Who pays for SNAP now?

The Congressional Budget Office puts the law's SNAP cuts at about $187 billion through 2034. That does not mean $187 billion out of a single year. It means the program is set to spend roughly a fifth less than it was projected to over the next ten years. To see how big that is, look at what SNAP has actually cost.

Supporters of the law read the cut as a correction for real growth, and the growth is real: in today's dollars, SNAP spending roughly quadrupled from 2000 to 2023, with most of that growth coming during recessions and the pandemic. But that one number hides how the cuts work. The law does more than shrink the budget. It makes four separate changes, each starting at a different time and paid by someone different.

The four big changes get the headlines. The same law makes quieter ones that add up. Fewer immigrants qualify: the law makes refugees, asylees, and others admitted legally on humanitarian grounds ineligible, even those who work (in North Carolina, starting February 1, 2026). The benefit can no longer outpace inflation: the 2021 raise to the Thrifty Food Plan was the first real increase in 45 years, and the law now caps future ones at inflation, which slowly erodes what a benefit buys. And two deductions shrank, trimming benefits by about $100 a month for many households. The basic income and asset limits did not change. What changed is who counts as eligible and how much the formula gives them: fewer people qualify, and those who do get a little less.

Start with the strangest part. A number that once just tracked how accurately a state paid benefits is now a bill it has to pay.

Every year, the USDA measures each state's payment error rate: how often it paid a household the wrong amount, too much or too little. This is not fraud. It is bookkeeping: a wrong figure keyed in, an income update that came late, or a typo. These errors pile up fastest in the hardest cases, like families whose income changes every month. Even the USDA does not call it fraud.

Fraud is a separate, smaller problem. The government's own studies measure it at about a penny or two on the dollar, roughly $1 billion a year. A 2025 USDA review flagged up to $3 billion more in records with duplicate Social Security numbers, or numbers belonging to people who had died, but its own report calls those "possible issues rather than confirmed fraud," not proven cases. Either way, fraud is not what the error rate counts. The bill is pegged to the bookkeeping.

And the rate is an estimate, not an exact count. The state checks only a sample, about 1,150 North Carolina cases a year, so the USDA publishes the rate with a margin of roughly two points either way. The new penalty ignores that margin and treats the number as exact. And the brackets sit so close together, at 6, 8, and 10 percent, that sampling luck alone can tip a state from owing nothing to owing tens of millions.

And these are not abstractions on a state ledger: a family paid too much has to pay it back, even when the state's own mistake caused it, and a family paid too little is owed the difference. Case by case, the score is real money moving to and from real households.

For sixty years, that score was just a management number, something the state tracked. The new law turns it into a bill. Starting in October 2027, a state with an error rate under 6 percent pays nothing toward benefits. From 6 to 8 percent, the state pays 5 percent of its food bill. From 8 to 10, it pays 10 percent. At 10 or above, 15 percent.

On June 24, the USDA released the 2025 scores. The national rate was 10.62 percent, about $10.1 billion in wrong payments. North Carolina's was 7.36 percent, and most of the errors were payments to families that were higher than they should have been rather than lower. Either way it is error, not fraud.

That score now has a price. North Carolina hands out about $2.8 billion in SNAP benefits a year. At last year's error rate of 10.21 percent, the state would owe 15 percent of that, as much as $420 million a year. That is a bit over one percent of the state's general fund: not the whole budget, but real money that comes out of the same fund where every school and Medicaid dollar is already fought over. At 7.36 percent, the bill drops to the 5 percent tier, about $150 million. One year of cleaner paperwork cut the worst case by about $270 million.

And the bill is not final yet. For the first year, a state can use whichever is lower, its 2025 or its 2026 error rate. If North Carolina's 2026 rate comes in under 6 percent, its share for 2028 is zero. Because the 2029 bill uses the 2026 rate too, that one score would wipe out both years. So the real question is simple: the federal year that ends September 30, 2026 is the one being graded. The accuracy of North Carolina's paperwork in it will decide whether the state owes about $150 million a year for food, or nothing.

And basing millions on one or two years is shaky. A rate pulled from a small sample bounces around, and one that jumps high usually drifts back down the next year on its own. North Carolina's fall from 10.21 to 7.36 percent is part real cleanup, since it closed thousands of old cases, and part that drift. It is not proof the state suddenly got better at paying people right.

Supporters of the law have a fair point here. A state should pay households the right amount, and North Carolina's error rate had been climbing for years. That is real. If you move billions of dollars, you should move it accurately. And North Carolina is not alone: on the 2025 scores, 41 states and the District of Columbia would face a penalty, which supporters read as overdue accountability for states long shielded from the cost while Washington paid more than ninety percent of the tab. Nationwide that first bill comes to about $9 billion, and close to half the states would owe $100 million or more.

But accuracy is not the only way to lower the score, and that is the problem. A state that cannot find $150 million has a cheaper way out. The score counts only the money paid to people who got the wrong amount. It does not count the people a state wrongly turns away. So the surest way to bring the error rate down is not to pay more carefully. It is to make SNAP harder to get and keep. The Center on Budget and Policy Priorities and the Food Research and Action Center both warn the law pushes states that way, toward more paperwork, shorter deadlines, and more frequent renewals that quietly thin the rolls. Illinois and Georgia have already started making some families renew twice as often. No state has said out loud that it is dropping people to lower its bill. It does not have to. The incentive does that on its own.

The error-rate bill is the big one, so it gets the attention. But a second bill comes a year earlier, in October 2026, and it lands closer to home.

Running SNAP costs money: the caseworkers, the computer systems, the phone lines, the fraud checks. Washington used to pay half. Under the new law it pays a quarter. In most states, the state government covers the difference. North Carolina is one of the ten where counties run the program, so that extra 25 percent lands on county taxpayers. State health officials told a legislative committee in January 2026 that this will cost North Carolina's counties about $69 million more a year, starting in October 2026, with about $16 million more falling on the state itself. Mecklenburg County's share alone is estimated at over $8.1 million, the most in the state.

There is a number for Buncombe County, and it comes from the state. When the health department worked out the cost county by county, Buncombe's share of the new cost came to about $2.8 million a year in additional money the county must find, once the change is fully in effect. The county's own 2027 budget puts it a little lower, closer to $2 million of new cost, because the cut starts in October and only part of the budget year takes the hit. Either way it is money the county did not have to spend before. And Buncombe is not alone in the mountains of Western North Carolina: the same shift adds about $760,000 a year in Henderson County, about $410,000 in Haywood, and about $280,000 in Jackson, all of it money Washington used to split.

The state is not making the counties whole: its 2026 budget left them to absorb close to $52 million of it in the first year. In June, a group of state senators asked Congress to delay the change. For now it stands, and it lands in October.

About 29,000 Buncombe residents were on SNAP in spring 2025, getting about $171 a month each. County caseworkers, not the state, read the documents, approve or deny, and enter the data that becomes the error rate. Starting this October, the county pays more for all of that work.

All of this lands somewhere, and here it lands on Buncombe.

If you pay taxes in Buncombe County: you are about to pay more. County taxpayers now cover a bigger share of the cost of running SNAP, about $2.8 million a year. Washington used to split that cost down the middle. And the food bill is now partly the county's too: North Carolina's 2026 budget covers the state's share by withholding a slice of every county's sales-tax revenue. That share runs from nothing up to $420 million a year, about $150 million at today's rate. It is not only a tax question. About 25,000 residents here are on SNAP now, down from about 29,000 in spring 2025. The roughly $50 million a year in benefits they spend goes to local grocers and tailgate markets. As the rolls shrink, that money leaves local tills too.

That is who pays. But a law that moves this much money also decides who eats, and the rolls are already shrinking. If you get SNAP, or you want to know who is losing it and why, that is the other half of this story: who loses SNAP now →

The people who need the help did not change this year. The bill did, and part of it now has Buncombe's name on it.