When Deutsche Bank's economists asked leading AI models whether technology would solve the inflation problem, the machines delivered a sobering reality check that contradicts Silicon Valley's favorite narrative.
For two years, the consensus among tech's wealthiest investors has been almost evangelical: artificial intelligence is the great disinflationary force that will tame prices, supercharge productivity, and give the Federal Reserve room to keep interest rates low. Marc Andreessen and Vinod Khosla have championed this view with the conviction of converts. The logic feels airtight. Cheap technology replaces expensive human labor. Startups multiply and compete on price. Inflation retreats. There is just one problem. When Deutsche Bank's research team, led by Chief U.S. Economist Matthew Luzzetti, actually asked the AI systems themselves, the robots fundamentally disagreed.
The bank ran a straightforward experiment, documented in a research note published in late March. They posed a structured probability question to three leading models: Deutsche Bank's proprietary dbLumina, OpenAI's ChatGPT 5.2, and Anthropic's Claude Opus 4.6. The prompt asked each system to assign probabilities to four inflation outcomes over one-year and five-year horizons, ranging from AI meaningfully reducing inflation to actively raising it. The results landed like a bucket of cold water on the bullish consensus.
At the one-year mark, all three models agreed the most likely outcome was minimal impact. The striking part was what came next. Every single model rated AI raising inflation as more probable than AI meaningfully reducing it. Deutsche Bank's dbLumina put the odds of AI lifting inflation at 40 percent, compared to just 5 percent for a meaningful decline. Claude assigned 25 percent versus 5 percent. ChatGPT came in at 20 percent versus 5 percent. These are not pessimistic human analysts projecting their fears onto the data. These are the machines themselves identifying the flaw in their own disinflationary narrative.
The culprit cited consistently across all three models is the infrastructure binge that AI demands right now. Data centers are multiplying across the globe. Semiconductor demand has surged to historic levels, driving up prices for advanced chips from the likes of Nvidia and AMD. Electricity consumption from AI workloads is climbing sharply, with the International Energy Agency projecting data center power demand could double by 2026. That kind of demand-pull pressure does not lower prices. It raises them. And it is happening faster than the productivity gains that might eventually offset those costs.
As Fortune recently reported, even at the five-year horizon where models shift more toward disinflationary outcomes, the dramatic deflationary collapse some forecasters have predicted remains firmly in the realm of tail risk. The machines see a gradual cooling, not a paradigm shift.
The Ghost of White-Collar Futures
The Deutsche Bank findings gain weight when placed alongside more apocalyptic scenarios circulating in financial circles. James Van Geelen's Citrini Research, currently the top finance Substack, rattled markets in February with a thought experiment written as a dispatch from 2028. He described what he called "ghost GDP," a scenario where AI inflates national economic figures while mass layoffs hollow out household incomes. His logic is unsettling: machines spend zero dollars on discretionary goods. If corporate AI adoption triggers white-collar unemployment, which in turn triggers more AI adoption as companies seek cost savings, the result is a negative feedback loop that could push unemployment to 10.2 percent and crash the S&P 500 by 38 percent.
A March 2026 study from Anthropic itself added empirical weight to concerns about white-collar displacement. Researchers found that AI tools like Claude are theoretically capable of automating the vast majority of tasks in high-paying professional fields: 94 percent of computer and mathematical work, 90 percent of office and administrative roles. Yet actual adoption remains a fraction of that theoretical potential. No systematic rise in unemployment has materialized yet, but the gap between what AI can do and what companies are currently using it for represents a massive overhang of latent disruption.
For startup founders and business leaders, the takeaway is nuanced but important. The disinflationary promise of AI is real, but the timeline is considerably longer than markets have been pricing in. The capital expenditure required to build AI infrastructure is itself inflationary in the near term, and the productivity gains that could offset those costs remain years away from full realization. Companies betting on AI-driven cost reductions in the next 12 to 18 months may find themselves on the wrong side of this equation.
Watch the semiconductor cycle and data center buildout costs closely. Those are the leading indicators of whether AI remains a net inflationary force or finally begins delivering the deflationary dividend that Silicon Valley has been promising.
