DeepSeek R1: A Plot Twist for Silicon Valley

Shockwaves from the Deep (Seek) - there's more to the story!

Author

Sebastian Muehl
January 30, 2025

Good morning. NVIDIA lost a cool $465 billion market valuation on Monday this week. A monumental AI model launch shook up the entire US tech industry. Deep Seek is entering the tech stage not just to compete but to reshape what is currently happening.

The conventional thesis was that developing AI models that could compete with the leaders would require loads of expensive hardware — and the comfortable feeling for the US that Chinese companies could not get their hands on this hardware.

However, a Chinese AI startup achieved just that last week using loopholes and young, inexperienced engineers—a wake-up call to Silicon Valley.

Aside from that, it is worth taking a look at their launch and market positioning strategy. A bold move that puts them in control.

Shockwaves from the deep seek

💡Trend Alert: 📉The Big Drop

For a while now, the markets have believed that artificial intelligence is the new poster child for a golden era in the US and steeply rising tech stocks, all led by NVIDIA. Suddenly, the mood turned.

Necessity Is the Mother of Invention.

The Chinese AI startup DeepSeek released a model that performs on par with leading AI models from OpenAI and others. The difference is that it was trained with much less computing power due to restricted access to the needed chips.

Why It Matters:

  • The narrative just changed: AI development doesn’t need the hardware monopoly we thought. Could this mean the market’s enthusiasm for AI has been overinflated, and less investment is required in order to build cutting-edge AI?

  • Less capital, same results: If DeepSeek’s approach becomes a trend, it will reshape how investors think about capital allocation in the AI space. I think more capital will flow into solutions rather than the tools. What matters more now will be network effects, deep integrations and hard to get data from businesses.

  • A wake-up call from China: Chinese startups no longer follow US tech leaders - they can compete toe-to-toe. They work hard, and they know how to focus on what matters — it sets a powerful example. I am looking at you, Europe!

❓For the Curious: How they did it

DeepSeek’s constraint on resources made them inventive. They were able to slash training costs by rethinking how AI processes information. Instead of pre-reading everything, like existing AI models, DeepSeek finds the right information when needed.

It’s like DeepSeek didn’t read every book in the library (like other models) but it knows where to look if the question come up.

It also uses a mixture of experts system. Topic specific questions are getting delegated to sub-models. Instead of one generalist AI model.

The trade-off? Less power needed upfront but more computing power and time needed to generate the answers.

The one other thing they did was to use special chips that NVIDIA specifically made for the Chinese market. Like a loophole around the export regulations. AND there are rumors that DeepSeek isn’t all the honest about the exact computing power they used.

How DeepSeek is defining the market on its own terms

I thought there was another noteworthy story in this whole chaos. I loved their launch strategy. They didn’t just launch to compete; DeepSeek defined the AI market on its own terms.

You see, launching as just another product vendor would have been the conventional launch method—one more model to pick from.

What they did instead was they positioned themselves as an ecosystem builder with an open-source strategy.

They are taking a page from the playbooks of RedHat (Linux) and Android. There are many benefits to that, like higher trust, faster adoption, and the potential for setting industry standards.

So, with the launch, DeepSeek invited the entire AI community to build and innovate together.

Massive business opportunities arise.

DeepSeek is not the first open-source model. Meta’s Llama is also considered open-source. However, Llama has restrictions on commercial use, while DeepSeek is completely open-source.

This enables anyone to start, for example, a B2B business. Selling the on-premise setup and training of an in-house AI model. These can be tailored to specific company use cases, and the company data stays within the company's gates. It can run wholly protected. Exactly what businesses are looking for these days.

How will DeepSeek make money?

This is where the story goes completely wild. I would have thought that classic open-source business models have led to riches by providing support services, deployment, etc., and that this would be their genius plan as well…

However, based on this tweet, they might not even be looking for any of that. It’s just a side project??? They didn’t even plan to make some beer money with this?

Conclusion

No matter what it is, it is a huge win for open source and a great gift to the world. By giving everyone access to open-source, state-of-the-art models, we can all win, not lose.

DeepSeek's playing offense with a community-driven model is excellent news for AI and its advancements. Everything else is a heated, further-developing debate that I can’t help but follow for more juicy details.

So what do you think? Do you think they have more GPUs than they admit? Or was it really a side project that cost a bunch of “smart math guys” $5 million to run? Let me know in a reply or comment.

Have a great rest of the week,

How did you like this edition?

Login or Subscribe to participate in polls.

Reply

or to participate.