The Comfortable Bubble: Silicon Valley's New Favorite Idea
"Bubbles are good."
That is not a headline from The Onion. It is a direct quote from Hemant Taneja, the CEO of General Catalyst, one of the world's most prominent venture capital firms. In a recent interview with The Atlantic, Taneja acknowledged that AI is a bubble but welcomed the idea. Bubbles lead to "some spectacular failures," he said, but also to "enduring companies that change the world forever."
He is not alone. Jeff Bezos has argued that AI might be a "good" kind of bubble. Sam Altman has said similar things. Even Howard Marks, the legendary investor who famously anticipated the dot-com crash, told The Atlantic: "If investors remained dispassionate, it would take a lot longer for a new unproven technology to be adopted."
The remarkable thing is not that Silicon Valley is in a bubble. The remarkable thing is that Silicon Valley has stopped denying it and started insisting that there is nothing wrong with it. The question is no longer "Is this a bubble?" but rather "Is this a good bubble?" and the answer, increasingly, is yes.
Why the railroad is the new dot-com
The intellectual scaffolding for this position comes from Boom: Bubbles and the End of Stagnation, a 2024 book by Byrne Hobart and Tobias Huber published by Stripe Press. The book has been praised by Peter Thiel and Marc Andreessen and has quietly become something like the unofficial doctrine of the current AI build-out.
The core argument is that there are two kinds of bubbles: good ones (railroads, dot-com) and bad ones (2008 housing). Both cause damage when they burst. But good bubbles leave behind infrastructure that transforms the economy. The railroad mania of the 19th century caused devastating depressions but also gave the United States a freight network that reshaped the continent. The dot-com crash wiped out trillions in market value, but the fiber-optic cables laid during the frenzy became the backbone of the modern internet. Without the bubble, the thinking goes, these technologies would have taken decades longer to arrive.
Apply this to AI, and the conclusion is straightforward. The current spending spree is not a bug. It is the mechanism by which transformative technology gets built. "Stop trying to make bubbles go away," an entrepreneur recently wrote. "The benefits of innovation outweigh the costs of volatility."
The numbers have stopped being numbers
The scale of the bet is hard to process. The four largest technology companies (Alphabet, Amazon, Meta, Microsoft) are on track to spend roughly $650 billion on AI infrastructure in 2026 alone. That is more than the GDP of most countries. In Q1 2026, AI startups captured about 80% of all global venture capital, with three companies (OpenAI, Anthropic, xAI) accounting for two-thirds of that. OpenAI raised $122 billion in a single round, the largest private raise in history, pushing its valuation toward $1 trillion.
When numbers reach this magnitude, they stop functioning as financial data and start functioning as signals of commitment. The "good bubble" frame provides a moral license to keep spending, because the downside (a crash) is reframed as a cost of doing business for the future.
The part that does not fit the analogy
There is one difference between AI infrastructure and historical precedents that the good-bubble argument tends to gloss over: durability.
Railroad tracks and fiber-optic cables last for decades. A GPU cluster, by contrast, becomes obsolete in three to five years. The chips that power today's frontier models will be e-waste before the debt used to buy them is paid off. If the bubble bursts before the next generation of hardware is fully amortized, the infrastructure left behind will not be a lasting asset. It will be a depreciation schedule.
Carlota Perez, whose work on technological revolutions is cited by both optimists and pessimists, has warned that the "installation phase" of new technologies tends to concentrate wealth and cause financial fragility before the "deployment phase" distributes benefits more broadly. In the AI case, the concentration is already extreme: 83% of global AI venture capital went to US-based companies in Q1 2026. The gap between frontier labs and everyone else is not narrowing.
Howard Marks, in his characteristically plain-spoken way, put it: "The investor doesn't say, 'Well, yes, I lost my money, but thank God it advantaged society.'" Accepting short-term pain as the price of progress is easy for people whose net worth is in the billions. It is a harder sell for someone watching their 401(k) get caught in the downdraft.
One second take-away
The "good bubble" is an interesting ideological shift. It provides pre-emptive absolution for speculative excess, wrapped in a flattering historical analogy. Whether it is right depends on one thing: whether the infrastructure left behind lasts long enough to matter. Rails and fiber did. GPUs might not. The doctrine is elegant. The physics is less obliging.
Links: Even Silicon Valley Says That AI Is a Bubble (The Atlantic) | Boom: Bubbles and the End of Stagnation (Stripe Press) | AI Funding in 2026 (The AI Insider) | Big Tech $650B AI Capex (Tech Insider) | Technological Revolutions and Financial Capital (Carlota Perez)