Chinese AI lab DeepSeek has sent shockwaves through Silicon Valley with the release of open versions of AI models that rival those of tech giants OpenAI, Meta, and Google. The development has not only rattled the tech industry but also raised concerns at the highest levels of the US government, which fears that China is pulling ahead in the AI arms race.

DeepSeek claims to have built its models highly efficiently and quickly, offering them at a fraction of the price charged by American AI companies. The move has sparked a frenzy of activity, with AI CEOs, founders, researchers, and investors telling TechCrunch that DeepSeek's models have major implications for American AI policy.

The rise of DeepSeek marks an inflection point for Silicon Valley's AI landscape. Experts say the models serve as an indicator of the accelerating rate of AI progress, with some even calling it an "AI Sputnik moment" – a reference to the Soviet Union's launch of the Sputnik spacecraft, which spurred the US to invest heavily in its space program.

One of DeepSeek's key innovations is its use of "pure reinforcement learning," a trial-and-error approach that allows the model to learn through experience alone. This approach has been hailed as a breakthrough, with experts saying it represents one of the most promising ways to improve AI foundation models today.

DeepSeek's models, including its R1 model, have been shown to match or beat OpenAI's o1 model on several AI benchmarks. The speed with which DeepSeek achieved this has taken the industry aback, prompting calls for increased investment in open-source AI to compete with China.

Martin Casado, a general partner at Andreessen Horowitz (a16z), tells TechCrunch that DeepSeek proves the US is not alone in its technical capability and that very competitive solutions can come from anywhere, including China. Casado argues that instead of hampering US innovation, the country should invest strongly in open-source AI.

President Trump has weighed in on the issue, saying that DeepSeek should be a "wakeup call" for American AI companies and praising the Chinese AI lab for its open approach. This lines up closely with a16z's stance on AI, with the VC firm having argued that the US should prioritize innovation over preventing "outlandish" AI doomsday scenarios.

Not everyone is convinced, however. Some analysts are skeptical of DeepSeek's claims, including its assertion that it trained one of its frontier models for just $5.6 million using older Nvidia GPUs. Additionally, tests have shown that DeepSeek's models are flawed, providing inaccurate answers or non-answers 83% of the time when asked about news-related topics.

There are also claims of IP theft, with OpenAI alleging that DeepSeek used its AI models to train its own – a violation of OpenAI's terms. OpenAI is currently being sued by several parties for allegedly committing copyright infringement in training its models.

Despite these concerns, DeepSeek's innovations have moved the needle, and its models have the potential to change the way AI is developed and used. As the US and China continue to compete in the AI arms race, one thing is clear: the stakes have never been higher.