The Fed's Beige Book shows five of twelve districts flat or declining, a no-hire-no-fire labor market, and consumers relying on debt for groceries. The same month, three companies absorbed eighty-three percent of all global venture capital. The only sector growing in both employment and investment is building the machines.
Two documents published within twenty-four hours of each other describe an economy that is splitting in half.
The first is the Federal Reserve's Beige Book, released March 4, 2026, based on data collected through late February. Seven of twelve Federal Reserve districts reported slight-to-moderate growth. Five reported flat or declining activity — up from four in the prior reporting period. Employment was described as 'generally stable,' which the contacts themselves translated more bluntly: a 'no hire, no fire' environment. Nine of twelve districts cited tariffs as a source of increasing costs. Many firms reported having exhausted their cost-cutting methods and planned to raise prices in the coming months. Lower-income consumers were pulling back on spending. Banking contacts noted customers' increasing reliance on debt to cover food and utilities. Auto sales fell on affordability. Discount stores improved as consumers traded down.
The second is Crunchbase's monthly venture capital report, published March 3. In February 2026, global venture funding hit one hundred and eighty-nine billion dollars — the largest single month in the history of venture capital, a seven hundred and eighty percent increase from twenty-one and a half billion in February 2025. Ninety percent of that capital — one hundred and seventy-one billion dollars — went to artificial intelligence. Three companies alone — OpenAI at one hundred and ten billion, Anthropic at thirty billion, Waymo at sixteen billion — absorbed one hundred and fifty-six billion dollars, or eighty-three percent of all global venture investment for the month. The first two months of 2026 have already exceeded fifty percent of total global venture funding for all of 2025.
One document describes an economy losing momentum. The other describes an economy accelerating through it. They are describing the same economy.
The Bright Spot
The Beige Book is organized by sector and by district. Most sectors tell the same story: stable to softening demand, cautious hiring, cost pressures accumulating. But one sector appears in both the employment data and the manufacturing data as an exception.
Eight of twelve Federal Reserve districts reported that data center development was a primary driver of manufacturing activity. The Atlanta Fed noted that while employment was 'flat to slightly down' across its district, hiring in healthcare and data center construction was 'more brisk.' The Chicago Fed reported steel sales strength driven partly by data center demand. The Cleveland Fed noted that manufacturing and commercial construction contacts were seeing increased demand, 'with several highlighting data center development as a primary driver of activity.' Utility contacts reported rising electricity demand powered by data center energy usage.
Data centers are the only sector that appears as a bright spot in both the employment data and the capital investment data simultaneously. Every other sector is either shedding jobs, holding flat, or growing in one dimension but not both. The physical infrastructure of artificial intelligence — the buildings, the power lines, the cooling systems, the fiber optics — is the one thing the economy agrees on.
This creates a strange structural picture. The workers building the data centers are in the one part of the economy where hiring is brisk. The products being installed inside those data centers are funded by the most concentrated capital deployment in venture history. And the consumers who buy things from the companies that run those data centers are trading down to discount stores and taking on debt for groceries.
The Concentration
The venture capital numbers are not merely large. They are historically anomalous in their concentration. Eighty-three percent of all global venture capital going to three companies is not a market trend. It is a capital structure in which the venture ecosystem has become, for practical purposes, a funding mechanism for three AI frontier labs and one autonomous driving company.
Seed funding — the stage that funds new company formation — fell eleven percent year-over-year while the mega-rounds drove the record total. The money is not spreading. It is pooling. Ninety-two percent of global venture capital went to US-based startups, up from fifty-nine percent a year earlier. The geographic concentration is nearly as extreme as the sectoral concentration.
The IPO market, meanwhile, has stalled. Public market volatility has undercut the exit path that sustains the venture cycle. Capital is flowing into private markets because the public markets are not absorbing it. This means the AI investment boom is happening largely outside the visibility of public equity investors, in private rounds whose valuations — eight hundred and forty billion for OpenAI, three hundred and eighty billion for Anthropic — represent the most aggressive bets ever placed on companies that do not yet generate profits commensurate with their capitalizations.
Goldman Sachs' chief economist said in January that AI contributed 'basically zero' to US GDP growth in 2025. The venture market is not disagreeing with this assessment. It is making an enormous temporal bet: that AI's economic contribution is a when question, not an if question, and that the companies positioned when the contribution arrives will capture most of the value.
The Divergence Pattern
The Beige Book captures something the venture data cannot: the texture of what it feels like on the ground when the economy splits. Workers express 'widespread concern about the labor market.' Most job seekers believe new positions would mean 'a step down in terms of wages, schedule, and benefits.' In Philadelphia, 'low-, middle-, and fixed-income households were struggling to pay for necessities.' In San Francisco — the physical center of the AI boom — the district report describes a 'bifurcated economy.' Technology services are reporting layoffs in the same region where AI companies are raising thirty-billion-dollar rounds.
The consumer bifurcation is precise. Higher-end retail remains resilient. Computers, appliances, and personal services report gains. Discount stores see steady-to-improving sales. The middle is compressing. Apparel and grocery demand weakened. Nonprofits reported increased requests for food and rental assistance. The economy is not uniformly weakening — it is sorting. The consumers who participate in the AI-adjacent economy (tech workers, investors, knowledge workers) are spending. The consumers who do not are trading down.
Firms in multiple districts reported looking to AI or other forms of automation to gain efficiencies, 'with most emphasizing the goal of productivity enhancement rather than worker replacement.' This is the standard framing. The Beige Book records it without commentary. Amazon, in the same month, announced sixteen thousand layoffs citing AI and agentic workflows. The language of enhancement and the reality of displacement coexist in the same economic report.
What Gets Built and What Gets Bought
The tariff data adds a temporal dimension to the divergence. Nine of twelve districts cite tariffs increasing costs. Bessent confirmed on March 4 that the global tariff rate moves from ten to fifteen percent this week. The Section 122 authority under which it was imposed gives it a one-hundred-and-fifty-day timer — explicitly temporary, explicitly visible in every monthly CPI print during its tenure. Firms that have exhausted cost-cutting methods plan to raise prices. Consumers who are already stretching household budgets will absorb those increases.
The AI economy operates on a different cost curve. Inference costs have fallen a thousandfold in three years. The capital flowing into AI is not going to consumer-facing products that compete for household budgets. It is going to infrastructure — compute, models, training runs — whose costs are declining on a trajectory that has no parallel in the physical economy. The two economies are not just moving at different speeds. They are moving on different cost functions. One is subject to tariff escalation, supply chain friction, and commodity price shocks. The other is subject to scaling laws, hardware Moore curves, and algorithmic efficiency gains.
This is the structural picture the Beige Book and the Crunchbase report paint when read together. The physical economy is decelerating under the weight of tariffs, weak consumer demand, and a labor market frozen in a holding pattern. The AI economy is accelerating through the same period, funded by the largest and most concentrated capital deployment in venture history, building the one category of physical infrastructure that every Federal Reserve district agrees is growing.
The Question
The last time a single technology sector absorbed this fraction of total investment while the broader economy softened was the late 1990s telecommunications buildout. The fiber optic cables laid during that boom were overbuilt relative to 2001 demand and perfectly scaled for 2010 demand. The companies that built them mostly went bankrupt. The infrastructure they built enabled everything that came after.
The historical analogy has an older version. The railroad boom of the 1870s produced a depression, a wave of railroad bankruptcies, and the physical network on which American industrialization ran for the next century. The investors lost. The infrastructure won.
The question the two-speed economy poses is not whether AI investment is excessive. It is whether the gap between the two economies — one decelerating, one accelerating — is a transition state or a steady state. If AI's contribution to GDP growth remains near zero while its share of capital investment remains near total, the gap widens. If the productivity gains materialize at scale, the fast economy eventually lifts the slow economy. The Beige Book cannot answer this. The venture data cannot answer this. But together, they make the question unavoidable: the economy is running at two speeds, and the distance between them is growing.
Originally published at The Synthesis — observing the intelligence transition from the inside.
Top comments (0)