The Trump II Theater

On the afternoon of Monday, March 3, 2025, in the Roosevelt Room of the White House, an engineer named C.C. Wei stood beside Donald Trump and read a prepared statement off a single sheet of paper. Wei, who nine months earlier had succeeded Mark Liu as chairman as well as chief executive of Taiwan Semiconductor Manufacturing Company, the first executive to hold both titles since Morris Chang, wore a dark suit and a slightly nervous smile. Howard Lutnick, the new commerce secretary, hovered to one side. A row of American flags formed the backdrop. Wei announced that TSMC would invest at least one hundred billion dollars in the United States, on top of the sixty-five billion already committed to Arizona, to build three new fabrication plants, two advanced packaging facilities, and a research and development center. Trump, looming over the podium, cut in to translate the news for the cameras. “This is a tremendous move by the most powerful company in the world,” he said. “Semiconductors are the backbone of the twenty-first century economy. Without semiconductors, there is no economy.” A reporter asked whether the deal had been driven by the threat of tariffs. Trump nodded. The threat, he said, had worked.

Three months earlier, on the Joe Rogan podcast at the end of October 2024, Trump had said the opposite of everything implicit in the Roosevelt Room scene. “Taiwan, they stole our chip business, okay?” he had told Rogan. Taiwan, he said, wanted American protection but did not pay for it. The CHIPS Act, the bipartisan industrial-policy bill Joe Biden had signed in 2022, was a “horrible thing.” All Washington had needed to do was charge tariffs and the chip business would have come back on its own. The clip had run for under two minutes inside a three-hour interview, and almost no one in the chip industry had taken it as policy. By March, with Wei standing next to him and the most powerful chipmaker on earth pledging the largest single foreign direct investment in American history, Trump was treating the same line as vindication. The grievance had produced the deal. The deal had erased the grievance. Both could be true at once. The contradiction was the point.

The Trump II chip story, in the four months between the inauguration and the rescission of the AI Diffusion Rule in May, was the story of an administration discovering that the chip industry was both a foreign-policy lever and a domestic-political prize and not deciding, in any consistent way, which it valued more. Individually, the moves looked like extensions of the first Trump administration’s playbook. Taken together, they fractured the chip war along a fault line the original architects of the Section 301 era had not anticipated. The line did not run between the United States and China. It ran through the United States itself, between the people who wanted to use the chip industry as an instrument of statecraft and the people who wanted to use it as an instrument of patronage and tariff revenue. The second front of the chip war was no longer in Beijing or Hsinchu. It was in Washington.

The signs had been there from the beginning. At the January 20 inauguration, moved indoors because of the cold, the seats directly behind the Trump family went to Elon Musk, Jeff Bezos, Mark Zuckerberg, Sundar Pichai, and Tim Cook. Marco Rubio, sworn in as secretary of state that afternoon, sat further back among the cabinet. The visual was not subtle. The new administration’s center of gravity ran through Silicon Valley, through founders who had decided in the previous twelve months that the second Trump term was a more comfortable fit than the first had been. David Sacks, the former PayPal Mafia investor Trump had named in December as White House AI and Crypto Czar, was already running an informal coordination group with venture capitalists. Sriram Krishnan, the Andreessen Horowitz partner Sacks had recruited, was joining the Office of Science and Technology Policy as senior AI policy adviser. The first Trump administration had treated Silicon Valley with suspicion. The second was building cabinet rooms inside it.

The day after the inauguration, Trump appeared in the East Room with Sam Altman, Larry Ellison, and Masayoshi Son to announce a venture they called Stargate. The headline number was five hundred billion dollars, of which one hundred billion was committed up front. The first data center, Ellison said, was already under construction in Abilene, Texas. SoftBank and OpenAI each held forty percent, Oracle and Abu Dhabi’s MGX each held seven, and the actual financing was opaque. Within hours Musk, who had not been invited, declared on his social network that SoftBank had under ten billion dollars secured and the announcement was vapor. Altman replied that Musk knew better. Whatever the truth of the funding stack, the political theater was unmistakable. The new president, on his second day in office, had presented himself as the impresario of American AI infrastructure, and the country’s largest private capital pool had agreed to play along.

Six days later, on January 27, Trump flew to his golf club in Doral and addressed the annual Republican congressional retreat. The chip line was not rambling. He told the assembled lawmakers that “in the very near future, we are going to be placing tariffs on foreign production of computer chips, semiconductors, and pharmaceuticals.” The tariff on Taiwanese chips, he said, could be twenty-five percent, fifty percent, or a hundred percent. Companies would build in America rather than pay it. The CHIPS Act, he added, had been the wrong way to do this. “We don’t want to give them billions of dollars like this ridiculous program that Biden has.” On the same day, six thousand miles away, the Chinese AI lab DeepSeek released the first of its R1 reasoning models, and Nvidia, in a single trading session, lost roughly six hundred billion dollars in market capitalization. The administration’s first chip-policy day produced two stories at once. One was about whether American export controls were working. The other was about whether American industrial policy was about to be torn down.

The man Trump had nominated to oversee both questions was, by January 27, in the middle of his confirmation hearings. Howard Lutnick, chairman of the bond brokerage Cantor Fitzgerald, had been chosen for Commerce in November. He was sixty-three, best known to the public as the executive whose firm had lost six hundred and fifty-eight employees, including his brother, on September 11. He had spent the years since rebuilding Cantor and running its relief fund. He had never overseen an export-control system or a multibillion-dollar grant program. In his hearing on January 29, when Senator Maria Cantwell pressed him on whether he would honor the multiyear award contracts the previous Commerce Department had signed with Intel, TSMC, Samsung, and Micron, Lutnick said he would “rigorously enforce” the documents already signed. He declined to commit to enforcing the spirit of the law. “I work for the President,” he told the committee.

The Senate confirmed Lutnick on February 18 by 51 to 45, on party lines. Treasury Secretary Scott Bessent had been confirmed earlier, on January 27, 68 to 29. Jamieson Greer, the Lighthizer protégé named U.S. Trade Representative, was confirmed on February 27. The cabinet was now formally in place. Its center was Lutnick, who controlled Commerce and therefore BIS, the CHIPS Program Office, and the Section 232 investigative apparatus all at once. The hawks were Greer at USTR, Rubio at State, and Pottinger’s intellectual heirs scattered across the NSC under Mike Waltz. The libertarians were Sacks, Krishnan, and the venture-capital coalition that had backed the campaign. The industrial-policy continuity was Bessent, who privately favored letting the CHIPS Act run as written and renegotiating only at the margins. The four wings did not agree on what the chip industry was for, and the new president did not require them to.

While Trump’s confirmations were proceeding, the outgoing Biden administration handed its successor a problem on the way out the door. On January 13, 2025, with one week left in office, BIS published an interim final rule titled “Framework for Artificial Intelligence Diffusion.” It was the work of Alan Estevez, Biden’s undersecretary for industry and security, and Tarun Chhabra at the NSC, and it was the most ambitious export-control architecture any administration had attempted. It divided the world into three tiers. Tier 1 contained eighteen close American allies, the Five Eyes plus Japan, Korea, the Netherlands, and a handful of European partners, and faced no caps on advanced AI chip imports. Tier 2 contained most of the rest of the world, including India, Israel, Singapore, Saudi Arabia, and the UAE, and faced country-by-country compute caps. Tier 3 was the embargoed list: China, Russia, Iran, North Korea. The rule was set to take effect on May 15. Almost no one outside the administration liked it. Nvidia called it “sweeping overreach.” Allies in capitals from Riyadh to Warsaw read the tier structure as a unilateral American imposition of where they sat in the international hierarchy of trust.

The Trump administration did not move on the rule for three months. The May 15 effective date crept closer. By April, the chip industry’s Washington lobbyists had concluded the new administration would not enforce the rule but had not yet decided how to replace it. Then on May 13, two days before the framework was scheduled to bite, BIS announced its rescission. The accompanying statement called the rule “overly complex, overly bureaucratic” and likely to “stifle American innovation.” The same day, Trump boarded Air Force One for a four-day trip through Riyadh, Doha, and Abu Dhabi, accompanied by Lutnick, Sacks, Jensen Huang of Nvidia, Lisa Su of AMD, and Sam Altman. By the time the plane landed in Saudi Arabia, Nvidia had announced it would supply Humain, the kingdom’s new state AI venture, with eighteen thousand of its newest Blackwell GB300 systems. AMD signed a separate ten-billion-dollar collaboration with Humain. In the Emirates, Trump announced a deal allowing the UAE to import as many as five hundred thousand of Nvidia’s most advanced AI chips per year through 2027, almost all of them flowing through G42, the Abu Dhabi national champion that two years earlier had been the subject of intense congressional scrutiny over its alleged ties to Chinese intelligence. The diffusion rule had tried to set Gulf chip access inside a global framework keyed to common security standards. Its replacement was a series of bilateral handshakes negotiated personally by the president on a state visit.

Inside the United States, the same fragmentation was reshaping the CHIPS Act. On March 4, the day after the TSMC announcement, Trump delivered his first joint address to Congress and called the CHIPS Act “a horrible, horrible thing.” He demanded that Speaker Mike Johnson find a way to repeal it. Republican applause was conspicuously thinner than for the rest of the speech. Within days, Senate Republicans were briefing reporters off the record that there was no appetite to revisit the law. Ohio Republicans, with Intel’s stalled twenty-billion-dollar New Albany site in their districts, were not going to claw back funds that had been the basis of their own re-election ads. Arizona, Texas, Idaho, and New York members had similar local stakes. The CHIPS Act had become the kind of bipartisan industrial program political scientists had once said America could not produce, because the act of distributing the money had created the coalition that defended it. The president could call it horrible. He could not get it repealed.

What he could do was renegotiate the awards from inside Commerce. Lutnick, in the weeks after his confirmation, made clear he intended to treat the existing CHIPS Act contracts as the floor of a new deal rather than the ceiling of the old one. On March 31, Trump signed an executive order creating a new “U.S. Investment Accelerator” inside Commerce and folding the CHIPS Program Office that Gina Raimondo had built into it. The order directed Lutnick to “renegotiate” agreements toward “improved terms.” Some contracts, like TSMC’s, were rewritten upward; the additional hundred billion dollars came without an additional dollar of American subsidy, an exchange Lutnick held up as the model. Others were left to wither. The career staff that had built the awards architecture, and the NIST workforce that supported them, had been hollowed by parallel DOGE-driven reductions in force; Commerce terminated more than seventy NIST employees on March 3 alone. The technocratic backbone of the original CHIPS implementation was now, deliberately, smaller and more politically pliable.

The tariff theater ran on a parallel track. On April 1, Lutnick formally initiated two Section 232 national-security investigations, one on semiconductors and semiconductor manufacturing equipment, the other on pharmaceuticals. Section 232 was the same authority Trump had used in the first term to impose steel and aluminum tariffs; it gave Commerce up to 270 days to recommend duties, though Lutnick promised the report within two months. The next day, April 2, Trump declared “Liberation Day” in the Rose Garden and unveiled a chart of country-by-country reciprocal tariff rates. Taiwan’s was thirty-two percent. South Korea’s was twenty-five. China’s was thirty-four. The chart excluded raw semiconductors from the reciprocal regime, on the theory that those would be addressed under Section 232, but Taiwan and Korea now stared at duties on the televisions, cars, machinery, and finished electronics that surrounded the chips. President Lai Ching-te promised in a televised address to negotiate. By April 9, with bond yields spiking and equity markets in retreat, Trump announced a ninety-day pause on the country-specific rates, leaving a ten percent baseline. Markets posted their largest one-day gain since 2008 within minutes. The president told reporters he had paused because people were “yippy.”

In the same April week, Lutnick’s office told Nvidia that exports of the H20, the Hopper-derived inference chip Nvidia had designed specifically to comply with the October 2023 thresholds, would now require an export license. Nvidia disclosed on April 15 that it would take a five and a half billion dollar charge against inventory and purchase commitments. Jensen Huang, in meetings in Washington and Beijing, argued that the ban was self-defeating, that it would hand the Chinese market to Huawei and accelerate exactly the indigenization the controls were meant to prevent. By July, after a meeting with Trump, the policy had reversed. Nvidia and AMD were told they could resume H20 and MI308 sales to China in exchange for a fifteen percent revenue share with the U.S. government, an arrangement of doubtful legal grounding that Lutnick defended as a “fee” rather than an export tax. Huang, who had spent years calling Biden’s controls “a strategic misfire,” now described Trump’s approach as “visionary” and “bold.” The export-control consensus that had held since October 2022 was, by midsummer, disintegrating in real time.

The fragmentation showed up most starkly at Intel. Pat Gelsinger, who had bet the company on a foundry pivot and on the CHIPS Act, had been forced out by his board on December 1, 2024, eight weeks before the inauguration; the stock had lost more than half its value over his tenure. On March 12, 2025, ten days after the TSMC ceremony, Intel named Lip-Bu Tan, the Malaysian-born venture investor who had run Cadence Design Systems for thirteen years and had served briefly on the Intel board before resigning in protest at the slow pace of the foundry plan. Tan was sixty-five, respected across the industry as the rare leader who had made money in EDA, in venture capital with Walden International, and on chip-company boards on both sides of the Pacific. The appointment was read as the board’s recognition that the foundry strategy required someone who could speak fluently to TSMC’s customer base and to TSMC itself. By August, the Trump White House had converted Intel’s CHIPS Act grants into a roughly nine-billion-dollar equity stake of about ten percent. SoftBank had separately invested two billion. The U.S. government was, for the first time since the Reconstruction Finance Corporation era, the largest shareholder of an American semiconductor manufacturer. Lutnick described the conversion as a renegotiated contract. The financial press described it as something closer to nationalization. Tan called it a “partnership.”

By late summer, the shape of the second front had become visible. Taiwan settled into a fifteen percent baseline tariff with most-favored-nation language carving out exemptions for chips fabricated by companies investing in the United States; TSMC’s hundred-and-sixty-five-billion-dollar Arizona commitment turned the company into its own tariff shield. South Korea reached a similar fifteen percent settlement, with Samsung’s thirty-seven-billion-dollar Taylor fab and SK Hynix’s Indiana HBM packaging plant doing the same work. The Section 232 investigation produced a tariff regime that exempted companies with U.S. manufacturing footprints, turning capacity-localization commitments into the central currency of trade policy. The AI Diffusion Rule had been replaced not by a global framework but by a rolling series of bilateral arrangements, each negotiated by the president personally, with national champions designated by name.

The Section 232 outcome itself arrived on January 14, 2026, when Trump issued a presidential proclamation imposing a narrowly targeted 25 percent tariff on a defined category of advanced computing chips not destined for the United States supply chain. The carve-outs were the heart of the rule. Chips imported “to support the buildout of the United States supply chain” were exempt, including silicon used in U.S. data centers and U.S. public-sector applications. Nvidia’s H200 and AMD’s MI325X were named, by example, as covered items in the absence of an exemption. The structure made the tariff operational against shipments destined for non-U.S. customers, while leaving most of the actual flow into American hyperscaler data centers untouched. Five weeks later, on February 20, 2026, the Supreme Court ruled six-to-three that the broader IEEPA-based reciprocal tariffs Trump had declared on Liberation Day were unlawful, a decision that struck down a substantial portion of the global tariff architecture but left Section 232 actions intact, on the grounds that 232 had specific statutory authority that IEEPA did not provide. The chip tariff survived the constitutional ruling that had unwound much of the rest of the Trump tariff regime. The architecture that had emerged from the spring of 2025 was, by early 2026, narrower in legal scope than its drafters had originally envisioned and broader in commercial effect than its critics had predicted.

The chip war’s American politics, in the eight months after Trump’s return, had not produced a coherent doctrine. It had produced four overlapping ones: a tariff doctrine that treated chip imports as bargaining leverage, an industrial-policy doctrine that took equity stakes in domestic champions, an export-control doctrine that monetized waivers as revenue shares, and a bilateral-diplomacy doctrine that handed AI compute to Gulf allies in exchange for sovereign-wealth commitments to American projects. Each had its own constituency inside the administration and was capable, on a given day, of contradicting the others. In Beijing, the changes were read as evidence that the American chip-policy edifice had become improvisational and exploitable. In Hsinchu, they were read as confirmation that the silicon shield was now a function of how much Taiwan was willing to invest in American soil rather than how much American protection Taiwan could buy. In Washington, the same officials who had spent the late Biden years building a global architecture were spending their early Trump years dismantling it.

Lutnick, in television appearances through the summer, kept returning to the same metaphor. The CHIPS Act had been a giveaway. The new approach was a deal. America would no longer pay companies to build here. America would charge a price for access to the American market and use the leverage to extract investment, jobs, and equity. The metaphor had a certain coherence. It also had a blind spot. The chip industry the United States was negotiating against was the chip industry the United States had spent forty years offshoring. The Beijing officials watching the H20 reversal, the Riyadh officials watching the Humain deal, and the Hsinchu officials watching the Roosevelt Room ceremony were all drawing the same conclusion. American chip policy was no longer a function of a stable bipartisan consensus. It was a function of who had walked into the Oval Office last. The first front had not gone away. The second front, the contest inside the United States over what the chip war was even for, was now the one that determined how the first was being fought.

In the Roosevelt Room on March 3, after the cameras had been turned off, Wei had told a small group of American officials that TSMC’s commitment had been finalized only in the previous week. The Taiwanese company had spent the first six weeks of the new administration trying to read what kind of deal Trump would accept. The hundred-billion-dollar number was the answer Wei’s team had concluded the president needed to hear. Whether it would actually be spent, and on what timeline, was a separate question, the kind the previous administration’s CHIPS office staff had been built to answer and the new one no longer had the people to. The answer, in the Trump II theater, was the announcement. The implementation was what came after, when the cameras had moved on and the next deal was being negotiated by the same small circle who had decided that the chip industry’s politics, after forty years of bipartisan technocracy, was now theirs to make up as they went.