s 2.3 million-person workforce.
Discussion
  | January 22, 2025Breaking down the Trump admin’s new immigration agenda
✍️🇺🇸 President Trump signed a series of immigration-related executive orders hours after being sworn into office on Monday.
Discussion
  | January 17, 2025The Israel-Hamas ceasefire is poised to take effect
🤝 Israel and Hamas agreed to a preliminary ceasefire deal, in a step that could mark the beginning of the end for the 15+-month war in Gaza.
You've made it this far...
Let's make our relationship official, no 💍 or elaborate proposal required. Learn and stay entertained, for free.👇
All of our news is 100% free and you can unsubscribe anytime; the quiz takes ~10 seconds to complete