A. I. Sputnik moment: what just happened?

Key Markets update for Tuesday, 28 January 2025

China's DeepSeek AI released a new AI model, DeepSeek R1 causing what Marc Andreesen called, "a Sputnik moment" for the industry because just as Soviet Union's launch of the Sputnik satellite in 1957 upended the assumptions about the American technological dominance at the outset of the space race, DeepSeek is challenging such assumptions in the AI tech race today.

What is DeepSeek R1? It's a new language model designed to offer performance that seems to match or even exceed the performance of its much larger rivals. It is capable of answering questions, generating text and understanding context. But what sets DeepSeek apart is the way it was built: it is cheap, efficient and very resourceful.

As the retired Microsoft software engineer David Plummer reported in this short explainer video, DeepSeek matches or even exceeds the performance of best American AI models even though it was allegedly developed for under $6 million and without access to the most advanced Nvidia hardware.

DeepSeek: a distilled language model

By contrast, American AI companies invested tens of billions of dollars in their models. Plummer drives the point home with a Ferrari analogy: it's like building a Ferrari in your garage using spare Chevy parts. And if you can throw together a Ferrari in your garage and it's just as good as the regular Ferrari, what do you think that does to Ferrari prices? Plummer explained DeepSeek's architecture: it's a 'distilled' language model. When you train a large AI model, you end up with something massive; hundreds of billions, if not a trillion parameters using terabytes of data and requiring entire data centers worth of GPUs just to function.

For most tasks, however, you may not need all that brute force, so what the Chinese developers have done is, they used the distillation approach, where you use a larger model like GPT 4 or the 671 billion parameter behemoth R1 and you use it to train smaller ones. It's like a master craftsman teaching an apprentice: the latter doesn't need to know everything about everything but just enough to do the tasks at hand well. In this sense, DeepSeek used multiple AI models as master craftsmen in a process that's a bit like assembling a panel of experts to train one exceptionally bright student. In this way, DeepSeek developers compressed the knowledge and reasoning capabilities of much bigger systems into something far more lightweight in terms of resources.

As a result, DeepSeek doesn't need massive data-centers to operate. Users can run smaller variants on decent consumer-grade CPUs or even on a beefy laptop, and that's a game-changer, as it dramatically lowers the entry barrier for AI. This is great news for smaller companies, research labs or even hobbyists looking to experiment with AI without having to make a large investment in it. By focusing on cost and accessibility, DeepSeek R1 creates an important niche as a practical, cost-effective alternative to larger AI models.

Cracking entirely new possibilities open

To explain this effect, David Plummer draws a parallel to the early days of computing when the industry was dominated by mainframe computers. When the PCs were introduced, initially there was a lot that they couldn't do and were only good enough for some types of work. By today, it is clear that the lowly PC in fact revolutionized computing.

DeepSeek could catalyze similar changes in the AI industry by making advanced AI tools accessible to many kinds of smaller users who could tailor AI solutions for specific uses or industries, running on local hardware or even be embedded in devices like smartphones. In terms of great power competition, Deep Seek R1 signifies that China is not just a spectator in the global AI race but a formidable competitor capable of producing cutting edge open source models.

That creates a dual challenge for American AI leaders: maintaining technological leadership and justifying the price premium in the face of increasingly capable, cost-effective alternatives. Open source models like DeepSeek R1 will enable developers around the world to innovate at a lower cost. That could undermine the competitive advantage of proprietary models particularly in areas like research and small-to-medium size enterprise adoption. Companies and governments around the world will be able to build upon such foundations without fear of restrictions imposed by dominant firms.

US A.I. industry’s new financial trouble

This could accelerate AI adoption globally but reduce demand for U.S. models impacting revenue streams for companies like Open AI or Google Cloud. According to Sequoia research, the US AI industry needs to reach $600 billion in annual revenues to justify current levels of investment. So far, it's only reached around 10% of that sum and DeepSeek may just have blown up their business plans along with investors' assumptions about the future of the AI industry.

For those who are inclined to look further into DeepSeek, below are the links to two short videos I'd recommend:

• "DeepSeek Fully Tested - Isane Performance" (15 min)

• "NEW Deepseek AI Good For Creating Trading Strategies in TradingView and PineScript? (FREE AI)" (30 min)

To learn more about TrendCompass reports please check our main TrendCompass web page. We encourage you to also have a read through our TrendCompass User Manual page.

Trading signals for Key Markets, 28 Jan. 2025

With yesterday’s closing prices we have the following signals: