CHIPS Act, Promises and Friction

The conversations Gina Raimondo held in her conference room on the fifth floor of the Herbert Hoover Building, on Constitution Avenue across from the White House, settled into a pattern. A chief executive flew in. A team from the company’s Washington office assembled around a long table. Across from them sat the Secretary, accompanied by the small group she had recruited to run the CHIPS Program Office, Mike Schmidt and Sara Meyers and Todd Fisher and a rotating cast of analysts. The visitor opened with a number. The number, in almost every case, was several billion dollars. Raimondo, who had spent two terms as governor of Rhode Island and a career before that in venture capital, listened politely and then looked across the table.

“My conversation with every CEO goes roughly as follows,” she told an audience at the Center for Strategic and International Studies in late February 2024, recounting the script she had been running for nearly a year. “They come in asking for billions of dollars. I look at them across the table and say, you’ll be lucky if you get half of that. And when they come in the next time and learn they’re getting less than half of that, they all tell me, Secretary, I’m not feeling so lucky.”

The line landed as a joke. The substance was not. Raimondo’s office had received applications totaling more than seventy billion dollars from companies seeking grants out of the thirty-nine billion that Congress had appropriated for fab construction under the CHIPS and Science Act. The math required that almost every applicant absorb a discount. The negotiations through which those discounts were imposed, and the additional concessions they secured, were the actual machinery of the largest discretionary industrial-policy program the United States had run in two generations. By the early spring of 2024, with the program’s preliminary award announcements arriving at a pace of roughly one a month, the shape of the bet was beginning to come into view.

The bet had a specific premise, articulated by Raimondo in the same CSIS speech and in variations through 2023 and 2024. By the end of the decade, the United States would manufacture roughly twenty percent of the world’s leading-edge logic chips, up from zero. It would expand mature-node production and build advanced-packaging capability that no longer had to fly wafers across the Pacific. It would do this without nationalizing fabs, without picking technology winners directly, and without spending more than the fifty-two and seven-tenths billion dollars Congress had approved.

The administrative scaffolding was assembled fast. Raimondo had hired Mike Schmidt, a former Treasury official who had stood up the American Rescue Plan’s expanded Child Tax Credit, in September 2022 to run the new CHIPS Program Office. His team grew from a handful of people in a borrowed conference room to roughly two hundred and fifty by the time of the first major award. Their internal mandate was to extract the maximum public benefit per grant dollar, concentrating the appropriations where they would catalyze the most additional private capital rather than distributing them widely.

By the inauguration of 2025, the office’s published scoreboard would read approximately as follows. Direct grants of around thirty-three billion dollars committed. Loan authority of about six billion announced. Roughly twenty final award agreements signed. The induced private capital expenditure, the figure semiconductor industry trade groups began quoting in their year-end summaries, ran above four hundred fifty billion dollars across roughly ninety announced projects in twenty-eight states. The Semiconductor Industry Association, working from its own count of corporate announcements going back to 2020, would put the cumulative pull on private investment near six hundred billion. Schmidt himself, after leaving government in early 2025 and starting a Substack with his former staff, would describe the leverage as roughly ten dollars of private capital per federal dollar committed. No one in Washington had constructed an industrial-policy lever like that in living memory.

The earliest, largest, and politically most fraught award negotiation involved Intel. Pat Gelsinger, who had taken the chief executive’s chair in February 2021 with the explicit pitch that he would rebuild Intel into a foundry to compete with TSMC, had announced in January 2022 that the company would build a twenty-billion-dollar fab complex on a thousand-acre site in Licking County, Ohio, half an hour east of Columbus. He called it the Silicon Heartland. The two fabs there were to begin production in 2025. Gelsinger held the groundbreaking ceremony in September 2022, six weeks after the CHIPS Act became law, with President Biden in attendance. By 2024, the Silicon Heartland’s schedule had begun to slide. The first fab’s start date moved from 2025 to 2026 and then, by February 2025, to 2030. The second fab moved to 2032. Intel’s new construction in Magdeburg, Germany, and in Wrocław, Poland, was paused indefinitely. The Israel project at Kiryat Gat was delayed. A foundry strategy that had projected straight-line growth across four geographies, half a dozen sites, and five process generations in four years had collided with a market in which the company’s external customer base was much smaller than the business plan had assumed.

In March 2024, Raimondo’s office announced a preliminary award to Intel of up to eight and a half billion dollars in direct grants, plus eleven billion in low-interest loans, plus eligibility for a twenty-five percent investment tax credit on as much as a hundred billion dollars of capital outlay across Arizona, New Mexico, Ohio, and Oregon. The announcement was the largest single award in CHIPS history. By November, after Schmidt’s team had completed its due diligence and after Intel’s investment plans had visibly compressed, the grant was finalized at seven point eight six billion. Roughly five hundred million had been removed because of Intel’s slipping schedule and softening demand outlook, and roughly three billion of the original total had been redirected to a separate Department of Defense program called Secure Enclave, which would procure trusted Intel chips for military applications. Two days after the finalization, on December 1, 2024, Intel’s board removed Gelsinger as chief executive. He was replaced by an interim duo, the chief financial officer David Zinsner and the products group lead Michelle Holthaus, while the board began searching for someone willing to inherit a foundry strategy whose architect had just been pushed out. Lip-Bu Tan would arrive in March 2025. By the time he did, the Silicon Heartland’s first wafer date sat eight years out from the day Gelsinger had announced it.

The friction at TSMC’s Arizona project was different in kind. TSMC was building, not rebuilding. It had committed to a twelve-billion-dollar fab in Phoenix in 2020, in the closing months of the first Trump administration, and had broken ground in April 2021. The first-tool-in ceremony on December 6, 2022, was the Biden administration’s marquee photo opportunity on chips: the President arrived alongside Tim Cook of Apple, who had committed his company to becoming the new fab’s largest customer, alongside Jensen Huang of Nvidia, Sanjay Mehrotra of Micron, and an eighty-nine-year-old Morris Chang flown in from Taipei. TSMC announced at the same event that it was upgrading the Arizona project to two fabs and lifting the planned investment to forty billion. Within two years the figure would be sixty-five billion across three fabs; within three years, after the Trump return, one hundred sixty-five billion.

Underneath the announcements, the construction was hard. By the spring of 2023, TSMC’s Arizona team had begun to file paperwork seeking to bring in roughly five hundred Taiwanese workers on EB-2 visas, technicians and equipment installers whom the company described as essential to bringing leading-edge process tools online. The argument from TSMC was technical: nobody in Arizona, or in the broader American labor market, had ever installed a five-nanometer chemical-vapor-deposition tool in a production environment. Local Arizona Pipe Trades Union Local 469, the trade union that covered roughly three thousand pipefitters, plumbers, and welders, did not accept the framing. Its business manager, Aaron Butler, who also chaired the Arizona Building and Construction Trades Council, called the visa request an attempt to substitute foreign labor at lower wages for qualified Americans. In late July 2023 the union filed a petition titled “Block TSMC Foreign Worker Visas.” Local 469 took the position, in a series of statements that summer and fall, that TSMC was blaming American workers for management problems of its own creation: shifting safety protocols, inconsistent supervision on the construction site, and a Taiwanese supervisory style that the union’s members reported as adversarial. Workers told reporters from The American Prospect, Rest of World, and Bloomberg that injuries were unusually frequent on the site, that subcontractors had reported pressure to compress schedules below safe limits, and that turnover was running high.

Governor Katie Hobbs visited the construction site in August 2023 and announced a worker-safety partnership that put the Arizona Division of Occupational Safety and Health on the project with expanded oversight. Through the autumn, TSMC and the Arizona Building and Construction Trades Council negotiated a memorandum that, by December 2023, included additional training programs for American hires, joint oversight on safety, and a more limited footprint for the foreign technicians. Roughly five hundred Taiwanese employees did come, working under H-1B and EB-2 status, but on a more time-limited basis than the original request had implied. The dispute did not end with anyone declaring victory. It ended with a project that ran several months behind schedule and that produced, when its first fab started up, a pool of American technicians who had now spent two years in the company of TSMC’s process culture, for better or worse.

The cost story was more revealing than the labor story. The first Arizona fab’s construction budget grew from twelve billion to something in the range of twenty-five billion across 2022 and 2023. C.C. Wei, who had succeeded Mark Liu as TSMC chairman, told an audience at National Taiwan University in early 2025 that each phase of construction in Arizona had taken at least twice as long as the equivalent stage in Taiwan. He named permitting as the central reason. Inside TSMC, the comparisons ran further: line workers in Phoenix earned four to five times Hsinchu wages; total fab construction was on the order of fifty percent more expensive than in Taiwan; tool commissioning ran slower because the local supplier ecosystem was thinner. Morris Chang, who had been calling the project an “exercise in futility” since 2021, gave an interview in December 2022 in which he called himself “extremely naive” for ever expecting American costs to converge with Taiwan’s.

And yet the first fab worked. Trial production of TSMC’s N4P process began in late 2024. By early 2025, TSMC’s U.S. division president Rick Cassidy was telling industry analysts that the Phoenix yields on a four-nanometer test vehicle were running roughly four percentage points higher than at the equivalent Taiwan facility, in the low nineties. The chips going out the door were Apple’s A16 application processors, fabricated for the first time on American soil and packaged in Taiwan and Malaysia. By the end of 2025, Apple had committed to purchase more than a hundred million Phoenix-built chips per year. Whatever else the Arizona project was, it was no longer an open question whether it could fabricate at the leading edge. It could.

The TSMC final award negotiation produced its own theater. Preliminary terms in April 2024 set the direct grant at up to six and six-tenths billion dollars and offered as much as five billion in loans, against the now sixty-five-billion-dollar Arizona commitment. The third Phoenix fab was added at the same time, with planned production at the two-nanometer node beginning in 2028. The final agreement was signed on November 15, 2024, twelve days after Trump’s election victory and weeks before he would take office. Raimondo and Lael Brainard, the National Economic Council director, made no secret of their motive. Locking the binding contract before January 20, 2025, was the surest available method of ensuring that the funds and the milestone schedule survived a change in administration. The Bloomberg and Reuters wires both quoted unnamed officials describing the work that fall as “Trump-proofing.” TSMC’s award was the first major final agreement under the program. The remainder of the marquee deals followed in a compressed cascade: Intel’s seven point eight six billion finalized November 26, Micron’s six point one six five billion finalized December 10, Samsung’s renegotiated four point seven four five billion finalized December 20, Texas Instruments’ one point six one billion the same day, GlobalFoundries’ one and a half billion finalized in November, GlobalWafers’ four hundred six million in December.

Each of those finalizations was a real-time scrub of the preliminary terms against actual due diligence, and the gaps between proposal and agreement said something about how the Commerce Department was pricing risk. Intel’s grant fell. TSMC’s held but was reorganized to push specific milestones onto contractually defined dates. Micron, which had announced a twenty-year commitment of as much as a hundred billion dollars in New York and twenty-five billion in Idaho, secured the headline six-billion grant on a path to its first New York fab in Clay coming online by 2028. Samsung’s award fell from six point four to four point seven four five billion, a reflection of the Taylor, Texas, project’s halting trajectory. The Taylor fab, which Samsung had announced in 2021 with seventeen billion dollars and projected to begin operations in 2024, had by mid-2024 become a forty-four-billion-dollar facility delayed to 2026. The most-cited reason was structural rather than technical: Samsung’s foundry business had not signed enough customers in the four-nanometer and two-nanometer process generations to fill the line. Reports through 2025 described the company as halting commissioning of EUV scanners that had already arrived and reconsidering whether to reformat the fab around its newer process before any high-volume external customer existed.

Micron’s New York project carried a different political weight. The decision to put a hundred-billion-dollar memory complex in Clay, on a site north of Syracuse, had been the product of a multi-year campaign by Senate Majority Leader Chuck Schumer to direct CHIPS spending into upstate New York. Schumer’s argument to Sanjay Mehrotra and to Micron’s board through 2021 and 2022 had been that the region offered cheap power, abundant water, and a workforce being trained at SUNY’s polytechnic institutes. Schumer, who had made the CHIPS Act his signature legislative achievement and who had partnered with Republican Senator Todd Young of Indiana to assemble a bipartisan coalition through the bill’s torturous two-year passage, treated the Micron site as the proof case for whether industrial policy could fall outside the traditional Sun Belt. The April 2024 preliminary announcement was made jointly with Schumer present. The December finalization, which Mehrotra called “the largest private investment in New York’s history,” locked in a project whose first wafers were targeted for 2028 and whose full buildout would extend into the 2040s. By the time of Trump’s inauguration, Schumer was telling reporters that he believed the act would survive a Republican Congress because, in his framing, “the chips don’t care which party is in power.”

The political turbulence was not hypothetical. Through the 2024 campaign, Donald Trump had described the CHIPS Act in a series of podcast appearances and rally speeches as a giveaway to foreign companies and as evidence of how Washington had managed to lose the chip industry to Taiwan in the first place. On Joe Rogan’s podcast in late October 2024, he said: “We gradually lost the chip business and now it’s almost exclusively in Taiwan. They stole it from us.” On the same circuit he proposed that tariffs on imported chips would have produced the same domestic investment without any subsidy outlay. Speaker Mike Johnson, at a campaign event in Syracuse in October 2024, told a reporter that Republicans would “probably” try to repeal the act if they won; he walked the comment back within forty-eight hours after the local outcry made plain that Micron’s fab was now a vote-relevant fact in central New York. By March 4, 2025, in his first joint address to a Republican-controlled Congress, Trump called the CHIPS Act a “horrible, horrible thing” and asked Johnson, on camera, to repeal it. Johnson stood and clapped. Other Republicans did not move. Aides to Senate Republicans told NBC News in the days afterward that there was no appetite for a repeal vote, both because the awards had already been finalized and because the projects were now jobs in real Republican congressional districts.

Howard Lutnick, who replaced Raimondo at Commerce in February 2025, took a narrower approach. Rather than seeking a legislative repeal of awards already in contract, his department began renegotiating individual agreements where the underlying milestones had slipped. By June 2025, in testimony before the House Appropriations Committee, Lutnick said that Commerce had been able to push the federal share of project capital expenditure down toward four percent or less, from the roughly ten percent target his predecessor had set. He described the original deals as “overly generous.” The renegotiations, in the cases that became public, included revised milestone schedules, expanded clawback provisions tied to delivery dates, and in at least one case the conversion of grant funding into the federal government’s first equity stake in a CHIPS recipient: in late August 2025, Trump’s administration announced that the Department of Commerce would convert a portion of Intel’s award into a roughly ten-percent equity position in the company. Whether that arrangement was an act of nationalization, of leverage, or of political theater was a question on which honest analysts disagreed. What was not disputed was that the original framework, the one Raimondo and Schmidt had built around grants and tax credits without ownership, had been reshaped within seven months of leaving the building.

The 25-percent advanced manufacturing investment tax credit, codified as Section 48D of the Internal Revenue Code, drew less attention but moved more money. By the time Treasury issued final regulations in October 2024, the credit was projected to flow more than a hundred billion dollars in tax expenditure to fab construction over a decade, on top of the appropriated grants. Intel, TSMC, Samsung, Micron, and GlobalFoundries each structured their American capital plans to maximize qualifying spend before the Section 48D construction-start cutoff at the end of 2026. The credit’s relative permanence, embedded in the tax code rather than dependent on annual appropriation, made it the part of the program least vulnerable to political reversal.

By the spring of 2026, the honest scoreboard read like this. The first Phoenix fab was producing leading-edge silicon at competitive yields. The second and third Phoenix fabs were on accelerated schedules, with TSMC’s commitment up to one hundred sixty-five billion. Micron’s first New York fab was under construction with a 2028 first-wafer target. Samsung’s Taylor project was running roughly two years late and looking for customers. Intel was reorganizing under a new chief executive, with Ohio targeted for 2030 and the foundry business now under a federal equity stake. Induced private capex was running well above any 2022 forecast. The twenty-percent leading-edge target was neither obviously out of reach nor obviously in reach. The political coalition that had passed the act, fragile from the start, had survived a change of administration with its core appropriations intact. The mid-game scoreboard was a project plan written in pencil.

Raimondo, after leaving Commerce, told an interviewer in mid-2025 that she counted the program a success not because every award would deliver on its preliminary terms, but because the direction of capital had shifted. The semiconductor industry’s twenty-year drift toward concentration in Taiwan and Korea had stopped. Whether it had reversed was a question no honest observer could answer before the early 2030s. The work of those years would be done by line workers in Phoenix and Clay and Taylor, and by the tool installers rotating through American sites from Hsinchu and Pyeongtaek and Yokkaichi, and by the slower buildup of a domestic supplier ecosystem the grant program had only begun to seed. Industrial policy at this scale, Raimondo liked to say, was not measured in any one fiscal year. It was measured in fab generations. The first generation was coming online. The next ones would arrive on schedules nobody, in the spring of 2026, could yet honestly forecast.