DeepSeek, a Chinese AI Startup, Shocks Silicon Valley With Low-Cost AI Model

The US tech giants believed their lead in AI was unassailable, at least in the short term. Today, they’re reassessing that assumption, which could lead to major upheaval in the burgeoning AI tech ecosystem. Chinese startup DeepSeek has exploded in popularity with a model that cost almost nothing to build and can outperform some of the most capable models to come out of US firms.
DeepSeek appeared on big tech’s radar quite recently when its iPhone app rocketed to the top of the App Store. The app and associated website allow access to the company’s latest DeepThink R1, a reasoning model similar to OpenAI’s GPT-o1 or Google’s Gemini Flash Thinking. It also offers a more traditional generative model called DeepSeek V3, but it was the power of R1 that led it to dethrone ChatGPT in the App Store.
By all accounts, DeepSeek’s models meet or exceed the standards of the most capable open source models like Meta Llama. They also challenge even the biggest closed models like OpenAI’s GPT-o1. At the same time, DeepSeek is working without the benefit of the latest and greatest hardware.
For the last several years, the US has been restricting the export of powerful AI accelerators and chip manufacturing equipment to China. The government hoped this would slow China’s efforts to use the technology in military applications. Nvidia complied with the restrictions, creating less powerful versions of its cards, like the H800 and later the H20. These hobbled accelerators were supposed to keep Chinese AI at bay, but it’s not working out that way.
According to Axios, DeepSeek managed to create R1 using Nvidia H800 cards for as little as $6 million. OpenAI, Google, and the rest of big tech are spending hundreds of millions of dollars on each new model. These models inflate quarter by quarter, adding billions of parameters, but maybe that’s not the only way to advance generative AI.
News of DeepSeek’s sudden success has sent ripples across Silicon Valley. Nvidia, which has seen its valuation soar in recent years, is in the midst of a stock drop. Since Friday, the company’s stock price has dropped about 20%, wiping out more than half a trillion dollars in value.
While DeepSeek is reportedly still running Nvidia hardware, its use of slower Chinese-specific accelerators suggests companies might not need to spend big on the company’s new Blackwell components. Each Blackwell Superchip is expected to cost upward of $70,000. It’s uncertain how US AI firms will react, but they’ll have to do something.
© 2001-2025 Ziff Davis, LLC., a Ziff Davis company. All Rights Reserved.
ExtremeTech is a federally registered trademark of Ziff Davis, LLC and may not be used by third parties without explicit permission. The display of third-party trademarks and trade names on this site does not necessarily indicate any affiliation or the endorsement of ExtremeTech. If you click an affiliate link and buy a product or service, we may be paid a fee by that merchant.

source