DeepSeek Derails the Big Tech Trade

It’s incredible to see how quickly a narrative can change. Last week, the US seemed to establish itself as the leader (and winner) in the AI race when Donald Trump announced a $500 billion AI Infrastructure investment through The Stargate project with OpenAI, Oracle, and Softbank while Meta committed another $65 billion to its own AI advancements.

Investors saw the news as inherently bullish for the US market and big tech companies in particular. The Nasdaq gained 1.65% while Meta and Alphabet both reached new ATHs.

I wrote about it as recently as yesterday. A lot can change overnight though. Here’s the brief, overly simplified, non-tech version of what just happened.

‍

Enter DeepSeek

So what changed? In short, DeepSeek entered the stage.

DeepSeek is a Chinese AI company. Its core product is an AI chat app similar to ChatGPT. In late December, it unveiled a new large-language model called DeepSeek-R1 that it says took just 2 months to and less than $6 million to train. It’s allegedly running on just 2,048 Nvidia H800 GPUs (at a cost around $40 - 50 million).

Last week, on January 20, it released this model in a user-friendly app that’s since skyrocketed to the top of the App Store.

The new DeepSeek model is performing on par with ChatGPT and other established players in the field. Its quality is impressive enough in and of itself.

What’s really remarkable, however, is that most people hadn’t even heard of DeepSeek until a few days ago. Most investors and analysts didn’t expect Chinese tech companies to be at this level yet. In fact, the US has restricted China from buying the most advanced chips, which means that DeepSeek is running on “outdated” technology.

Furthermore, the company appears to have achieved at a fraction of the cost what US companies have spent billions of dollars to do.

‍

The implications

The success of DeepSeek puts into question the massive AI spend that’s been happening in the US over the past couple of years. And the additional spend that’s planned for 2025.

If a Chinese startup can achieve this at such a low cost, why have big US tech companies spend so much money on AI? And why should they continue to do so?

The market implications are pretty clear:

• Nvidia won’t sell nearly as many chips as expected if DeepSeek is built on fewer and older chips than what the big US tech companies have been investing in recently.
• US tech giants like Meta, Microsoft, Alphabet, and Apple may not deserve their current multiples if they’re actually behind China in the AI race.

As of this writing, about half an hour before the US market opens, the tech sector is a bit of a bloodbath. Nvidia is down 11% premarket, dragging the Nasdaq 100 down by 3.5%. Both Alphabet and Amazon are down more than 3%.

The crypto market seems to have been collateral damage as Bitcoin fell around 7% over night before recovering some of the losses. The new US administration’s bundling of AI and crypto may have contributed to the fall.

To be fair, investors may also just be derisking ahead of the FOMC meeting on Wednesday after a really strong period. The Bank of Japan’s rate hike on Friday may also cause some concerns about the sustainability of the yen carry trade.

Regardless, the DeepSeek situation is definitely causing some concerns. It will be interesting to see how the markets react after digesting the news.

‍

Can we trust the DeepSeek story?

Some investors are skeptical of DeepSeek’s claims. They fairly point to the fact that we can’t always trust news and data coming out of China.

One point of contention is whether DeepSeek has actually managed to train its model on just $6 million when US counterparts are spending billions.

Related to that is the hardware being used and if China has perhaps managed to acquire some newer chips from the US despite the restrictions.

Achival’s post on X sums up what many people are thinking.

This story clearly isn’t over yet and will be worth following closely over the coming days and weeks.