Federal Reserve Chairman Kevin Warsh: AI Will Drive Down Prices of Most Goods and Services
Kevin Warsh, Chairman of the U.S. Federal Reserve, stated in a speech at Jackson Hole that artificial intelligence could become a new factor of production, leading to lower costs for businesses and consumers if productivity experiences a sustained leap.
Warsh noted that capital is continuously flowing into AI infrastructure, with the pace of innovation exhibiting characteristics of 'super Moore's Law'; users of large language models acquire model capabilities by purchasing tokens, with annualized token sales from two leading labs exceeding $100 billion, a growth of over 500% compared to a year ago.
He suggested that the current value of AI may primarily flow to holders of scarce assets, including AI labs, chip manufacturers, energy suppliers, and cloud service providers; however, the long-term key variable is whether productivity gains can be transferred to business clients and consumers.
Warsh specifically pointed out that the pricing paths of cutting-edge models and older models may diverge: access to high-performance models can maintain a high premium, while the token prices of older models may eventually drop to near marginal costs, resulting in ongoing technological deflation.
In the short term, AI does not solely push prices down. The construction of large-scale data centers is driving up costs for construction and storage chips, and AI capital expenditures have become a significant part of the growth in U.S. investment in equipment and intangible assets; Federal Reserve data shows that related investments are driving a quarterly growth rate of about 9% in such capital expenditures, with over half of the growth attributable to AI construction.
Warsh emphasized that research on AI productivity will not alter current monetary policy decisions. The U.S. PCE inflation rate over the past 12 months was 3.7% and 4.1% over the past 6 months, both above the Fed's 2% target; 54% of the goods and services in the PCE basket saw price increases exceeding 3% in the past 12 months, thus the Fed's current priority remains to reduce inflation.
In market mechanisms, funds are concentrating on buying scarce supply segments such as AI labs, chips, cloud computing, electricity, and data centers, flowing into infrastructure through corporate capital expenditures and credit expansion; if token prices continue to decline with model maturity, application companies and end users will benefit, while supply chain companies reliant on high-priced computing power, storage, and construction orders will face pressure on profit margins and pricing power. The market is simultaneously trading two opposing main lines: long-term AI deflation expectations favor productivity assets, while short-term investment demand and inflation stickiness support higher interest rate expectations.
Source: Public Information
ABAB AI Insight
This is not the first time Kevin Warsh has incorporated technological advancements into the framework of monetary policy. He previously served as a Fed governor and participated in normalizing forward guidance after the 2008 financial crisis; however, he has now explicitly criticized long-term, detailed interest rate guidance as potentially trapping markets and policymakers in a 'mirror hall effect'. In his speech, he referenced reflections on monetary policy from 2021: pre-committed communication may slow the response to sudden inflation changes. Today, he views AI as a variable that could change the supply side, but refuses to let unfulfilled productivity promises replace real inflation data.
The capital path has shifted from software subscriptions to heavy asset production systems. The value chain identified by Warsh includes AI labs, chip manufacturers, energy producers, and cloud service providers: labs first productize model capabilities through token sales, while cloud providers and chip manufacturers handle training and inference loads, and electricity and data centers accommodate the ongoing physical demand for expansion. The Fed assesses that more than half of the growth in U.S. investment in equipment and intangible assets this year is related to AI construction; this means that the current AI boom is primarily driven by capital expenditures, credit availability, and infrastructure development, rather than by a decline in end prices across the industry.
This is more akin to the capital structure of the early 19th century railroads, power grids, and the internet, rather than ordinary software upgrades. During the railroad era, capital was first concentrated in tracks and trunk lines, with fare reductions and widespread productivity improvements following; in the internet era, massive fixed investments in fiber optics, servers, and cloud computing were made first, followed by declines in unit computing and distribution costs. AI is currently in a stage of 'infrastructure expansion, with returns yet to be determined': leading models may maintain prices due to capability scarcity, while older models may be pushed toward marginal costs like general cloud computing resources. Warsh himself acknowledges that it remains uncertain where capital returns will ultimately fall—whether with model companies, chip manufacturers, energy providers, cloud vendors, or users.
Essentially, this represents a transfer of pricing power: when model capabilities are scarce, pricing power is concentrated in training capital, advanced chips, electricity access, and cloud infrastructure; as model performance spreads, inference efficiency improves, and alternative models increase, pricing power will shift to application-layer companies that possess customer relationships, business data, and distribution channels. Short-term capital expenditures will elevate construction, storage, and financing demands, coexisting with inflationary pressures; only when AI replaces enough repetitive labor and reduces enough unit service costs will productivity gains potentially lower the general price level. The Fed's current stance of not easing policy based on AI cost-reduction expectations is precisely because long-term supply expansion and short-term demand, investment, and price pressures can coexist.
ABAB News · Cognitive Laws
Infrastructure prices rise first, then prices fall after technology becomes widespread
Scarcity determines profits, substitutes determine prices
Productivity is long-term deflation, capital expenditure is short-term inflation