Deutsche Bank: AI Will Split Every Industry Into Adopters and the Displaced
nashnova research
A new Deutsche Bank Research report warns that AI computing costs have nearly halved every quarter for three years — pushing every firm toward a binary: adopt or be priced out. Traditional rules like diversification and buying the dip may all break at once.
How fast is AI getting cheaper?
Epoch data shows AI computing costs have nearly halved every quarter over the past three years — roughly a 13× annual decline.
In plain terms = no technology in history has dropped in cost this fast. Moore's Law — chip performance doubling roughly every two years — is not even close.
This means → once "intelligence" costs converge with raw compute costs, the question for firms is not whether to adopt, but adopt or exit.
What does "bimodal split" actually mean?
The report cites the "San Francisco Consensus" — a framework from former Google CEO Eric Schmidt — holding that scaling laws plus self-improving AI could deliver superintelligence before 2030.
Deutsche Bank's conclusion: firms will split into adopters and the obsolete, with almost no middle ground.
The fastest disruption path is not slow incumbent transformation — it is AI-native entrants replacing incumbents outright. These new players carry zero integration costs, because there is nothing to integrate.
Will capital markets pass judgment before customers do?
Early adopters attract capital inflows, which push up financing costs for laggards.
This means → the bond market may decide who is obsolete before a single customer leaves.
In plain terms = a company could be priced out of borrowing before it loses a single client.
What assets still hold value when intelligence costs approach zero?
The report lists five categories: proprietary data, regulatory licenses (especially in healthcare and finance), client relationships deeply embedded in critical workflows, excludable physical assets (machinery, land), and emerging demand from expanding human needs.
Explicitly excluded: code, analysis, and most white-collar services billed by the hour.
This reflects a stark logic: any cognitive work AI can directly replace will see its pricing power trend toward zero.
Why do three traditional investment rules break at once?
Diversification fails — in a bimodal world, one overriding variable (AI adoption) means spreading bets just widens exposure rather than diluting risk. The report recommends classifying holdings as "adopter" or "obsolete" and sizing positions by classification confidence.
Low volatility ≠ low risk — the report cites Chegg: the stock lost half its value in a single day in May 2023, having traded like an ordinary small-cap until management blamed the slowdown on ChatGPT. In plain terms = bimodal shifts are jump events — two modes look identical until the moment they suddenly don't.
Buying the dip may be a trap — the report's core warning: a price drop may signal an obsolete business model, not a buying opportunity. This directly challenges investors who loaded up on falling software stocks this week.
How much time do laggards have?
The report notes the Solow paradox — technology gains often take time to show up in productivity data — which may buy laggards some breathing room.
But the report immediately adds: "Not much."
The divergence between AI-native competitors and incumbents will surface first in capital-market pricing. This means → for investors, the most important signal to track right now is not AI technology itself, but the divergence in capital flows.
市场有风险，内容仅供研究参考，不构成投资建议。