Anthropic is reportedly set to tell potential investors its addressable market tops $30 trillion surpassing the $28.5 trillion figure SpaceX cited around its own IPO, according to the WSJ. The raise itself could top $100 billion at a roughly $2 trillion valuation.
Check that number against the real economy. The IMF estimates 2026 global nominal GDP at about $126 trillion, so this claimed market equals roughly 24% of everything the planet produces this year about the size of the entire US economy on its own. Worldwide labor compensation runs a bit over half of global GDP, around $65 trillion, meaning the pitch effectively claims AI could capture close to half of all wages paid to every worker alive. A TAM figure like this is a theoretical ceiling: it assumes near complete substitution and full price capture, neither of which has ever actually occurred.
Two prior booms were pitched on similarly outsized numbers. British railway mania: by 1847, capital poured into railways reached roughly 7% of GDP and swallowed close to half of all investment in the economy £44 million in the peak year alone, more than twice the Crown's entire military budget. The underlying predictions about rail transforming commerce turned out correct. Yet railway stocks still lost about two thirds of their value from the 1845 peak, and dividends collapsed from 7% to 2%. Track mileage tripled between 1844 and 1850 and remained in productive use for the next hundred years.
Telecom, 1996–2001: over $500 billion poured into fibre buildout, mostly funded by debt. By most estimates, 85–95% of that fibre sat unused ("dark") after the crash, wiping out roughly $2 trillion in market value. That same unused fibre is what carries today's video calls.
In both episodes, the underlying market opportunity was genuine — but the investors who funded the buildout weren't the ones who captured the value. The gains flowed instead to whoever ended up closest to the customer relationship: mail order retailers monetised the rail network; AWS and Netflix monetised the leftover fibre. That's the lesson a CIO should take from a $30 trillion headline. If AI models really can handle most knowledge work, the margin will flow to whoever controls the proprietary context, the routing layer, and the evaluation system deciding which model handles which decision not necessarily to whoever built the models. Compute itself is becoming cheaper and more commoditised by the day. What's actually being priced into someone else's market size slide is your organisation's own institutional knowledge.