What DeepSeek teaches us about Moats!
Today, we discuss what DeepSeek means for investing in Highly-Innovative Industries and the Moats of Nvidia, Google, Meta and co.
You may have heard of DeepSeek - a Chinese AI model that has been talked about all over the industry in recent days.
DeepSeek unveiled a free, open-source AI assistant that uses lower-cost chips and less data than its American counterparts. For comparison, DeepSeek spent $5 million, yes million, not billion, on training their AI model.
OpenAI, the company behind ChatGPT, recently raised $6.6 billion and borrowed an additional $4 billion to develop its model. $10 billion vs. $5 million…
Google, Microsoft, Amazon, and Meta spent a combined $300 billion on AI research in 2024—a number that is expected to grow significantly in 2025 and the years ahead.
If you look at the performance results between DeepSeek and OpenAI models, it’s not surprising that the huge investments of Big Tech are now being questioned:
DeepSeek performed on par or better in every category of the test. While I was initially a bit cautious about the numbers DeepSeek reported, dozens of industry experts, even competitors, supported these numbers.
There’s still a question about the costs. Maybe DeepSeek understated the actual costs to make it seem even more impressive. However, it’s widely accepted by now that it operates at only 3% to 5% of the costs of OpenAI’s model, and the difference is so huge that a couple of million more or less does not make a difference.
Why is DeepSeek so much cheaper?
In short, because the US sanctioned chip exports to China. Precisely, Nvidia’s H100 chips were not allowed to be sold to China. Thus, DeepSeek was trained on the weaker H800 chips. Apparently, the difference is not as big as experts in the industry thought.
Consequently, it is no surprise that Nvidia stock is down significantly on this news (-12% pre-market).
It does not look much better for the overall tech industry:
Within just six months, China’s AI companies caught up to their American counterparts, even without access to the same chips. Initially, US experts thought America was years ahead. Former Google CEO Eric Schmidt said that:
In the case of artificial intelligence, we are well ahead, two or three years probably of China, which in my world is an eternity.
Recently, he said that China has caught up much faster than expected and that there is an open race between the US and China.
So, back to the original question. What does this teach us about Moats in highly innovative industries?
1. Leaders get Replaced quickly
The more innovative an industry is, the more likely it is that the current market leader will be replaced sooner rather than later. Just a couple of weeks ago, it looked like Nvidia chips would be the standard for decades to come.
Now, it’s questionable if you really need Nvidia chips to train AI models when even the old Nvidia chips can do it. Certainly, other companies can make chips with the same quality as “outdated” Nvidia chips.
→ We don’t know how this will play out. Fact is, you pay 36x forward earnings for a company whose growth outlook became much more uncertain within just 5 days, and no one, not even the experts, saw it coming.
I’m not saying Nvidia will be a bad investment going forward, but who can really say Nvidia or AI companies are in his circle of competence when not a single expert saw the developments of just the last 8 weeks coming…
2. High Risk Investments
The American tech giants invest hundreds of billions in AI technology. If other companies can do it without spending a single billion, we have to question the return of these investments…
→ Everyone wants to be the first mover to introduce new innovations. This leads to drastic overspending and burning their (and their investor’s) money.
3. How much Money will be generated in the End?
The more high-quality AI models there are, the more they become a commodity. Pricing power diminishes drastically. When ChatGPT was without an alternative, OpenAI could set the prices. Now, many models are at about the same level, and there will only be more from here on out.
How profitable will the business be when it’s out of its growth state?
→ Many industries have changed our lives (think of airlines), but the companies operating in those industries have been bad investments. Who knows how it’ll be for AI?
Bottom Line:
I’m not saying there is no money to make investing in AI. I’m pointing out, though, that there’s always a price too high (shoutouts to Howard Marks). DeepSeek is a beautiful example of how fast alleged moats can be disrupted.
How will all of this turn out? I don’t know yet. I’m happy, though, for the volatility in the markets (hopefully, Flow Traders can do some good business 😉)
I also want to do some more research regarding this development and the immense price reductions Alibaba and its competitors have recently made for their cloud and AI services. Maybe there have been cost advantages already that led to them initiating these price reductions.
Well, that’s it for today’s short article; let’s call it a market update. I hope you enjoyed it!
On Friday, a new Deep Dive is covering my latest Portfolio addition. Paid Subscribers know what company I’m talking about.
Till then, have a great time and think about the Risk/Reward ratios of your investments 😉.
Best
Daniel
All information about the Big Update coming in March: