Deep Seek vs Nvidia in the Dip

China is good at Trolling

For the last two weeks, I have been discussing Money and how it has become the most common language around the world. After the first week of the Trump administration, we saw tariffs implemented, the announcement of major AI investments, and the involvement of tech tycoons such as Elon Musk and Sam Altman in this technology development. Nevertheless, during the weekend, China made its move, a move that erased hundreds of billions in market value across major US Tech Companies. I can’t stop asking myself if that money even exists or is just a metric of the damage that could be done.

To dig into what happened this weekend, let’s begin by understanding the main difference between a distilled model and a full model. Imagine you have a massive, all-you-can-eat AI buffet, that's your full model, but you want to create a keto vegan version that still tastes good. That's distillation! Taking a large, complex model and creating a smaller, more efficient version that maintains most of the original flavor. DeepSeek's genius lies in making this keto vegan version taste suspiciously close to the premium buffet.

Another important clarification for the sake of the conversation is to address why DeepSeek is not comparable with Meta's Llama if both are Open Source. But that's like comparing apples to, well, whatever fruit China claims apples have always been a part of. DeepSeek uses a mixture of expert architecture that activates only 37B, which is less than a tenth of Llamas architecture. Again Deep Seek is a very capable distilled model. This is something similar to Zero Knowledge proof when it comes to data liquidity, you make the request but you don’t show your parameters. And boy stock market took a hit. I am glad that Jensen publicly claims to enjoy a little bit of pain as a source of motivation. We witnessed Nvidia's stock doing its best impression of a skydiver without a parachute, plummeting 17% in a single day almost six hundred billion worth erased. The tech-heavy Nasdaq dropped 3.6%.

Why I don’t trust Deep Seek

Here's a fun game: Ask DeepSeek about Tiananmen Square, and watch as it performs the digital equivalent of "Look, a squirrel!". The AI starts writing the truth, then suddenly remembers its party membership and replaces it with "Sorry, that's beyond my current scope”. After seeing TikTok's capabilities, and the data concerns I wouldn’t feel comfortable turning into my personal assistant.

The Money Paradox

Here's the kicker: Wall Street just lost hundreds of billions in market value because a Chinese company claimed to build something for $6 million that American companies spent billions developing. This is why Mark Andressen called this moment the “Sputnik Moment”. The market reaction wasn't about technology; it was about fear. Fear that the AI gold rush might not require all that gold after all, for me is just another signaling game. In the end, we knew this was just the beginning of the game, and there were going to be many more cards to be played from both sides.

And isn't that just ironic? In our quest to create artificial intelligence, we've demonstrated just how naturally unintelligent our financial markets can be.

But hey, at least we can ask ChatGPT about Winnie the Pooh without it having an existential crisis.