Last week, financial markets trembled as Oracle's stock plummeted, wiping out tens of billions in market cap. But the interesting part wasn't the numbers—it was the reason. Oracle didn't lose revenue because nobody wanted their products. They couldn't deliver fast enough. Labor shortages and material constraints slowed their data center expansion, leaving the tech giant struggling to keep pace.
My personal experience with Google's API this week reinforced this reality. An 8-second video from Veo 3.1—something that used to take minutes—now requires 30 minutes of waiting. Even worse, daily image generation limits on premium models dropped dramatically from 100 to just 15-20 images.
We're facing a paradox: on one side, fear of a financial bubble about to burst; on the other, an unprecedented hunger for computing resources. Here's my perspective on what's really happening.
The Supply-Demand Paradox: A "Good Problem to Have"
Unlike the dot-com era of 2000—when companies like Pets.com were valued sky-high despite having no users or business model—AI faces the opposite challenge: supply can't keep up with demand.
Andrew Ng, a pioneer in the AI field, recently made a thought-provoking observation. He argues that we don't have too much technology—we have a severe shortage of inference infrastructure. Oracle, Google, and Microsoft struggling to expand isn't a sign of decline. It's proof of massive real-world demand. Google recently emphasized they need to double their supply every 6 months (1000x in 5 years) just to keep up!
According to Ng, this is "a good problem to have." It's better to be limited by production capacity than to build products nobody wants. However, he also warns that too much capital is flowing into "building the railroad tracks" (training and inference infrastructure) while ignoring the real gold mine: the Application Layer—where actual value is created for end users.
The Creative Destruction Bubble
Despite real demand, we can't deny the excessive enthusiasm of investment capital. Howard Marks from Oaktree Capital categorizes bubbles into two types, and this distinction is crucial right now:
Mean Reversion Bubble: Destroys wealth and leaves nothing behind (like tulip mania or worthless crypto tokens).
Inflection Bubble: Destroys investor wealth BUT leaves behind magnificent infrastructure for society.
The 19th-century railroad boom is a perfect example. Most early railroad investors went bankrupt due to fierce competition and massive capital costs. But when the bubble burst, the tracks didn't disappear. They remained, connecting territories and reducing transportation costs to a minimum for all of society.
AI is a textbook Inflection Bubble. Even if today's data center builders over-invest and suffer losses, millions of GPUs and massive data centers will still exist. They'll become cheap assets, enabling the next generation of applications to explode.
Who Gets the "Economic Surplus"?
Charlie Munger's thinking reveals a brutal truth: technology that changes the world doesn't necessarily make investors rich.
The aviation industry completely transformed human civilization, yet over 100 years, airline profits have been razor-thin—sometimes the entire industry operated at a net loss. Why? Brutal competition eroded every profit margin.
The latest "State of AI" report from OpenRouter shows the gap between closed and open-source models is shrinking rapidly: open models now capture about one-third of actual usage, with competitive or superior performance in many key areas. This creates strong pressure pushing AI service prices down toward pure energy costs.
When this happens, "economic surplus" won't end up in the pockets of model-building company shareholders. It will shift to end users—to society. We'll enjoy superintelligent AI at prices as cheap as electricity and water. That's a victory for humanity, even if it's a disaster for portfolios over-invested in hardware or foundation models.
When Giants "Burn Money," Society Benefits
The most impressive evidence I saw this week was the launch of GPT-5.2. This model increased economic efficiency by 390 times compared to last year. Specifically, the computing cost to solve the extremely difficult Arc-AGI-1 benchmark at a high score dropped from $4,500 to just $11.64.
Clearly, OpenAI and its investors are still burning money in the infrastructure race. But the result? A dramatically smarter model reaching ordinary users at a fraction of the cost.
This is the shift of "Economic Surplus" in action. When giants rush into an arms race, burning trillions of dollars to reduce costs and increase computing power, the ultimate beneficiary is society. If the financial bubble bursts, investors may lose money, but the AI infrastructure they leave behind will become cheap, ubiquitous "electricity" for the entire world.
What This Means for You
Instead of worrying about macroeconomics, focus on harnessing this stream of intelligent electricity to boost your productivity:
If you're a Doctor: Turn AI into your diagnostic imaging assistant and medical literature researcher. Your value will shift from "memorizing" to empathy and final decision-making.
If you're a Teacher: Use AI to personalize lesson plans and automate grading, freeing precious time for mentoring and guiding student thinking.
If you're an Engineer: As Andrew Ng suggests, don't just stop at writing code. Transition to designing "Agentic Workflows"—automated task processes that solve bigger problems.
The Bottom Line
As Warren Buffett once said: "In the short run, the market is a voting machine (driven by emotions), but in the long run, it's a weighing machine (measuring real value)."
The real value of AI isn't in the numbers on stock tickers. It's in the productivity it brings to your work today.
So, what about you?
How are you using AI to "upgrade" your career? I'd love to hear your stories and perspectives.
What are your thoughts on the AI infrastructure boom? Share in the comments below.