| Silicon Valley is awash in cash for artificial intelligence. So when DeepSeek's open-source model called R1 surprised the world, Jackie Davalos writes, it revealed that money doesn't automatically secure dominance. If this email was forwarded to you, click here to sign up. While tech executives were rubbing shoulders in Davos, Switzerland, last week, discussing the billions of dollars it would take to stay ahead of China in the artificial intelligence race, a relatively obscure startup released a model that proves throwing money at AI won't guarantee a US lead forever. The biggest AI companies have burned through billions to develop large language models. Chief Executive Officer Mark Zuckerberg said Friday that Meta Platforms Inc. would spend up to $65 billion this year. OpenAI, SoftBank, Oracle and others pledged $100 billion last week as part of the Stargate AI venture. And Microsoft Corp. CEO Satya Nadella said he's "good" for his $80 billion to build out its Azure platform. Then there's DeepSeek, a Chinese AI startup founded just last year that claims it was able to train a model comparable to those from OpenAI and Meta for a cool $5.6 million. The low price tag was a big surprise for the investors who've been bankrolling the AI boom. They'd assumed the endless sums being spent were just the cost of doing business. Tech stocks tanked Monday, and AI chipmaker Nvidia Corp. had the biggest wipeout in US stock market history, with its market value falling $589 billion. Despite the reaction, DeepSeek's capabilities aren't that deep. The startup's R1 model can do a lot of things we've already seen from leading chatbots including those from OpenAI, Meta and Anthropic. DeepSeek's AI assistant can answer questions in a conversational way, handle reasoning and do some tasks, particularly ones involving math or coding, with an accuracy that's on par (or better in some cases) than AI incumbents. The shock caused by DeepSeek has less to do with what it can do and more with how it does it. Its models use more efficient techniques and aren't as dependent on the most powerful (and costly) Nvidia graphics processing units that the US government has worked so hard to keep outside of China's borders.  DeepSeek's cheaper model shook Wall Street on Monday. Photographer: Mladen Antonov/Getty Images This breakthrough—clever engineering trumping brute-force computing power—is in some ways a reckoning of the US's own doing. In its quest to undercut China's ability to build advanced AI by choking access to the cutting-edge chips, Washington helped speed up the Chinese government's building of its own AI ecosystem. President Donald Trump has vowed to "make America the world capital in artificial intelligence," but DeepSeek's debut reveals that cordoning off our homegrown technology comes with trade-offs and could breed a sense of complacency among US companies coasting on sky-high valuations. Trump called DeepSeek's progress a "good" thing and said the release of the AI assistant from a Chinese startup "should be a wake-up call for our industries that we need to be laser-focused on competing to win." OpenAI CEO Sam Altman seemingly got the message. He said in a post on X that "we will obviously deliver much better models and also it's legit invigorating to have a new competitor! we will pull up some releases." He added that "compute is more important now than ever before to succeed at our mission." If Altman is right, the US still holds the upper hand. Silicon Valley's AI companies have proved they can outspend China—it's whether their investors will want to keep footing the bill. | 
No comments:
Post a Comment