Can AI Save Us From Inflation (And Make US Rich) 2/2?

Is the AI Boom a Bubble, or Are We Underestimating the Future?

If you’ve been watching the stock market lately, especially anything related to tech or semiconductors, you know things have been absolutely wild. For months now, people have been tossing around the B-word: Bubble . Many investors who pulled cash out early, waiting for a "correction," or those who went all-in on inverse positions, are probably feeling pretty stressed right now as the market keeps soaring past historic highs . But here's the thing: what if those fears—the feeling that we've gone "too high, too fast"—are actually rooted in an old way of thinking, and we are missing the scale of change happening right now?

What’s fascinating is the debate over valuations. Back in the fall, when indices were much lower, the bubble argument was everywhere; yet, now that prices have climbed even higher, some reports are actually calling the market undervalued ! This counterintuitive insight suggests that many early bubble predictors likely failed to grasp the magnitude of the underlying shift in earnings and technology, particularly in the semiconductor space . I've found that often, our personal experience—the limits of the markets we've seen before—can blind us to truly unprecedented cycles. For someone who lived through the dot-com era, like AFW Partners CEO Lee Sun-yeop, the current situation feels different; back then, investors didn't have the real, tangible earnings growth we are seeing today driven by AI and data needs.

Let's look at the hard data, because numbers don't lie. When major players like SK Hynix and Samsung announced their recent earnings, the results were staggering, and they give us the clearest picture of what's happening. SK Hynix, for example, saw its revenue increase by 66% year-over-year in Q4, but its operating profit shot up an incredible 137% . Samsung’s profits saw an even more dramatic increase, up 200% while revenue rose 23% . This massive disparity—where profit growth outpaces revenue growth by such a significant margin—tells us exactly one thing: companies are heavily leveraging their pricing power (P) over volume (Q) . This is the surest sign of a severe supply shortage. The demand for semiconductors is so high right now that, just like during the Christmas season depicted in one widely circulated image, DRAM became a tragically scarce commodity in the U.S. market, with prices soaring five to tenfold in a short period.

Will the AI Demand Cycle Really Last Three Years?

If you were wondering how long this insatiable demand could possibly continue, we got a shocking hint straight from the top. When Tesla released its earnings, Elon Musk dropped a huge piece of information during the conference call that essentially reset the timeline for the entire industry: he projected that memory shortages would persist for at least three years, even if major suppliers like Samsung, SK Hynix, and Micron ramped up production under a "best-case scenario" . This statement is incredibly significant because it means that even the most conservative analysts' predictions for supply and demand might be dramatically off.

Musk’s insight, though, goes beyond just the three-year horizon. He suggests that the constraints in computing and memory packaging are so severe that Tesla will need to build its own dedicated computing facilities (Dojo or similar) just to remove these limiting factors, further emphasizing that the scale of AI computation required is far larger than previously estimated . What's crucial here is that when people actively building the world’s most advanced AI models talk, their forecasts about future demand are often much more reliable than those coming from traditional analysts who might be anchored to historical norms . The conclusion is simple: this isn't a quick, nine-month market cycle; we are in a long-term infrastructure investment phase that is barely getting started.

This leads us to a crucial realization shared by other tech giants: the world is entering an unprecedented technology investment cycle . Jensen Huang, CEO of Nvidia, echoed this sentiment, arguing that the talk of an AI "bubble" is misplaced because what we are seeing is the construction of a massive, historically unmatched infrastructure . He provides compelling data to back this up, noting that while hundreds of billions of dollars have been invested so far, the industry will require trillions more in future investment—a tenfold increase from current levels just to keep up with the trajectory of AI development . This means that if we apply the old metrics of P/E ratios or market caps to this investment cycle, we are likely misunderstanding the sheer potential scale of growth.

If Semiconductors Are Scarce, What's the True Bottleneck?

If major tech leaders are publicly admitting that they need 10 times more investment and that supply will be constrained for years, where is the ultimate pressure point? What’s the weakest link in this chain? Interestingly, the answer is not the chips themselves. Elon Musk identified the true constraint—the ultimate bottleneck that could slow down AI adoption—as power and energy supply.

You see, running these massive AI models, training them, and deploying them in data centers demands astronomical amounts of electricity, far surpassing what current global power grids can reliably provide . In fact, Nvidia itself warned in its earnings call that energy bottlenecks could slow down the rate of AI deployment . This counterintuitive insight—that the ability to power the chips is more limiting than the ability to manufacture them—shifts the focus away from just memory and GPUs toward power generation and distribution. Google's CEO also confirmed the exponential demands, stating that computing capacity will need to double every six months, and increase by a factor of 1,000 in just five years.

This enormous energy requirement points directly to major infrastructure plays. When the CEO of BlackRock, Larry Fink, engaged in a discussion with Musk, he specifically asked what the biggest hurdle would be for widespread AI benefits, and the answer was definitively energy . The race to secure this energy is evident. For instance, companies that can quickly construct reliable power sources, like those specializing in nuclear energy (like the South Korean firm D-Ena ), suddenly hold immense leverage. As we saw with the extreme delays in power plant construction in the US, Finland, and France (up to 14 years in some cases), the ability to deliver infrastructure on time is incredibly valuable, giving the quickest builders enormous pricing power . Ultimately, while both chips and power are vital, if the power goes out, the chips stop running, making electricity the most fundamental bottleneck for the entire AI future.