Persistence Analytics' Osnato: AI capex outruns verified evidence

Osnato grounds his argument in his own background: 13 years on the floor of the New York Stock Exchange, where he says a trade could be executed and documented while the assumptions behind it proved less durable than expected. He applies that lesson to AI infrastructure by pointing out that the industry's commitments run on mismatched clocks — compute contracts lasting years, debt maturing decades out, servers on short technology cycles, and power infrastructure outliving several generations of the equipment it supports — while enterprise customers are still learning what they will use and pay.

The piece is explicit that none of this proves the investments are unsound. Instead, Osnato separates claims that are often bundled: contract existence versus economic durability, announced capacity versus energized capacity, and a guarantee versus the continued usefulness of the asset behind it. He warns that apparently distinct exposures may rest on a small set of interconnected customers, infrastructure providers, funding sources or technology assumptions, so diversifying contracts may not diversify the dependency underneath.

His proposed remedy is a revalidation gate rather than a forecast about AI's success — a documented chain from the representation relied upon through its embedded assumption, supporting evidence, dependencies, the party bearing downside and the material change that would force reassessment. He notes that engineers, lawyers and finance teams each own different pieces, and that independent decision assurance asks whether those records together establish the proposition a capital allocator is relying on.