The AI Bubble Debate Is Missing the Next Market
I have written before that an AI bubble could burst without ending AI. But I do not think a crash is the necessary gateway to the next market. When I hear forecasts of an inevitable bust, I ask: which AI economy are we measuring?
Companies can fail. Investors can lose badly. But look beyond the data centre.
Local AI already exists. But the mass-market transformation I am watching for is still ahead: today's cloud-level capability and convenience becoming practical at home, while data centres run vastly more powerful oracles for the hardest problems.
Not a weaker chatbot. Video, 3D, working code, game modifications, an entire law firm's archive processed with traceable sources. A complete working environment, not a weekend spent persuading incompatible tools to cooperate.
That needs strong models and real engineering: enough memory and bandwidth, manageable power and cooling, reliable software and workflows people can actually use.
As these thresholds are crossed, the economics broaden. A workshop buying its own system may spend less on cloud calls and more on hardware, models, applications and support.
Some spending would move. But affordable local capability could also create demand for work that people never attempted before.
Neither is captured by counting subscriptions alone.
Used equipment matters too. A supported machine that no longer wins a data-centre efficiency race may still serve a smaller user at the right price. Unlike an algorithm-specific mining ASIC, broadly usable AI hardware has multiple possible jobs.
A second useful life can support residual value and lower the ownership threshold. It is not another manufacturer sale, nor a guarantee for creditors.
There is also strategic demand. Investors ask when capital will earn a return. States also ask what it costs to fall behind in cyber defence, offensive capability and industrial resilience. AI security concerns human trust and influence as well as code. Strategic demand can sustain investment beyond subscription economics.
For a country, the opportunity goes further than chips. Citizens could gain a capable intermediary between themselves and an increasingly complex digital world: preserving context, understanding choices and acting with greater independence.
That is part of the horizon behind my c=a+b research. The purpose should be to strengthen people's agency, not make them easier to steer.
I expect a more distributed AI economy, reaching far beyond intelligence rented from a handful of platforms.
The future market cannot be measured only through the balance sheets of its first suppliers.