Yesterday I was discussing AI investing with a friend, and one thought kept coming back.
One of the riskiest investments today might not be Bitcoin. Or venture capital. Or early-stage AI startups.
It might be betting that the world’s largest public companies will continue looking roughly the way they do today.
That’s a much bigger assumption than most investors realize.
AI Doesn’t Need to Kill Companies
People often imagine disruption as extinction. Kodak disappears. Blockbuster collapses. A new winner takes everything. That’s dramatic.
Reality is usually more subtle. Companies survive. Their multiples don’t. AI doesn’t necessarily need to eliminate a business. It only needs to reduce the value of what made that business special.
Software Is First
The first obvious casualty is software itself.
If your product is primarily code…
…and your competitive advantage is simply writing better code…
…you’re entering a very uncomfortable decade.
Modern AI agents are already compressing software development costs at astonishing speed. The result isn’t that SaaS disappears. Databases remain. Systems of record remain. Identity remains. Enterprise integrations remain. But many application-layer businesses become dramatically easier to replicate.
When differentiation shrinks, multiples usually follow.
Then Knowledge Work
Banks won’t disappear. Insurance companies won’t disappear. Consumer brands won’t disappear. Neither will airlines or railroads. But look inside almost any large corporation.
How much of its operating cost comes from office work? Finance. Legal. Compliance. Operations. Marketing. Customer support. Internal software. Documentation. Analysis. Planning. Much of that is already becoming partially automatable.
If 80% of knowledge work gradually becomes AI-assisted—or fully automated—the economics of those businesses inevitably change. Not overnight. But permanently.
“Who Will Buy Anything?”
Whenever this comes up, someone asks the same question.
“If everyone loses their jobs, who buys the products?”
I don’t think that’s the right framing. Most people today already can’t afford the best healthcare. Or the best education. Or premium financial services. Or luxury housing. Or advanced robotics. The interesting possibility isn’t that demand disappears. It’s that many things become dramatically cheaper to produce.
Lower production costs create entirely new consumers. We’ve seen this pattern repeatedly throughout industrial history.
AI may simply compress it into a much shorter timeframe.
Physical Industries Move More Slowly
Software changes in months.
Construction doesn’t. Factories don’t. Logistics doesn’t. Housing certainly doesn’t.
The robots already exist. Autonomous warehouses exist. Construction robots exist. Industrial automation exists.
The bottleneck is no longer technological. It’s deployment. Economic diffusion takes time.
A robot capable of building houses doesn’t immediately mean hundreds of thousands of houses get built next year.
Infrastructure has inertia. Software has almost none.
What About Crypto?
Crypto presents an interesting thought experiment. Its entire value ultimately depends on software, cryptography, incentives, and security.
Large language models are becoming increasingly capable at vulnerability discovery, exploit generation, protocol analysis, and automated security research.
That doesn’t mean Bitcoin or Ethereum suddenly become worthless. But it does suggest that software-native assets may face risks that traditional valuation models were never designed to consider.
The Real Bet
None of this means the S&P 500 suddenly collapses. Companies like NVIDIA, Microsoft, Alphabet and others may continue benefiting enormously from AI itself. The point is different.
Markets may still be pricing many businesses as though the next decade will resemble the last one.
I’m increasingly convinced it won’t. The biggest investment risk today isn’t volatility. It’s assuming stability where technological acceleration is becoming exponential. Because in an AI economy, standing still may become the fastest way to fall behind.