The Maze: AI is being sold as a universal deflation machine. The better analogy is narrower. Since 2000, U.S. prices fell hardest where goods became digital, tradable, quality-adjusted, and scalable. They rose fastest where capacity, credentials, regulation, and third-party payment still control supply. That is the AI lesson for commerce: models can lower production cost, but markets decide whether customers see lower prices.
The spread is the story, not the average. From January 2000 to December 2025, the BLS-based Mark J. Perry visual puts overall U.S. inflation at +92.6%. Hospital services rose +281.4%, college tuition and fees rose +196.7%, textbooks rose +176.9%, child care rose +158.8%, and medical care services rose +147.0%. On the other end, cell phone services fell -44.1%, computer software fell -73.0%, toys fell -74.2%, and televisions fell -98.1%. That is not one economy. It is two pricing systems sharing the same CPI wrapper.
Technology deflates when supply can scale faster than demand. Software, electronics, toys, and telecom services sit closer to markets where production can globalize, features can improve quickly, and marginal cost can collapse. BLS also adjusts CPI for product replacement and quality change, including through hedonic methods, so the television number should not be read as a simple shelf-price receipt. It is still useful. It shows how brutally competition and quality improvement can compress a category when the market lets output scale.
Healthcare and education are different animals. BLS medical CPI methodology tracks total reimbursement to providers, not only the patient co-pay, and hospital services include inpatient, outpatient, room, board, lab work, and other hospital-provided services. Eligible payers include self-pay, private insurance, and Medicare Part B. That structure matters. If the buyer, payer, provider, regulator, and credentialing body are not the same economic actor, lower production cost does not automatically become a lower retail price.
This is the missing caveat in the AI price debate. AI can reduce the cost of drafting, searching, coding, designing, translating, triaging, and advising. But the visual's winners and losers imply a harsher filter: prices fall fastest where buyers can switch and suppliers can scale without institutional permission. In accredited education, clinical care, legal services, and regulated financial advice, AI may make professionals more productive while the official price remains protected by licensing, liability, reimbursement, or brand trust.
Commerce will feel this first in software-like work. Product content, search relevance, creative testing, customer support, catalog cleanup, seller onboarding, analytics, and merchandising assistance behave more like software than hospitals. AI should push those costs down and make mediocre tooling harder to defend. Inventory, last-mile logistics, returns, compliance, and category expertise will deflate more slowly because humans, assets, risk, and local execution still sit inside the workflow.
Why it matters: The useful question is not "Will AI be a bubble?" Dot-com infrastructure crashed and still rewired the economy. The useful question is where AI turns scarcity into abundance. If the answer is software-like commerce operations, margins will move from tools and agencies toward operators who can redesign work around cheaper intelligence. If the answer is regulated services, expect productivity gains first, price cuts later, and plenty of institutions trying to keep the spread.
