원문 보기
Are you worried that China will catch up with us and leapfrog ahead? For digital intelligence, it all comes down to how many chips you have and how much energy. China has more electricity than the US, Europe and India combined. And at some point, they'll probably have a lot of computers. So, they will become leaders.
There is a high probability that they will become leaders someday. I don't know if you saw it recently, but Elon Musk gave a very poignant interview to The Economist. He was in despair. I would simply like to warn you and the media against the absurd characterization of the far right. Zero. People. I understand you perfectly.
Zero. Do you realize that the media is disrespectful? This is untrue, misleading, and nonsense. Please leave this part. Yes, he was ready for a tirade. That's right. But earlier in the interview, one of the most important topics they discussed was, of course, AI. Of course, it's AI. He is everywhere.
He drives my car. He writes my letters. It even calculates how many calories are in a glass of milk. I actually checked this. But more importantly, AI provides a very high percentage of stock market returns. And thanks to these seven companies, the stock market looks like this , with this valuation.
That's why I found Elon's thoughts in this interview interesting, because they exactly match what Steve Eisman said recently . A person who is quite adept at predicting bubbles in the stock market. So in this video, I want to talk about three elements of the problem that Elon discusses in the interview, and what we as investors should pay attention to.
The first problem is probably the biggest structural problem of this AI boom in America, which is very , very difficult to solve, and that is energy. But before we get to where the energy actually ends , listen to Elon describe this whole AI race, because honestly, the rest of this video is centered around that one sentence.
You could say that the limitations to AI are mostly electricity and AI chips. And really, it's just two resources: energy and chips. But the scariest thing is that in one of them, electricity, Elon says that the US is simply uncompetitive in the long term. And this is a problem. And if you look at electricity production, AI depends on which factor is limiting.
Whether it's chips or electricity, China has significantly more electricity than the United States. In fact, China already has more electricity than the US, Europe, and India combined. And I predict that China will achieve four times the electricity production of the United States, which is roughly proportional to the population.
This is where the real “bottlenecks” of AI begin to emerge in the long term. We know that China is already ahead of the US in terms of generation, but take a look at this Bloomberg article. They note that in one year, China added capacity equivalent to 40% of the entire US power grid, and in the last 4 years, it has built more generating capacity than the entire US power grid combined.
The energy gap is not narrowing. In fact, according to almost all forecasts, it is increasing. China is breaking further and further ahead. And this is a huge problem, because AI absorbs electricity in huge quantities. We need a lot of energy for the future of this technology, and this raises the obvious question: why can't the US just build more energy capacity?
To be fair , the US is certainly not ignoring its problems. They are trying to build energy infrastructure and recognize that energy-dependent data centers are becoming a problem. For example, in February, Trump announced a pledge to protect taxpayers, essentially telling tech companies, "Build your own power sources for your data centers." But even with this pressure, bringing new capacity onto the grid remains an unfortunate obstacle.
As of this year, projects with about 2,000 gigawatts of capacity are in the queue for connection in the United States . And the average waiting time for a real connection is about 5 years. In Texas alone, ERCOT manages a 474 gigawatt queue for large consumers, 90% of which are data centers. Some projects have been waiting for connection for up to 12 years.
So, the reality is that launching a new generation is a very long road. And that's why a lot of these AI hyperscalers are turning to things like this—backup gas turbines. I talked about this in a video earlier this year , but even Elon can't wait for the grid. He bought about 35 of these gas turbines for xAI's Colossus supercomputer in Memphis.
Typically, such turbines are used one or two at a time as backup generators, for example, for a hospital. But xAI had as many as 35 of them working as the main source of energy for the data center. So even Elon brought his own mobile power plant instead of waiting for a connection to the grid. This should tell you everything you need to know about how overloaded the system is.
For digital intelligence, it all comes down to how much AI computing power do you have? Well, how many chips do you have and how much energy do you have? How much electricity do you have? In my opinion, the limitation right now is a bit more about power and cooling than the AI chips themselves, as the power requirements for them are very, very high.
We'll talk about water needs, but in reality, the use of water for AI is negligible. The demand for electricity is very high. And it's not just about Elon. Everyone is trying to get these gas turbines now. So much so that it disrupted the supply chain. The waiting time for large turbines has stretched to 5 years.
GE Vernova's order backlog has grown to 116 gigawatts. Siemens Energy's order portfolio is around 69 gigawatts. Prices for turbines for power plants that will be launched in 2030 and 2031 are rising rapidly. They grew by 75%. That is, this is a real strategy of " selling shovels during a gold rush." In general, these are GE Vernova, Siemens Energy and Mitsubishi Heavy.
These are the three "big players" that sell these turbines, and you can see on Investing Pro that, despite the usual metrics, these companies are quite overvalued, but analysts still consider them fair or undervalued due to the prospect of AI development. Energy infrastructure is in high demand. When you ask Warren AI what's going on, you only see two factors.
The data center boom has begun, and so gas turbines, which are a transitional stage, provide dispatchable power while renewables and grids scale. And because of this trend, GE Vernova shares have risen 68% over the past year, Siemens Energy—by 65%, and to follow the energy infrastructure topic even more closely, Investing Pro also has ProPicks AI—a model built on over 50 financial signals that identifies market leaders.
In this tab we can see a strategy called Energy Elite. So, if you're looking for ideas for this particular topic, which is the energy infrastructure boom, you can take a look here and see how some of these companies compare to, say, market benchmarks like the S&P 500 Energy Index. So this is a really interesting feature.
I like to explore it for ideas. I want to mention that the great news right now is the summer sale on investing.com. This means you can get a 55% discount plus an additional 15% with my referral link in the description and pinned comment. So be sure to check them out. Now is a great time for this . And as always, thank you to investing.com for sponsoring these videos.
But let's get back to business. So this is a story about energy. This is Elon's first argument in the AI bubble debate. The US loses, and China wins. And that brings us to Elon's second argument— good old China suffering from a chip shortage. And since the US has banned the export of the latest AI chips to China, China is experiencing a shortage to some extent.
So, China has a chip shortage, but it could be a blessing in disguise for them. For those who don't know: because Nvidia makes the best AI chips in the world, and the US controls what Nvidia is allowed to sell to China, we have seen restrictions imposed on the export of the most powerful chips. This year the rules have been relaxed a bit.
In December, Trump approved a limited sale of the H200 to China. It started in small quantities in February, but there was one caveat. The US is imposing a 25 percent tariff on these chips heading to China. It's also worth noting that Nvidia's latest chips, such as Blackwell and the upcoming Rubin architecture, are still completely banned from China.
But here's why this is actually a problem for America. As the US imposed these restrictions on China, it encouraged China to develop its own technology. And now they are more competitive than ever. Although China's own deep ultraviolet lithography machines, which print microscopic circuits on silicon wafers, still lag behind ASML—the gold standard for performance and reliability.
But SMIC, which is the Chinese counterpart to TSMC, has found ways to get around some of the restrictions. They have already demonstrated 7- nanometer production using older DUV technology, and are now starting to master 5- nanometer. Massive efforts are underway across China to bridge this technological gap.
So while export restrictions have undoubtedly hurt China in the short term, they have given the country a huge incentive to create its own semiconductor ecosystem. China may be experiencing a chip shortage today, but the situation may change in a few years. And this brings us to Elon's third and final argument for why the US AI narrative may be a bit overblown: while China is currently experiencing a chip shortage and may remain so for years to come, it is producing AI models that are already very high-quality and cheap.
What do you think about Chinese models and the geopolitical aspect of this issue? Are you worried that China is catching up and breaking ahead? Given that Chinese AI companies are achieving such success with relatively little computing power, it seems that if they had a lot of resources, they could well become leaders.
And at some point, they will probably have a lot of computing power. So, they—ergo— will become leaders. There is a high probability that sooner or later they will become leaders. Listen to this. Elon is confident that it is only a matter of time before China becomes a world leader in AI. The reason is that despite the shortage of chips in China's AI ecosystem, they still create very, very good models.
I think, realistically speaking, Fable is still the smartest model, and anyone objective will confirm that. But the Chinese model is getting pretty close to Fable. So even Elon, who has every reason to promote his own interests, says that the gap between the best US model and the best Chinese model is narrowing.
And this is where what I talked about in my video about Steve Eisman comes back, because Steve is not as concerned about the technical capabilities of Chinese models as Ilona is. He is concerned about what will happen to LLM pricing in the US when they become competitive . AKA, he predicts a price war that will squeeze OpenAI and Anthropic.
And now we have numbers to back up this argument. This is data from research firm Artificial Analysis. According to their tests, DeepSeek V4 Flash costs 3 cents per test compared to 86 cents for Moonshot AI Kimi K3, $1.86 for OpenAI ChatGPT 5.6 Soul, and $3.15 for Claude Fable 5. Steve Eisman estimates that OpenAI and Anthropic together generate about 70% of all AI revenue that goes to large hyperscalers: Microsoft , Amazon, Google.
So, take some time to think about it. If a cheaper, “ good enough” Chinese model starts luring customers with price, it will have a very bad impact on OpenAI and Anthropic’s margins. And if that means OpenAI and Anthropic have to cut hundreds of billions in future spending they’ve already promised hyperscalers— spending that’s already factored into their stock prices—it could be a catalyst for an AI market correction.
So, let's take a step back. Let's put the facts together. There was a lot of stuff in this video: energy, chips, prices, models, etc. But, in essence, Elon is talking about three elements of one system. The US now has chips, but is running out of energy . China has energy, but it lacks chips. This is the situation now.
But the problem for America lies in the future. The microchip shortage in China is a problem that could take years to solve. You know: lithography, end- of-life, manufacturing. However, America's electricity shortage is a problem that can be solved in a decade, because you can't just build a gas turbine, a transmission line, or a new power plant in a short period of time.
So you get a US energy bottleneck, plus China solving the chip problem, plus models that cost significantly less—that's a recipe for big losses, especially for OpenAI and Anthropic. And if they can't spend the hundreds of billions they promised to hyperscalers, the latter will suffer , and the entire narrative about artificial intelligence will begin to collapse.
Here is the story. And it's interesting that this is why Elon is trying to build his data centers in space. Solar energy is much more efficient in space. In addition, Elon has ready-made launch capabilities through SpaceX. So it may not be as incredible as it sounds. But with that in mind , these are Elon Musk's thoughts on the AI "bubble" we find ourselves in.
I hope this was all clear. I hope you enjoyed it. Many thanks again to investing.com for the sponsorship. Don't forget to follow my link in the description to get an additional 15% discount. Take advantage of their 25% off summer sale, but it ends in just a few days. I also wanted to mention that spots for the Rule One New Money live seminar, which will be held in Atlanta in September, are already filling up.
It will be held from the 18th to the 20th. Come to a complete 3-day training course on Warren Buffett's strategy, where you will learn how to find great businesses, evaluate companies, and how to use options strategies to get the most out of every investment. I think you will all really enjoy learning this.
I'll be there, Phil will be there of course, the whole Rule One coaching team will be there too. If this is something you are interested in, please join us. The ticket is free, so get yours today. Other than that, friends, thanks for watching, and see you in the next episode.
Comments 0