San Francisco is home to OpenAI and Anthropic, two frontier AI labs worth a combined $2 trillion, according to The Economist. Ninety-one additional AI unicorns cluster in the Bay Area, adding another $600 billion in private-market capitalization. By any measure of innovation capital, San Francisco stands as the most technologically advanced city on earth. Yet that same analysis found that AI wealth has not translated into broad urban prosperity. The middle class continues to shrink. The city holds more technology, more data, and more innovation capital than anywhere in history, and its operating systems still cannot convert any of it into adaptive economic outcomes.
This is not a San Francisco problem. It is an American city problem. CBRE’s Q1 2026 U.S. Office Market Report recorded eight consecutive quarters of positive net office absorption in New York City, with Q1 alone totaling 6.9 million square feet, the highest first-quarter figure since 2020. San Francisco posted 3.4 million square feet in new leasing during Q1 2026, up 43% year over year and its strongest quarter since 2019, according to Colliers’ Q1 2026 San Francisco Office Market Report. The predictions of urban collapse proved wrong. But the recovery exposed a deeper issue: cities equipped themselves relentlessly with sensors, dashboards, and real-time data, yet designed most of that infrastructure to measure performance rather than respond to volatility.
The Same Infrastructure, Overloaded and Empty in the Same Week
That gap between measurement and responsiveness shows up in three places. The MTA now publishes daily ridership data showing pronounced midweek peaks and sharp Monday and Friday declines, a pattern that has hardened into a new normal as hybrid work stabilizes. Kastle Systems, which tracks access-control data across 2,600 buildings and 41,000 businesses, reported in December 2025 that A+ office buildings hit 95.5% occupancy on a peak Tuesday, while Friday occupancy across all tracked buildings averaged just 31% of pre-pandemic levels. The same cities experience infrastructure that is simultaneously overloaded and underloaded depending on the day of the week.
Permitting tells a parallel story. San Francisco sits in the middle of a well-documented housing crisis. Between January 2024 and August 2025, the city cut its median housing-permit processing time from 605 days to 280 days, according to a KQED analysis of Planning Department data. That represents a genuine, hard-won improvement. But 280 days still means a city that responds to a housing shortage in nine-month increments while rents move in real time. KQED also found that more than 1,300 applications remain in the backlog, with an average wait of 1,489 days as of October 2025. The system got faster. It still cannot keep pace with the problem.
In New York City, the explosion of rideshare pickups and last-mile delivery overwhelmed curbside infrastructure built for a different era. Mayor Mamdani launched a new Office of Curb Management in April 2026 to oversee 6,300 miles of streetside lanes, a bureaucratic response to a problem that existed for years before the governance structure caught up. That lag captures the pattern.
Governance Built for Stability in an Era of Constant Change
The issue is structural. Municipal systems evolved around deliberation, accountability, and risk mitigation when conditions changed slowly. Today, those systems face environments where conditions shift constantly, but governance structures fragment authority and delay coordination.
Every building in a city functions as a node in this system. The office tower that empties on Monday. The mixed-use development waiting 1,489 days for permits. The lobby competing with delivery drivers for curbside access. When those nodes generate real-time data about how people actually use space and feed it to the systems around them, the city learns. When they sit dark, the city guesses. Most cities still guess.
Singapore offers a counterexample. Its Smart Nation initiative connected transport systems across government departments and deployed the GLIDE adaptive signal network, which adjusts green-light timing based on real-time traffic demand. Singapore’s Land Transport Authority reports that the system cut average travel times by 30% and reduced congestion at major intersections by 15%. That result did not come from better sensors. Singapore had sensors for decades. It came from a governance architecture that enables real-time decision-making across functions, something most American cities still lack.
What Programmable Cities Actually Require
Singapore demonstrates that adaptive urban systems work when governance enables them. The question for American cities is what it actually takes to get there. The answer goes beyond smarter sensors or better analytics. It requires genuinely programmable infrastructure that reconfigures itself based on real-time conditions rather than quarterly planning cycles.
That demands governance models that enable faster decision-making without sacrificing accountability. It demands procurement frameworks that let cities test, learn, and scale solutions in months rather than years. It demands cross-department coordination operating through a shared data infrastructure and buildings designed for continuous recalibration rather than fixed configurations that made sense two decades ago.
Most critically, it requires accepting that volatility is permanent. The Tuesday parking shortage and the Monday emptiness are not anomalies to correct. They are the new operating conditions. Cities that cling to stability as a design principle will keep building infrastructure for a demand pattern that no longer exists.
The technology already exists. The data already flows. The question is whether governance structures and procurement models can evolve fast enough to use them. The cities that figure this out will not just recover from the disruptions of the past five years. They will compound the advantages that made them powerful in the first place. The rest will keep publishing dashboards that show exactly what is going wrong while the underlying system runs on outdated code.
The opinions expressed in Fortune.com commentary pieces are solely the views of their authors and do not necessarily reflect the opinions and beliefs of Fortune.
The post San Francisco has $2 trillion in AI wealth and can’t fix its own city. That’s every city’s problem appeared first on Fortune.
