The emergence of Chinese AI startup DeeSeek upended Silicon Valley earlier this week, punching a massive $1 trillion hole through the tech industry. The company's highly efficient yet high-performing "reasoning' model, dubbed R1, suggested that Wall Street may be massively overspending on computing power to run its AI systems.
Now it's coming out that DeepSeek rattled employees at ChatGPT maker OpenAI so deeply that it's causing a major rift among the company's staff.
As Wired reports, insiders are concerned that OpenAI could soon fall behind DeepSeek, at least in part due to a power struggle between the company's research and product groups.
While OpenAI's latest o1 "reasoning" model — created by the research group — gets the most public attention, "leadership doesn’t care about chat," one former employee who worked on chat told Wired.
Another former OpenAI researcher added that DeepSeek did "similar" reinforcement learning for its own R1 reasoning model, "but they did it with better data and cleaner stack."
Put simply, OpenAI's o1 model remained experimental, allowing DeepSeek to come out ahead.
"It was like, 'Why are we doing this in the experimental codebase, shouldn’t we do this in the main product research codebase?" one employee in the research told Wired. "There was major pushback to that internally."
The AI industry's turmoil this week raised plenty of glaring questions: are investors really overpaying companies like OpenAI, which have been hellbent on scaling up their AI models' abilities by spending vast sums of money on power-hungry datacenters?
According to the Wall Street Journal, the Sam Altman-led company is now looking to raise yet another $40 billion, a round of funding that could balloon its already colossal valuation to a lofty $340 billion.
That's in part an effort to shore up enough cash for Trump's shiny Stargate AI infrastructure deal, which is aiming to raise a whopping $500 billion in just four years.
It's an astronomical amount of money to be pouring into unproven tech, and investors were clearly spooked by DeepSeek's groundbreaking advancements.
Whether OpenAI can regain the crown and calm investors remains to be seen. The company opened up its next-generation "reasoning" model dubbed o3-mini to all users for free today, a lighter-weight and therefore cheaper-to-run model.
It's almost certainly intended as a response to DeepSeek, whose open-source models are also available for free.
(OpenAI also has the advantage, by the way, of not being beholden to Chinese censorship or being a perceived risk to national security by the White House.)
Now that DeepSeek has thrown the gauntlet, investors are likely to push for greater efficiency. But that doesn't necessarily mean OpenAI is about to downscale its operations any time soon.
"You do need less compute per unit of intelligence, but people are still going to want more units to scale up even more," independent AI policy researcher and former OpenAI employee Miles Brundage told Wired.
More on OpenAI: OpenAI Strikes Deal With US Government to Use Its AI for Nuclear Weapon Security
Share This Article