"Only about 37% of those are getting the visas, so the remainder of that, which is 63%, they have to figure out what to do with these people."
That is an immigration attorney at Fragomen describing what happens at one major semiconductor employer to the foreign engineers in their final year of H-1B eligibility. Not marginal hires. Last-ditch cases, already trained, already working, and the company loses roughly two-thirds of them because a lottery said no.
Now hold that number against the other side of American industrial policy.
You can subsidize a fab into existence. You cannot subsidize it into producing chips with an empty clean room.
TSMC has laid out plans to invest as much as $265 billion across a dozen chipmaking and packaging facilities in Arizona. Micron is targeting $100 billion on memory production in New York. Samsung is building logic in Texas. Even Intel's delayed $28 billion Ohio project is slated for shortages the moment production ramps.
The concrete is being poured. The staffing plan is missing.
A joint analysis by McKinsey, the industry group SEMI, and the National Science Foundation puts the skilled-labor shortfall at as many as 157,000 full-time workers by 2030. Nearly three-quarters of chip employers already report significant difficulty hiring engineers.
America is building fabs it has no plan to staff.
The shortage is not spread evenly across the economy. By 2030, roughly 74 percent of the industry's unfilled roles sit in manufacturing and 60 percent in engineering. And the engineering hole is the structural one.
Here is why it won't close on its own: only about 3 percent of US engineering graduates go into the chip industry at all. The rest chase software and AI, where the money is better. As McKinsey's Taylor Roundtree put it, "There's just not enough talent to go around."
So the US built two policy engines and pointed them at each other.
Engine one is the CHIPS Act: subsidies that conjure demand for tens of thousands of high-skilled workers on a fixed construction timeline. Per CSET, engineering employment in the sector could grow by 19 percent over the next decade, with fabs coming online long before any domestic training pipeline can mature.
Engine two is immigration policy, which throttles the only supply that can arrive on that timeline. Per Marketplace, the H-1B cap sits at 85,000 for the private sector—barely changed from the 65,000 set in 1990—while applications have blown past 400,000. In 2014 roughly 70 percent of applications were approved. For fiscal 2025, the grant rate was under 20 percent.
If an American chip firm identifies a must-have engineer, the H-1B is now more likely to deny that engineer than grant them.
The standard objection to just issuing more visas is displacement: foreign workers take jobs Americans would otherwise hold. It sounds intuitive. It is also, on the best available evidence, wrong.
The cleanest test we have comes from the H-2B visa lottery, which the Peterson Institute studied precisely because it randomizes who gets foreign workers. Random assignment is the closest thing labor economics gets to a controlled experiment—it strips out the usual confound that thriving firms both hire more foreigners and more natives.
In 2021 and 2022, per the Peterson Institute, firms allowed to hire more foreign laborers through the expanded H-2B lottery saw revenue rise 12 to 14 percent. And across all firms, randomly getting the chance to hire foreign workers did not reduce US hiring.
The winning firms increased production, investment, and profits. Restrictions on H-2B hiring, PIIE found, either did not affect or actually reduced employment of low-skill American workers.
Read the mechanism carefully, because it inverts the entire displacement story. Foreign labor is not a substitute slotted into a fixed number of chairs. It is a complement that unlocks expansion the firm otherwise couldn't attempt.
A firm that can staff a new line builds the new line. Building the line requires supervisors, logistics, maintenance, sales, administration—adjacent roles that get filled disproportionately by natives. Deny the foreign worker and you don't reallocate that job to an American. You cancel the expansion, and the American-held adjacent jobs never get created.
The H-2B evidence sits in low-skill sectors—construction, care work, food production. But the logic scales up cleanly to fabs, where the missing input is the experienced engineer who can transfer tacit process knowledge that no classroom conveys.
CSET reportedly makes exactly this point about firms like TSMC: leading-edge manufacturers transfer experienced employees to new US fabs specifically to train and mentor American coworkers on the job. The foreign engineer is not displacing the American hire. The foreign engineer is the reason the American hire has a fab to work in and someone to learn from.
Which makes the contradiction plain. The US is subsidizing the fab and starving the staffing. It pays firms to build capacity and then denies them the one input that lets the capacity produce anything.
Self-defeating industrial policy has a specific failure mode here, and it is worse than delay.
The marginal engineer denied a green card does not vanish. Consider Harshad Surdi, the Intel engineer who spent nearly a decade in the US and described his last H-1B attempt as blood, sweat, and tears hanging in the balance of a lottery. He got lucky this cycle. Many don't, and Surdi has watched colleagues give up and leave for Europe and Canada.
The green-card math makes the leaving rational. Per-country caps mean Indian and Chinese nationals can each receive no more than 7 percent of employment-based green cards a year. CNAS notes that some categories of Indian and Chinese applicants filing today will likely die still in line.
Rivals are catching the talent America trains and rejects. China has reportedly stepped up recruitment of chip talent trained abroad, offering bonuses of half a million dollars or more.
And the capacity is migrating with the people.
On July 5, Prime Minister Modi inaugurated CG Semi's outsourced assembly and test facility in Sanand, Gujarat—foundation stone laid in 2024, chip testing commencing by August 2025, more than 5,000 direct and indirect jobs. Modi framed it as the cluster logic America is fumbling: "one industry gives birth to ten more and ultimately transforms the entire regional economy."
That is precisely the adjacent-jobs multiplier the PIIE evidence describes—the expansion that foreign labor unlocks. India is absorbing it, happily hiring the engineers the US trains, denies, and exports.
The reshoring bet was never really about concrete or subsidy dollars. Those are the easy parts, and the US is executing them at nine- and ten-figure scale.
The bet was about people—157,000 of them by 2030, concentrated in roles only 3 percent of American grads want, on a timeline no domestic pipeline can meet.
The immigration system that could bridge the gap is instead rejecting 63 percent of a chipmaker's last-chance engineers and handing them to Portland's ski hills, Canada, and eventually to firms in Shenzhen and Sanand.
You can subsidize a fab into existence. You cannot subsidize it into producing chips with an empty clean room.