California Spends $325 Billion a Year. Here's What It Gets.

In fiscal year 2025-26, California will spend $324.7 billion. That figure comes from SB 101, the enacted budget bill, with $231.9 billion flowing from the General Fund alone. For context, that is more than the entire GDP of Finland.

California charges a top marginal income tax rate of 14.4% (13.3% plus a 1.1% mental health surcharge on income above $1 million), the highest of any state. Its combined state and local sales tax averages 8.68%, among the top five nationally. The tax revenue is not the problem. Where the money goes, and what it produces, is.

$24 Billion on Homelessness. It Got Worse.

Between 2019 and 2024, California spent approximately $24 billion on homelessness programs across at least 30 separate programs, according to a 2024 Stanford SIEPR policy brief. Over that same period, the state's homeless population grew from roughly 151,000 to over 181,000, an increase of about 30,000 people. That works out to roughly $160,000 per homeless person based on the 2019 count. The number went up, not down.

In 2024, the California State Auditor released Report 2024-102, a damning assessment of this spending. The Interagency Council on Homelessness (Cal ICH), the state body created to oversee these programs, had not consistently tracked spending or measured outcomes. The auditor's conclusion: policymakers "are likely to struggle to understand homelessness programs' ongoing costs and achieved outcomes." The state wrote $24 billion in checks and did not bother to measure whether anything worked. The Legislative Analyst's Office (LAO) separately noted that the 2024-25 budget included a $15.3 billion multi-year plan to tackle homelessness, "the most significant investment in state history," while the metrics to evaluate whether prior billions achieved anything remained undefined.

A 2025 point-in-time count showed a 4.3% statewide decrease in homelessness. Progress, finally. But even with that reduction, California still accounts for roughly 28% of the entire nation's homeless population while comprising about 12% of its residents. As the Stanford SIEPR researchers noted, roughly 30% of Permanent Supportive Housing (PSH) beds are reserved for the chronically homeless, 30% for veterans, and only 1% for youth, leaving at least 39% of PSH units without targeted allocation for the populations driving the crisis.

A $128 Billion Train to Nowhere

In 2008, California voters approved Proposition 1A to build a high-speed rail system connecting San Francisco to Los Angeles. The ballot measure told voters the system would cost $33 billion.

Seventeen years later, not a single passenger has boarded a train. The California High-Speed Rail Authority's current cost estimate for the full Phase 1 system runs between $89 billion and $128 billion, depending on risk scenario. That is a 288% cost overrun at the high end, and construction is still years from completion on even the initial 171-mile segment between Merced and Bakersfield. That segment alone is $7 billion short of its funding needs. In February 2026, the Authority approved another large cost overrun, and the Inspector General's own reports have documented persistent management and contracting failures. The Hoover Institution calculated the per-mile cost at roughly $167 million per mile for the Merced-Bakersfield segment, more expensive per mile than any comparable system in the world.

For reference, Japan built the original Tokaido Shinkansen in 1964. It runs 320 miles from Tokyo to Osaka. Construction took five years and cost the inflation-adjusted equivalent of about $32 billion. Japan's train has been operating for 61 years with zero passenger fatalities. California has spent more money on a train that does not yet exist.

Below Average in 4th-Grade Math

California spends approximately $16,993 per pupil on K-12 education, according to NCES data. That ranks above the national average. Yet on the 2024 National Assessment of Educational Progress (NAEP), California's 4th graders averaged 233 in math, compared to the national average of 237. In 8th-grade math, California scored 269 against a national average of 272. Both grades placed California in the bottom half of states. The Public Policy Institute of California (PPIC) reports that between 2019-20 and 2024-25, state K-12 funding increased nearly 50% (22% inflation-adjusted), yet test scores have not improved proportionally.

Compare this to Florida, which spends $12,491 per pupil and scored 240 in 4th-grade math, ranking 4th nationally. Utah spends $9,575 per pupil, roughly half of what California spends, and outperforms it on NAEP math at both grade levels. The LAO's 2024-25 California Spending Plan analysis noted that the state's Proposition 98 guarantee now directs over $23,000 per pupil when all funding sources are combined, yet California still cannot close the gap with states spending far less. Money alone does not explain educational outcomes. What you do with the money matters more than how much you collect.

Infrastructure: C-Minus for the Fifth Richest Economy on Earth

In December 2025, the American Society of Civil Engineers released its 2025 Report Card for California's Infrastructure. The grade: C-minus. That is the same grade the state received in 2019, and below the national grade of C.

California's roads received a D. This is a state that collects a gas tax of 68.1 cents per gallon (the nation's highest) plus additional cap-and-trade fees. Despite this, 43% of California's roads are in poor condition according to the report. Bridges scored a C. Drinking water scored a C-minus. Parks scored a D-plus.

Where does the gas tax money go? California's Transportation Commission has struggled to explain that clearly. A portion funds the high-speed rail. A portion goes to cap-and-trade compliance. A portion covers Caltrans administrative overhead. At each step, the connection between the dollar collected and the pothole fixed grows thinner.

Then there is CEQA, the California Environmental Quality Act. A comprehensive study by Jennifer Hernandez at Chapman University found that in 2020 alone, CEQA lawsuits sought to block approximately 48,000 approved housing units statewide, just under half of the state's total housing production. Over the 2019-2021 study period, CEQA lawsuits challenged agency housing plans that would have allowed more than one million new housing units. California has the worst housing-adjusted poverty rate in the United States, and CEQA litigation is one reason it stays that way. The California Center for Jobs and the Economy noted that non-housing infrastructure projects, including transportation and water systems, face the same legal gauntlet.

The $265 Billion Pension Hole

California's state and local pension plans carry combined unfunded liabilities of more than $265 billion, according to the Reason Foundation. CalPERS alone accounts for $168 billion of that gap. CalSTRS, the teachers' pension fund, carries $67.2 billion in unfunded liabilities as of June 30, 2024. That is over $6,000 in pension debt for every person living in the state, including children. The Equable Institute's 2024 rankings placed CalPERS' funded ratio at roughly 72%, meaning it holds 72 cents of assets for every dollar it has promised to pay out, well below the 90% minimum threshold Equable considers "resilient."

CalPERS reported a funded ratio of about 72% at the end of fiscal year 2023, calculated using an assumed rate of return of 6.8%. CalSTRS uses 7.10%. If actual returns fall short, the gap widens and taxpayers cover the difference through higher employer contributions, which means less money for services. The Reason Foundation notes that CalPERS uses a layered amortization approach where new unfunded liabilities are paid off over 20 years, but the total can be stretched to 30 years as a level percent of payroll, pushing costs to future generations.

Every year, a larger share of the state budget goes to servicing pension obligations. These are contractual commitments that cannot be reduced under California law (the "California Rule"). When pension costs rise, they crowd out spending on schools, roads, parks, and every other category. This is not a future problem. It is happening now.

Where Does the Money Actually Go?

Of the $324.7 billion in total spending for FY 2025-26, the largest categories are health and human services ($85.1 billion), K-12 education ($77 billion), and higher education ($22 billion). Roughly $15 billion goes to debt service and pension contributions combined.

Consulting contracts absorb a significant but poorly tracked share. A GovTech investigation found that McKinsey alone held multiple state contracts, including one for DMV modernization. The state's IT modernization efforts have repeatedly involved nine-figure consultant engagements with mixed results.

California employs approximately 234,000 state workers. Compensation and benefits for state employees, including pension contributions and retiree health care, consume roughly $40 billion annually. On a per-capita basis, California has fewer state employees than the national average, but those employees cost significantly more due to higher salaries and pension obligations.

The Comparison That Hurts

Metric	California	Texas	Florida
State budget (total)	$324.7B	$321.3B (biennial: $160.6B/yr)	$116.5B
Population	39.0M	30.5M	22.6M
Top income tax rate	14.4%	0%	0%
Per pupil spending	$16,993	$12,861	$12,491
NAEP 4th grade math (2024)	233	237	240
Infrastructure grade (ASCE)	C-	C	C
Pension unfunded liability	$265B	$64B	$38B
Homeless per 100K	~464	~86	~127

Texas runs a state with a similar total budget, 78% of California's population, zero income tax, and produces better educational outcomes. Florida does the same with a third the budget. Neither state is a utopia. Both have their own failures. But the efficiency gap is not subtle.

Original Contribution: The Cost-Per-Outcome Gap

To quantify the efficiency problem, consider a simple metric: dollars spent per unit of outcome improvement. California spent $24 billion on homelessness over five years and saw the population increase by 30,000. If we naively divide $24 billion by 181,000 (the 2024 count), that is $132,596 per person experiencing homelessness, and the problem still grew. Houston, by comparison, reduced its homeless population by 64% between 2011 and 2022 using a housing-first approach that cost roughly $30,000 to $40,000 per permanent placement. California's per-placement cost for Project Homekey, where data is available, runs closer to $200,000 to $400,000 per unit.

On education, California spends 36% more per pupil than Florida ($16,993 vs. $12,491) but scores 7 points lower on 4th-grade NAEP math (233 vs. 240). If spending linearly predicted outcomes, California should score higher. Instead, the spending premium buys negative returns.

Five Ways AI Could Actually Help

California's efficiency problem is not primarily about bad intentions. It is about a system too complex for any human to track. There are 30+ overlapping homelessness programs, 395,000+ regulatory restrictions, hundreds of billions in annual spending across thousands of line items, and no coherent feedback loop connecting dollars to outcomes. That is exactly the kind of problem AI is built to address. Not with chatbots. With systems that process scale and detect patterns humans cannot.

1. Real-Time Budget Tracking with Anomaly Detection

The federal Digital Accountability and Transparency Act of 2014 (DATA Act) required every federal dollar to be tagged and published to USAspending.gov in a standardized format. A Brookings Institution analysis found that real-time tracking dramatically improved spending visibility and enabled automated auditing that would have been impossible manually. California has no equivalent. Spending data is fragmented across departments, reported annually at best, and formatted inconsistently.

An AI system modeled on the DATA Act framework could ingest every state expenditure in real time, tag each transaction to a program and outcome metric, and flag anomalies automatically: a program whose per-unit costs spike 40% quarter over quarter, a contractor billing at 3x the rate of comparable vendors, a homelessness program whose bed-nights-per-dollar is declining while funding increases. The State Auditor's office currently catches these patterns years after the fact, if at all. An ML-based anomaly detection pipeline could flag them within days.

2. Predictive Outcome Modeling Before Funding Decisions

The UK's What Works Network, a system of 14 independent centers established in 2013, evaluates policy interventions using randomized controlled trials and quasi-experimental methods before the government commits funding at scale. The UCL EPPI-Centre's decade review found this approach saved an estimated £3 billion by killing ineffective programs early and scaling ones that worked.

California does almost none of this. Programs get funded based on political negotiation, not evidence. An AI-assisted modeling system could ingest historical data from California's own programs plus comparable interventions nationally, generate probabilistic outcome projections for proposed funding allocations, and surface the counterfactuals: "Based on 14 years of California homelessness data and 47 comparable programs in other states, this $2 billion allocation has a 23% probability of reducing unsheltered homelessness by more than 5%." That number would be uncomfortable. It would also be more honest than the current approach, which is to fund, hope, and not measure.

3. Automated Cross-Agency Compliance Monitoring

The GAO's 2024 Annual Report on Fragmentation, Overlap, and Duplication identified 112 new actions across federal programs where agencies were doing duplicative or overlapping work. Since 2011, the GAO has identified roughly 2,000 such actions and estimates that addressing them has saved or generated $667 billion. California has no comparable systematic review of its own programs.

The Little Hoover Commission, California's independent state oversight agency, has produced nearly 300 reports since 1962 identifying redundancies, but its recommendations are non-binding. An AI compliance monitoring system could continuously scan the state's regulatory code, cross-reference program mandates across all 200+ state agencies, and automatically flag when two departments are funding the same activity, when reporting requirements conflict, or when a new regulation contradicts an existing one. The Little Hoover Commission's own 2024 report on AI and state government recommended exactly this kind of integration, noting that California's state IT infrastructure is decades behind where it needs to be.

4. Dynamic Resource Allocation (Quarterly, Not Annual)

California's budget cycle is annual. The Governor proposes in January, the Legislature debates for five months, and the budget is signed in June. If a program is failing in October, the money keeps flowing until the next June at the earliest. That is a 20-month feedback loop in a world where data can move in milliseconds.

A dynamic allocation system would set measurable outcome targets for every funded program at the start of each fiscal year. AI models would track progress against those targets monthly using real-time data feeds (bed-nights, test scores, road condition indices, permit processing times). When a program falls below its target trajectory for two consecutive quarters, funding would automatically shift to a pre-approved alternative intervention. This is not new in the private sector. Hedge funds reallocate capital daily based on performance signals. Advertising platforms reallocate budgets hourly. The concept is the same: stop funding what isn't working before the fiscal year ends. The Stanford SIEPR California Policy Research Initiative (CaPRI) has called for exactly this kind of evidence-based dynamic approach to state spending.

5. AI-Assisted Regulatory Impact Assessment

According to the Mercatus Center's RegData project, California's regulatory code contains over 395,000 individual restrictions, the most of any state and roughly 5.5 times more than the median state. The OECD's Regulatory Impact Assessment framework recommends that every new regulation be evaluated for its costs and benefits before adoption. California has a version of this (the Office of Administrative Law reviews regulations), but it is a manual, understaffed process that catches formatting errors more often than economic impacts.

An AI regulatory impact assessment system could, for every proposed regulation, automatically estimate the compliance cost to businesses and individuals by cross-referencing the new text against the existing 395,000 restrictions, flag conflicts with existing rules, model second-order effects (a new housing restriction's impact on construction permits, housing supply, and homelessness), and generate a plain-language summary of what the regulation actually does. The Cali-DOGE project has already begun cataloging the regulatory burden using Mercatus data. An AI system could go further: not just cataloging restrictions but modeling their cumulative effect on the outcomes California claims to care about.

Limitations

This analysis does not account for cost-of-living differences, which are substantial. A state employee in Sacramento costs more than one in Austin or Tallahassee because housing, labor, and construction all cost more. Adjusting for cost of living would narrow the efficiency gap, though it would not close it. California also faces unique demographic challenges: a larger undocumented immigrant population, more non-English-speaking students (California has the highest proportion of English learners in the nation), and a coastline that concentrates homelessness in visible outdoor encampments rather than dispersing it into shelters.

The NAEP comparison is particularly fraught. California's student body is more diverse and has a higher proportion of English learners than Texas or Florida. When adjusted for demographics using the Urban Institute's demographic adjustment model, California's gap narrows. But it does not disappear.

The AI proposals above carry their own risks. Dynamic resource reallocation could be gamed by programs that optimize for measurable outputs rather than actual outcomes (teaching to the test, moving the goalposts on homelessness metrics). Anomaly detection systems produce false positives that could paralyze small agencies with constant investigations. And any system that automates funding decisions inherits the biases of its training data. These are real concerns. But the current approach, spending $325 billion annually with minimal systematic outcome tracking, is not a safer alternative. It is just a more comfortable one.

Strongest Counterargument

The best case against this article is that California is attempting to do something fundamentally harder than Texas or Florida. California provides Medi-Cal to undocumented immigrants. It runs a cap-and-trade system that adds costs but addresses climate change. It has stronger worker protections, tenant protections, and environmental regulations. These cost money, and some of them produce real value that does not show up in NAEP scores or ASCE grades.

A California defender would argue that comparing California to Texas is like comparing Sweden to Singapore: both are functional, but they are optimizing for different things. California is trying to be a generous welfare state. The question is whether it is achieving its own goals efficiently, not whether it matches a fundamentally different governance model. That is a fair argument. But the state auditor's own reports suggest California is failing even by its own standards, spending billions without tracking whether outcomes improve.

The Bottom Line

California is not a poor state. It is the world's fifth-largest economy. It collects more tax revenue than all but a handful of nations. And it runs that money through a system so thick with pension obligations, consultant contracts, and unaudited programs that the connection between dollars in and outcomes out has become almost impossible to trace. A state that spends $325 billion a year should not score below the national average in 4th-grade math. It should not have C-minus infrastructure. It should not spend $24 billion on homelessness and watch the number go up. The money exists. The outcomes do not. Something in the middle is eating it.

The tools to diagnose the problem now exist. Real-time spending visibility, predictive outcome modeling, automated compliance monitoring, dynamic reallocation, and regulatory impact assessment are all technically feasible with current AI systems. The obstacle is not technology. It is institutional inertia. California's government would need to accept that spending decisions should be evaluated against measurable outcomes, and that programs which fail those evaluations should lose funding. That is a political problem, not a technical one. But at $325 billion a year, the cost of not solving it keeps compounding.