Summary of Key Points
This article challenges the common belief that an "AI bubble is a bad thing," arguing that it is an inevitable stage in a technological revolution. Essentially, an AI bubble represents a "ritual of burning" for excess capital, using reckless investment to lay the groundwork for the next generation of infrastructure. A collapse is not a disaster; rather, it is a "forceful tool" for clearing out old rules and creating new opportunities. The key is not to avoid a collapse but to prepare in advance for the world that follows: for example, by establishing a consensus on the meaning of human existence in the AI era, distributing the benefits of AI from the source (rather than redistributing them later on), and preventing chaos or monopolization by a few individuals after the bubble bursts.
1. Bubbles Are Not "Stupid"; They Are the "Fuel" for Technological Revolution
You might think it's crazy to build 30 data centers that consume大量 electricity in a cornfield, but this is a consequence of excess capital with nowhere to go. The French thinker Bataille noted that human society has never lacked resources for survival; what we lack are ways to consume those surplus resources. Ancient Egypt built pyramids, medieval Europe constructed cathedrals, and the Aztecs conducted rituals—all of which were forms of burning unused energy. Today, excess capital is burned through bubbles. Rational investment aims for a quick return on investment, but fanatical capital (that believes "this time will be different") invests billions in technologies with unclear business models for decades.
History shows that the 1840s British railway bubble led to the bankruptcy of many companies, yet the railway network remained and contributed to the Victorian era. The 2000s internet bubble caused numerous company failures, but the submarine fiber optic infrastructure survived, enabling platforms like YouTube and Netflix to thrive. Bubbles burn money, but what remains is the infrastructure—essential for technological progress.
2. To What Extent Has the Current AI Bubble Reached?
The current AI bubble is in the "final stages of frenzy":
- Real demand is increasing: In the past 12 months, global revenue from generative AI reached $110 billion (real money from customers), growing three times faster than mobile internet revenue. Moreover, for every 10% decrease in AI token prices, usage increases by 12-18% (similar to how more coal was used as steam engine efficiency improved).
- Investment has gotten out of control: Microsoft, Google, and Amazon plan to spend over $600 billion on AI infrastructure by 2026, with three-quarters of that going towards AI. Their own profits are not enough, so they need to raise an additional $1.5 trillion (competing with U.S. government bonds for funds).
A bubble is not "empty"; there is real demand and infrastructure in place, but prices have outpaced value—this is what defines a bubble.
3. Collapses Are "Cleaners": They Remove the Waste and Leave Behind the Essential
A collapse does three things that prosperity cannot:
1. Infrastructure is completed ahead of schedule: During a collapse, data centers, power grids, and people's skills in using AI are established (the 30 data centers in the cornfield will not disappear).
2. Capital returns to reality: During prosperity, investors ask if a story can fund the next round; after a collapse, they ask if a product can make money. This process filters out "unicorns" that rely on hype and retains companies that depend on cash flows.
3. It forces rule reconstruction: In prosperity, no one discusses issues like who should own AI copyright or computing power (because those making money don't want to answer these questions). But after a collapse, old rules are invalidated, and everyone is forced to rethink—this is the only chance to change the game.
4. A "Soft Landing"? Could Be a "Chronic Poisoning"
Everyone hopes for a "soft landing" (a gradual deflation of the bubble), but this might be worse:
- Fortification of profit-sharing alliances: Prolonged stability can lead to groups that hold resources without producing value (e.g., monopolistic industry associations), allowing them to further consolidate their power.
- A golden age, not a golden era: Growth may recover after a soft landing, but the distribution of wealth remains unchanged (e.g., the post-2008 recovery widened the wealth gap).
Only a collapse, with its disruptive impact, can break these old structures and make way for new rules.
5. What to Prepare For in Advance?
The key is to build the "fourth pillar" of stability: not to write a perfect blueprint for how humans should live in the AI era, but to create an environment that allows new solutions to emerge:
- What is this environment? It includes books, schools, open-source models, and public debates (where different perspectives can compete fairly without being suppressed). The New Deal in the U.S. did not arise suddenly; it was the result of decades of progressive and labor movements.
- Distribute benefits from the source: Don't wait for taxation and redistribution later on. Distribute access to AI computing power, models, and data before they are monopolized. For example, make open-source models accessible to everyone, provide schools with public computing resources, and ensure that everyone has a stake in the AI era (just as Paine argued in the 18th century that land is a shared human heritage).
A collapse is not scary; what's scary is to be unprepared for it. Start preparing now by creating an environment that fosters new rules.
In conclusion: Let the bubble burst, but don't be caught with nothing ready when it does.
(End of article)