The global artificial intelligence market reached approximately $601.9 billion in 2026, driven by massive hyperscaler infrastructure spending near $700 billion. Major tech firms and institutional players dominate hardware and cloud sectors, while retail traders increasingly use AI tools for automated research and trading strategies. Market size and growth valuation projections are heading toward $3.6 trillion by the early 2030s. But one narrative I’ve seen requires correction, a vague “history lesson” analogy being applied to AI markets during earnings season — one that implies the whole sector faces demand-supply collapse the way earlier tech cycles did.

The commoditization pattern is real, but it’s isolated to the model layer, not the market as a whole. Foundation model APIs are converging toward marginal-cost pricing because weights are replicable and switching costs at that layer are near zero. That is a supply-side correction in a single input, not evidence that AI value creation broadly is collapsing.

The durable margin is migrating to whichever layer controls something that doesn’t replicate at zero cost: proprietary data feedback loops, embedded workflow state, distribution into existing buyer relationships, or compliance and liability ownership. “Non-technical differentiation” is a misleading catch-all for these, because they decay at different rates. Workflow lock-in erodes as agents get better at data portability. Brand and distribution are durable, but only for firms that already have them. Regulatory moats are rents, not differentiation.

The correct read for earnings season is a bifurcation thesis, not a cautionary tale about AI broadly. Commodity model providers should be valued like utilities on thinning margins. Application-layer and infrastructure-adjacent companies that own a non-replicable input should be valued on their ability to compound that input faster than competitors can copy the interface around it.