The meteoric rise of Chinese startup DeepSeek has rattled Silicon Valley to its core.
The company demonstrated that it's possible to create top-tier AI at a tiny fraction of the cost of what outfits like OpenAI have been spending. By employing some clever tricks, DeepSeek says it squeezed a groundbreaking amount of performance out of a limited number of already obsolete Nvidia chips.
The news sent a jolt down Wall Street, leading to a massive tech selloff that wiped out over $1 trillion in market capitalization — not to mention almost $100 billion in losses for tech billionaires.
DeepSeek's seemingly competent use of "distillation," which is essentially training an AI on the output of another, has caught the attention of the AI industry, as the Wall Street Journal reports, sending a chill down the spines of tech leaders.
The process effectively allows much smaller entities to rapidly train AI models at a tiny fraction of the cost, an awkward development given the ungodly sums that the likes of OpenAI have been pouring into their models.
"It’s sort of like if you got a couple of hours to interview Einstein and you walk out being almost as knowledgeable as him in physics," data management company Databricks CEO Ali Ghodsi told the WSJ.
Before the emergence of DeepSeek, the modus operandi of AI companies had been to maximize the capabilities of AI models by maximizing the computing power behind them, a mind-bogglingly expensive endeavor that culminated in Trump announcing a half-trillion dollar AI infrastructure deal earlier this month.
But DeepSeek has demonstrated that smaller and more efficient rivals appear to be able to achieve similar — if not better — results.
OpenAI has since attempted to defend itself, suggesting to the Financial Times that DeepSeek may have broken its terms of service through distillation. But as experts quickly pointed out, that's a glaringly hypocritical position to take for the Sam Altman-led company, considering its own success was built on indiscriminately stealing other people's work as well.
Put simply, there's a chance that OpenAI may have completely lost its edge, leading to plenty of soul-searching among investors.
"Is it economically fruitful to be on the cutting edge if it costs eight times as much as the fast follower?" Hanabi Capital general partner Mike Volpi asked the WSJ.
Meanwhile, Anthropic CEO Dario Amodei attempted to underplay DeepSeek's significance. In a blog post, he argued that the startup's AI model "is not a unique breakthrough or something that fundamentally changes the economics" and that it's instead an "expected point on an ongoing cost reduction curve."