Thirty-five. For 1.2 million people.
I've built and analyzed federal disclosure records across four programs: H-1B visas for specialty occupation workers, H-2A visas for temporary agricultural workers, H-2B visas for non-agricultural seasonal workers, and the PERM labor certification process that is supposed to verify that permanent job openings cannot be filled by Americans. What I found is not four programs with different problems. It is one system with four faces.
The same structural failures run through every program simultaneously: employers certify the facts that only they control, staffing intermediaries extract fees from workers who can't leave, enforcement has ceased, and the path to permanent status is designed to take longer than a career. These are not bugs. They are features of a regime that was sold to Congress as a temporary bridge and became a permanent architecture.
And the domestic workforce we were supposed to train while the bridge operated? We never built it.
The People Inside These Programs
Roughly 600,000 people are living in the United States on H-1B visas right now. That is roughly one in every ten computer professionals working in this country, each of them tied to a single employer for their continued legal right to be here.
Add to that the workers who arrive each year through the agricultural and seasonal programs. In fiscal year 2024, approximately 385,000 people came under H-2A visas to work in fields, orchards, and packing houses. Another 192,000 arrived on H-2B visas for landscaping, hospitality, construction, and food processing. These workers are seasonal. They come, they work, they go home. But while they are here, they are as bound to their sponsor as anyone on H-1B.
These are not exotic skills the country cannot produce. They are farmworkers, groundskeepers, housekeepers, construction laborers, and restaurant cooks. Jobs communities across this country used to fill domestically, before these programs grew large enough to function as a substitute for a labor market that would otherwise have to compete on wages.
PERM is the gateway to a permanent green card through employment. Two-thirds of the people filing through this system are already here, most of them on H-1B visas, already working, already paying taxes, already embedded in American life. The PERM process is not about importing someone new. It is about whether the person already sitting at the desk gets to stay.
For tens of thousands of them each year, DOL certifies that no Americans are available for the position while the employer is reporting, in the same application, that it has conducted layoffs in that same job category within the past six months.
The checkbox exists. It asks the question. It does not change the outcome.
Microsoft, Tata Consultancy Services, Infosys, Capgemini, Oracle: the dominant users of these programs filed thousands of such certifications, year after year, simultaneously telling DOL that no Americans were available and their own HR systems that workers in those same categories had recently been let go. They are the largest players in the system, and the system certified their paperwork anyway.
Tens of thousands of PERM labor certifications were approved in our dataset despite the employer checking a box confirming it had conducted layoffs in that same job category within the past six months. Microsoft alone: over 9,500 such certifications.
The Architecture of Self-Certification
Every one of these programs runs on the same foundational mechanic: employers certify facts that only they control, submit those certifications to DOL, and DOL approves before verifying anything.
In H-1B, employers pick their own prevailing wage level from a four-tier menu. The lowest tier is the 17th percentile of the occupation's wage distribution, meaning 83 percent of Americans doing the same job earn more than a worker at that level is legally required to be paid. Nearly half of all H-1B positions certified in fiscal year 2024 were filed at the bottom two tiers, below the median market wage for the role. Nobody audits whether the job matches the level selected.
Goldman Sachs certified 11,851 H-1B positions in Utah between 2023 and 2025, 45 percent of every H-1B application filed in the state during that window. But applications are not workers. According to USCIS petition data, Goldman employs somewhere between 250 and 300 actual H-1B workers in Utah at any given time. Every worker Goldman does sponsor is paid against wage rates Goldman certified years in advance. Financial Risk Specialists at the bottom of that range were listed at $66,000 to $68,000 per year. The Bureau of Labor Statistics puts the median for that role at $88,000 to $110,000. Goldman Sachs, a company that cleared more than $14 billion in profit in 2024, told the federal government it intended to pay those wages. We don't have pay stubs. We have what Goldman put in writing.
In H-2A, employers write their own experience requirements. Cherryfield Foods, a blueberry operation in Washington County, Maine, required 55 months of experience for a farmworker position in fiscal year 2018, the highest such requirement for any fruit or farm harvesting job in the national H-2A database that year. Every other Maine employer doing comparable work required one to three months. The national ceiling for blueberry work is six months. Cherryfield dropped to three to six months in every subsequent year. DOL certified the 55-month application without challenge.
In H-2B, Montage Big Sky in Montana certified a hotel cook at 48 months of required experience in fiscal year 2022. Montage's own property in South Carolina listed the same job the same year at three months. Same company. Same position. Same luxury tier. DOL certified both.
The Body Shop Runs in Every Program
Each of these programs has produced its own version of the staffing intermediary: a firm that employs workers on paper, places them at client businesses for a fee, pays the worker less than that fee, and controls their continued legal presence in the country.
In H-1B, Cloudberg Tec Limited Company is incorporated in Salt Lake City. Between 2023 and 2025, it filed 99 certified Labor Condition Applications listing worksites in 23 cities across 12 states. These workers were not in Utah. They were contracted to clients nationwide, with Cloudberg as the visa sponsor controlling whether status continues. Dozens more operate the same way out of Utah addresses alone.
In H-2A, agricultural labor contractors account for roughly one-third of the entire program. Overlook Harvesting Company, based in Winter Haven, Florida, has certified more than 18,000 workers under 59 different entity names encoding the client corporation and the state: Bayer, Corteva, Pioneer, in Michigan, Illinois, Indiana, Ohio. One operation, invisible from any vantage point that evaluates employers one name at a time.
The leverage is identical across all three programs. The worker's continued legal presence depends on maintaining status with the sponsoring employer. The worker who can be sent home in 60 days does not push back on wages or working conditions the way a worker who can walk across the street to a competitor does. This is not an unintended consequence. It is the structure.
Enforcement Has Stopped
In 2017, the Wage and Hour Division opened 101 H-1B enforcement cases, 377 H-2A cases, and 108 H-2B cases. Combined backwage recovery: $11.2 million.
In 2025, it opened 6 H-1B cases, 28 H-2A cases, and 1 H-2B case. Combined recovery: $349,000.
That is a 94 percent decline in case volume and a 97 percent decline in dollars recovered, across all three programs simultaneously, while the worker count grew by hundreds of thousands.
One enforcement case opened for every 34,000 workers in 2025. The probability that any employer committing a violation faces an enforcement action is, for practical purposes, zero. PERM has no enforcement track at all.
The Trap That Holds the System Together
Two-thirds of PERM filers are already here, most working on H-1B, sitting at desks in American offices, raising children in American schools, paying American mortgages, and waiting. More than half of all permanent labor certifications go to workers from India. The country-specific quota gives India the same annual green card allotment as Iceland, regardless of how many people are in line. For Indian nationals filing today, immigration attorneys calculate the expected wait at 134 to 151 years.
Not a typo. One hundred and thirty-four years.
A person already living here, already working, already woven into American life, who files for permanent residency today, follows every rule, and never makes a mistake would, with near certainty, die before receiving the green card. For the entire duration of their career, they must maintain H-1B status through a sponsoring employer. The employer knows this. The wage suppression in the data reflects it. The worker certified at $66,000 accepts $66,000. The physical therapist with a doctorate placed at $20 an hour accepts $20 an hour. Not because they couldn't command more on the open market. Because the alternative is losing their home, their children's schools, and everything they have built here.
H-2A workers face a different trap but the same structure. Most have no path to permanent residence at all. The sponsor controls whether they come back next year. The worker who raises issues may not be invited back. Blaine Larsen Farms, a grain operation in Dallam County, Texas, escalated from zero emergency H-2A filings in fiscal year 2018 to 100 percent emergency filings by fiscal year 2023, for jobs that start February 1 every year. Nothing about an annual, predictable grain operation in the Texas Panhandle is an emergency. The designation is self-selected and nothing required Larsen to explain the escalation. The same year, the Wage and Hour Division opened 28 H-2A enforcement cases nationally. Not 28 for Larsen. Twenty-eight for the entire country.
The Case for Stopping, and What Comes Before Reopening
The reform conversation always starts in the wrong place. It asks what adjustments would make these programs work better. A higher wage floor. An extra audit step. A new rule on layoff disclosures. That framing assumes the programs are basically functional. They are not.
These are not programs with enforcement problems. They are programs with no enforcement. The infrastructure does not exist to protect 1.2 million working people with 35 cases a year. You would need enforcement to scale by roughly 4,000 percent just to reach the point where a violation carried a meaningful probability of detection.
The self-certification architecture was not designed to be audited. Adding auditors to a self-certification system is not reform. It is performance. Real reform requires changing the underlying design: wages set by independent determinations, not employer selection from a four-level menu. Experience requirements reviewed against actual job market data before certification. Recruitment verified by a third party, not attested by the employer who benefits from the outcome.
The green card queue cannot be fixed while hundreds of thousands of workers continue entering it every year under a country-specific cap calibrated for 1990. The queue is the trap. The cap is why the queue exists.
The wage floors were not designed to set market wages. They were designed to pass a legal test. The floor is not the wage. It is the ceiling that ensures wages never have to compete.
But the argument the data forces most directly goes further than fixing the program mechanics. We should not reopen any of these programs until we have first built the domestic training infrastructure that was supposed to make them unnecessary.
Every one of these programs was sold to Congress as a temporary bridge. H-1B was a bridge while the STEM pipeline caught up. H-2A was a bridge while agricultural wages normalized and domestic recruitment improved. H-2B was a bridge for genuine seasonal peaks. The bridge was never supposed to become the road. The domestic workforce development system was supposed to be built.
It was never built.
H-1B employers pay a mandatory training fee under the American Competitiveness and Workforce Improvement Act, between $750 and $1,500 per petition. The federal government collects roughly $250 million a year from these fees. Their stated purpose is to fund training programs for American workers in exactly the fields where H-1B is most heavily used. At $5,000 per participant, the cost of an average registered apprenticeship in technology or advanced manufacturing, that is 50,000 Americans per year who could be receiving paid, hands-on training in software development, data systems, and engineering. The Federal Reserve Bank of New York found that 6.1 percent of recent computer science graduates were unemployed and 16.5 percent were underemployed, working jobs that didn't require their degree, at the same moment those companies were filing hundreds of thousands of H-1B positions.
We are collecting the fees. We are not building the pipeline. We are using the fees as a nominal cost of admission to a captive labor system and telling the graduates it just didn't work out.
The agricultural side doesn't even have a training fee. The H-2A program has grown eight times over twenty years with no corresponding investment in domestic agricultural housing, no seasonal income support for American workers, no rural recruitment infrastructure. Employers certify that no Americans are available. That certification is technically defensible because the wage is set at the legal floor, the housing assumes workers arrive in organized cohorts from abroad, and the recruitment effort, in documented cases in our data, amounts to an ad that runs over Christmas and New Year's.
You cannot honestly certify that no Americans are available for jobs you have never invested in making available to them.
The honest goal is not to pause these programs and reopen them with better rules. It is to make them obsolete. A country that built the workforce it needed would not require hundreds of thousands of temporary agricultural workers to keep food on the table. It would not need to import physical therapists and software engineers because it chose, for decades, not to train its own. These programs exist because we built a labor market with holes in it and decided filling those holes with captive foreign labor was easier than filling them with investment, wages, and training. That is a choice. It remains a choice. And the longer these programs run, the harder it becomes to make a different one, because the employers who benefit have every incentive to ensure the domestic alternative is never built.
Stop. Not as a pause before the next version of the same system. As a reckoning with the question we have avoided for thirty years: what would it take to build a workforce development system that renders these programs unnecessary? That question cannot be answered while the programs run at full volume and the industry lobby argues that any disruption to the pipeline is a crisis. The political urgency to build the alternative disappears the moment the visa keeps the shortage from being felt. That is why the pipeline was never built. That is why it will not be built unless we stop.
What I Know for Sure
These programs do not function as emergency valves for genuine labor shortages. They function as a permanent architecture for below-market, captive labor across white-collar technology, agriculture, landscaping, hospitality, and construction simultaneously.
The workers inside these systems did not design them. They are being harmed by them. The American workers who never got a fair shot at those jobs are being harmed too. The computer science graduate working retail because the entry-level software job went to someone who had no choice but to accept less. The farmworker who applied and was turned away. The landscape crew member who watched wages stay flat for a decade because the AEWR made sure they had to.
The system is broken in exactly the same way across all four programs because the architecture is identical. Self-certification. No enforcement. Visa dependency that makes workers impossible to replace with people who can actually negotiate. A domestic training pipeline promised at the outset and never built, because the programs made building it unnecessary.
Stop. Build the infrastructure. Then ask whether we need these programs at all, or whether we finally built the thing that makes them unnecessary.
That is where the data brings me. All of it is visible in public records. All of it is findable. Most of it has never been assembled across programs and looked at as a single system.
Now it has been.