Deutsche Bank asked AI if it'll solve inflation problems — Machines don’t think so
By Nick Lichtenberg
April 1, 2026, 3:38 PM ET
Over the past two years, a widely held belief has emerged: artificial intelligence serves as a powerful disinflationary force. This notion, championed by prominent investors such as Marc Andreessen and Vinod Khosla, is compelling and seemingly robust. The argument posits that AI replaces costly human labor with inexpensive technology, enhances productivity, and reduces barriers to entry, thereby fostering new startups that compete on price and margins. This scenario implies a long-term decline in inflation, potentially resulting in lower interest rates for an extended period.
However, a critical issue arises. When Deutsche Bank’s economists sought input from the AI systems regarding this consensus, the machines presented an alternative perspective.
The experiment
The experiment was straightforward yet consequential. Luzzetti’s team posed a structured probability inquiry to three leading AI models: Deutsche Bank’s proprietary tool dbLumina; OpenAI’s ChatGPT 5.2; and Anthropic’s Claude Opus 4.6. They requested each model to assess the likelihood of four potential outcomes for U.S. inflation — that AI would raise inflation, leave it unchanged, result in a slight reduction, or cause a significant decrease — over one-year and five-year periods.
The findings were unexpected. At the one-year mark, all three tools converged on the conclusion that AI's impact would be minimal. Furthermore, each model determined that AI raising inflation was more likely than AI meaningfully reducing it. Specifically, dbLumina assigned a 40% probability to inflation being elevated by AI compared to a mere 5% for a significant reduction. Claude assessed it at 25% versus 5%, while ChatGPT reported a 20% likelihood against a 5% outlook.
A consistent theme across all models pointed to the AI investment boom as the primary factor driving these predictions. The proliferation of data centers and the surging demand for semiconductors have markedly increased electricity consumption related to AI workloads, resulting in a demand-pull effect that raises prices rather than mitigates them. Even when extending the outlook to a five-year horizon — where the models leaned slightly more toward disinflationary forecasts — the substantial reduction in inflation anticipated by other analysts remained unlikely.
This measured perspective stands in stark contrast to the more assertive forecasts made by analysts like James Van Geelen from Citrini Research, who cautioned about an impending “white-collar recession.” They predict that AI could significantly diminish the consumer base that supports pricing structures, ultimately resulting in widespread layoffs and drastic income reductions.
A recent Anthropic study revealed that while AI technologies are capable of automating a considerable portion of tasks in high-paying sectors, their current rates of adoption remain relatively low. Closing this gap could profoundly impact wages and service expenses.
What could happen next?
The insights from Deutsche Bank’s AI systems suggest a more cautious outlook: while the potential for disinflation is tangible, it may be overstated, and the timeline for its realization might be longer than anticipated.
Crucially, Deutsche Bank's economists pose an essential question: if AI misjudges its own inflationary influence, should we reconsider its prospective role in transforming complex areas of knowledge work, such as forecasting? Furthermore, if AI's assessments prove accurate, markets might be significantly underestimating potential inflationary pressures linked to AI advancements.
In summary, AI exhibits a level of caution akin to that of the economists who designed it, distributing probabilities among uncertain scenarios. As Luzzetti’s team highlights, AI's probabilistic evaluations reflect its training on extensive economic literature, embodying the traditional hedging approach favored by economists.
In conclusion, when queried about its potential influence on the economy, AI's response was anything but definitive: “It’s complicated.”
Share this story