5 Stocks Surging on DeepSeek’s Shocking Week
Euphoria in Silicon Valley turned to panic this week after a Chinese-based startup called DeepSeek released a new large language model (LLM).
It’s very good.
And it’s very cheap.
Not only did DeepSeek’s “R1” model beat all but the top models from OpenAI and Google in quality rankings. But it did so by spending just $6 million in model training – an amount that Andrej Karpathy, the former head of AI at Tesla Inc. (TSLA), called “a joke of a budget.”
Cost savings extend to its day-to-day use as well. Initial third-party tests suggest R1 costs a tenth of what it costs OpenAI to run its “o1” model.
That triggered a massive rout on Wall Street. Companies from AI chipmaker Nvidia Corp. (NVDA) to AI data center power producer Constellation Energy Corp. (CEG) saw their share prices tumble 20% or more, wiping out more than a trillion dollars in market capitalization.
Because if DeepSeek’s claims hold true, AI might not require as much power – and perhaps not as many cutting-edge GPUs from Nvidia – to train and run.
For retail investors, these headlines might feel like whiplash. One moment, we’re being told we need massive “hyperscaler” data centers and high-end chips to power next-generation AI. The next, a barely known startup rattles the very foundation of the 2024 bull market.
Still, there’s more to this story.
Since the release of R1, our team has examined how the system gets such fantastic results. We wanted to find out what makes this system so special… and if that look under the hood fundamentally changes our outlook.
