Winning the AI Race Part 5: President Trump on the AI Action Plan
Summary
Trump’s AI Action Plan is an industrial-capacity strategy built around permits, power and private capital. The administration wants rapid approvals for data centers, semiconductor facilities, transmission and dedicated generation, arguing innovators need “a green light, not strangled with red tape.” The investable thesis is that AI demand flows through the entire physical stack, from natural gas and nuclear to construction, HVAC and electrical work.
Trump framed electricity as the binding constraint, saying the AI industry may need roughly double current U.S. generation. He promised companies the right to build their own power plants, become “your own utility,” and sell surplus electricity into the grid. He said the U.S. would add at least as much capacity as China after China added 11 times more than America during Biden’s final year.
Immediate expensing materially strengthens the economics of AI infrastructure investment. Trump said companies can deduct 100% of capital expenditures—including factories, equipment and, newly, structures—immediately rather than over 38 or 42 years, with the provision available for 10 years. His message to capital allocators was direct: “Get out and take advantage of it.”
The plan couples federal preemption with a permissive approach to AI training data. Trump argued that developers cannot negotiate thousands of contracts whenever models learn from books or articles; copying or plagiarism remains out of bounds, but AI should be allowed to use the resulting “pool of knowledge.” He also demanded one federal standard superseding state regimes, warning that otherwise the most restrictive state effectively governs the entire industry.
The export strategy treats adoption of the American technology stack as a national-security asset. Trump said necessary protections would remain, but “the greatest threat of all is to forfeit the race” and drive partners toward Chinese technology. A new order directs expansion of exports spanning chips, software and data storage after repeal of the Biden-era diffusion rule.
Trump cited an enormous capital pipeline, mixing administration claims with company commitments. He cited almost $17 trillion entering the country, $320 billion or more this year from Meta, Amazon, Google and Microsoft for data centers and AI infrastructure, NVIDIA’s $500 billion four-year commitment, and $92 billion from 20 companies in Pennsylvania. If realized, those figures would make power availability, construction speed and equipment supply the near-term execution tests.
Federal procurement becomes the administration’s enforcement point against ideological bias, while startups are cast as the source of U.S. advantage. Trump ordered agencies to buy only AI he described as pursuing “truth, fairness, and strict impartiality,” while arguing that excessive regulation would centralize models among a few incumbents able to censor or control them. His counterexample was NVIDIA: Jensen Wong “started as very small,” illustrating why rules must leave room for the next major platform company.
Deep dive
1. Tariffs are meant to buy both market access and domestic factories
Trump opened with the newly signed Japan deal: the U.S. tariff falls from 25% to 15%, American companies enter Japan at zero tariff, and Japan provides what he called a “$550 billion signing bonus.” He added, without fully unpacking the economics, “To be exact, it’s 90% of that, but we control the whole lot of it.”
The broader trade framework is deliberately blunt: countries without negotiated deals could face “a straight simple tariff” between 15% and 50%. Trump said the EU could earn a lower rate by opening its market, while citing deals with the Philippines, Indonesia and the UK and ongoing work with China.
Market access may matter more than the tariff itself if U.S. companies compete successfully, in Trump’s framing. He also said automobile companies are moving plants into America because “they don’t want to pay the tariffs,” while separately arguing that cash-rich chipmakers need permits and rights to build rather than subsidies.
2. Winning AI requires permissive rules for knowledge and one national standard
Trump called “artificial intelligence” the wrong name—“It’s genius. It’s pure genius”—and cast leadership as civilizational: America cannot leave its children on a planet “controlled by the algorithms of adversaries advancing values and interests contrary to our own.” He claimed the U.S. was already leading China “very substantially.”
He preserved an important hedge around AI’s reach. Brilliant people tell him it will “dominate every industry that ever existed,” but his own answer was: “I don’t know if that’s true.” What he did treat categorically was the need to remove regulatory obstacles before the competition accelerates.
On training data, Trump distinguished reproduction from learning: “Of course, you can’t copy or plagiarize an article,” but reading and learning should not require separate payment or contract negotiations with every content provider. Otherwise, thousands of permissions could attach to each AI use while China operates under different rules.
State-by-state regulation is incompatible with a national model market, he argued, because “the most restrictive state of all will be the one that rules.” His proposed federal standard would supersede state requirements and avoid litigation across “43 states at one time”; foreign rules should likewise not become the toughest global default.
3. Permitting and tax policy turn the plan into a construction program
The first infrastructure pillar covers data centers, semiconductor plants, power stations and transmission lines. Trump said virtually all large capital investments “can be and should be made by the private sector”; government’s role is to provide certainty, environmental review and the ability to start building.
Speed is the organizing principle. Trump said he had instructed Lee Zeldin to target “one week for nuclear” approvals and “a couple of days for oil and gas,” while the new executive order fast-tracks federal permitting and reviews for major AI infrastructure.
The deregulatory formula is 10 old regulations eliminated for every new one. Drawing on his property-development experience, Trump noted that a six-year zoning process can outlast the market itself—occasionally saving a project when conditions collapse, but usually making approvals economically stale.
The tax accelerator is 100% immediate expensing for factories, equipment and structures; structures were not included in Trump’s first-term version. Instead of depreciating investment over 38 or 42 years, companies can deduct it immediately, and this time the window lasts 10 years.
4. Power becomes the decisive input—and companies become utilities
Trump said AI builders need “double what we produce right now for everything,” turning abundant electricity into the precondition for model leadership. His answer spans natural gas, oil, “clean, beautiful coal,” nuclear and company-owned generation rather than dependence on a grid he described as more than 100 years old.
Trump’s view of nuclear has changed: “I had problems with nuclear,” but advances have made it, in his current assessment, long-term, inexpensive and safe. He expects some AI companies to choose nuclear over oil or gas under strict but newly reopened approval pathways.
Every company would be allowed to build generation beside its project, operate as “your own utility,” and sell excess electricity back to the grid. Trump promised at least to match China’s capacity additions, citing 57 Chinese coal plants and an 11-to-one gap in new generation during Biden’s last year.
5. A capex wave is already testing the physical supply chain
Trump put almost $17 trillion of investment into the U.S. pipeline over a few months. For AI specifically, he cited $320 billion or more this year from Meta, Amazon, Google and Microsoft, plus NVIDIA’s $500 billion commitment over four years: “If you’re doing it, it’s good.”
In Pennsylvania, 20 companies announced $92 billion of energy and data-center projects, including what Trump said would become North America’s largest natural-gas power plant. He said it was already under construction, presenting it as an example of energy and compute capacity arriving together.
The employment case extends beyond engineers: Trump expects thousands of blue-collar roles and higher wages for energy workers, HVAC technicians and electricians. His underlying chain is straightforward—AI capex creates power demand, power and data centers require construction, and faster permits pull that spending forward.
6. Exports and procurement define the technology bloc Trump wants
Trump said Biden’s diffusion rule constrained American AI exports, alienated partners and pushed even friendly governments toward competing technology. He repealed it while promising “necessary protections for our national security,” because forcing partners onto rival infrastructure could ultimately cost America the race.
The export order directs Secretary Howard Lutnik and Secretary of State Marco Rubio to expand overseas sales “from chips to software to data storage.” Trump cited Middle Eastern leaders’ appetite for U.S. technology as evidence that partners want an American stack when Washington permits access.
The procurement order targets what Trump called “woke AI”: federal agencies should reject models infused with partisan agendas or critical race theory and procure only systems pursuing “truth, fairness, and strict impartiality.” A subsequent speaker, not identified by name in the transcript, summarized the same three orders—permitting, exports and ideologically neutral federal purchasing—as prepared for Trump’s signature.
Trump’s structural argument was that restrictive rules benefit a few large companies able to “centralize it, censor it, control it, weaponize it,” while American strength comes from startups and “small tech.” The accompanying bargain is explicitly national: Silicon Valley should put America first after building in China, hiring in India and shifting profits to Ireland.