Winning the AI Race Part 4: Scott Bessent, Howard Lutnick, Chris Wright, and Doug Burgum
Summary
Bessent’s fiscal thesis is that AI can turn a roughly $300 billion annual capex boom into the non-inflationary productivity growth needed to make his “3-3-3” arithmetic work. The plan targets a deficit near 3% of GDP from 6.7%, persistent 3%-plus growth, and 3 million additional barrels of oil equivalent; June produced Treasury’s first positive June since 2015. He sees AI potentially echoing the disinflationary railroad and 1990s IT expansions, with construction eventually giving way to use cases and productivity.
Bessent argued that tariff inflation has not materialized because foreign producers are cutting prices and US companies are absorbing margin pressure while investment moves onshore. China’s 30% rate meets what he called the “brooms and the water buckets from Fantasia” business model: keep cutting costs to defend market share. He expects at least $300 billion of tariff revenue over the next 12 months, while immediate expensing for equipment and factories supports projects such as AstraZeneca’s announced $50 billion US buildout.
Stablecoins could become a major new buyer of short-dated Treasuries and reinforce global dollar use even as China slowly reduces its holdings. Bessent said the GENIUS Act could generate “several trillion dollars of demand for T-bills” under 90 days, putting dollar-backed stablecoins into consumers’ hands without the direct controls of a central-bank digital currency. His contrast was blunt: governments can freeze a CBDC account, while stablecoins offer “unbridled choice.”
The administration’s near-term AI-power strategy is natural gas plus retained baseload, with advanced nuclear treated as a roughly 10-year industrial rebuild. Wright said planned closures could remove 100 GW by 2030 even as the country needs to add at least that much, so “the first thing to do is stop subtracting 100.” Burgum said gas and halted retirements must win the next 24 months; Wright expects Gen 4 and small modular reactors to reach free-market economics only after supply chains and production volumes mature.
Solar produced the panel’s clearest disagreement because Wright judged reliability at peak demand while a host emphasized annual contribution and falling costs. Wright said all US batteries could store only five minutes of national power, and that Texas wind and solar supplied 8% at peak despite representing 35% of capacity; the host replied that those constraints cover “two weeks” while the other 50 also matter. Wright’s answer was that those two weeks are “game time,” invoking the more than 200 deaths during Winter Storm Uri.
Power availability is already reshaping data-center siting, equipment supply chains, and skilled-labor economics. Burgum adopted Jensen Huang’s “AI factories” framing because the facilities repeatedly manufacture intelligence, and said turbine shortages may require coordinated supply mapping or the Defense Production Act. His fastest route is to colocate generation and compute beside stranded gas in the Marcellus, Permian, or Bakken, avoiding new pipelines and transmission while creating trades jobs he said can pay $120,000 initially and reach $150,000.
Lutnick presented tariffs as a mechanism for extracting both market access and infrastructure capital, while drawing a security line around frontier technology. Japan committed $550 billion to finance projects important to the president and US infrastructure, with lease economics split 90% to America and 10% to Japan; unresolved countries would face higher tariffs from August 1 while retaining the option to negotiate afterward. With China, ordinary trade belongs “below the line,” but H200s, H100s, and other leading chips remain above it; Lutnick’s emerging framework was cluster size and who controls it—“ally or not”—with trusted American operators and clouds part of the discussion, while TikTok must become American-owned and run on an American technology stack.
Deep dive
1. Bessent’s 3-3-3 plan depends on productivity outrunning the debt burden
Bessent defined “3-3-3” as reducing the deficit from roughly 6.7% to 3% of GDP, sustaining 3%-plus growth, and adding 3 million barrels of energy equivalent before Trump leaves office. June delivered Treasury’s first positive June since 2015 through higher revenue, including tariffs, and lower spending.
AI capex from the largest hyperscalers is already about 1% of GDP, or $300 billion annually. Bessent sees a possible transition from the current construction boom to an AI use-case phase and productivity-led growth that changes the debt trajectory while remaining non-inflationary.
His historical case ran from railroads making cross-country travel 10 times faster amid double-digit growth and inflation of negative 2% to negative 4%, through Reagan-era deregulation, to the 1990s IT boom. That last cycle eventually produced a surplus and discussion of what markets would do without enough government bonds: “But we fixed that.”
Asked whether the Fed should remain independent or replace Powell, Bessent did not answer directly. He expects perhaps one or two cuts after evidence that tariffs are not persistently inflationary, while limiting himself to “the mistakes they made, not the mistakes they’re going to make” and joking that the Fed resembles “universal basic income for PhD economists.”
2. Tariffs are designed to raise revenue, compress foreign margins, and pull factories onshore
Bessent said the feared tariff pass-through had not appeared yet: foreign suppliers cut prices, US companies accepted lower margins, and China’s business model, in his description, is an employment agency that keeps cutting costs to preserve market share. His distinction for the Fed was that a one-time price-level increase is not a continuing inflation spiral — hence “tariff derangement syndrome.”
The second-order effect is construction. Bessent cited AstraZeneca’s $50 billion US commitment and paired tariffs with five years of 100% immediate expensing for equipment, plus write-offs for factory structures: first comes the buildout, then “the factories get populated.”
Even if China slowly divests Treasuries, Bessent thinks GENIUS legislation could create several trillion dollars of demand for T-bills under 90 days. Dollar-backed stablecoins would also extend dollar use from Nigeria to Qatar while avoiding a CBDC’s capacity to freeze accounts over conduct the government dislikes.
Bessent also stressed permitting as an execution obstacle. TSMC’s Arizona complex might eventually supply up to 7% of US chip needs, yet changing designs can collide with inspectors objecting that “you said the pipe was going to be there, not there”; the lesson from earlier expansions is to “make it easy to build things again.”
3. The power race begins by cancelling planned capacity destruction
The host framed the gap starkly: US generating capacity is about 1 TW and may reach 2 TW by 2040, while China moves from roughly 3 TW to 8 TW — adding America’s entire capacity every 18 months. Burgum added that China installed 94 GW of coal last year and still gets over 60% of its electricity from coal while simultaneously building nuclear and hydro.
Wright said natural gas, nuclear, and coal provide 75% of US electricity and 90% of supply available regardless of weather. Previous plans contemplated removing 3.5 GW of hydro and closing 100 GW of power plants by 2030: “The first thing to do is stop subtracting 100 at the same time you want to add a hundred.”
The solar argument turned on timeframe. A host emphasized California and Texas output and battery cost declines; Wright countered that PJM’s wind, solar, and batteries delivered only 3% at peak, while Texas wind and solar supplied 8% at peak despite holding 35% of capacity. “Those are the two weeks that matter.”
Wright accepted that atmospheric CO2 has risen 50%, absorbs infrared radiation, and has contributed to warming, but said climate is not among the planet’s five largest problems. He favors unsubsidized solar, while arguing that cheap gas displacing coal and always-on nuclear have stronger decarbonization mechanics.
4. Gas owns the next 24 months while Gen 4 nuclear climbs a 10-year curve
Wright called natural gas the dominant source of new US electricity because it is cheap, abundant, fast to deploy, dependable, and easier on machinery. Regulatory priorities include reforming FERC’s clogged interconnection queue, restoring NEPA as an environmental process check rather than “an avenue for lawfare,” and removing what he called Clean Power Plan 2.0.
Burgum separated the immediate AI contest from nuclear’s longer payoff: four Trump executive orders had helped draw fresh capital to close to a dozen small-reactor startups, but “it’s not the thing we need in the next 24 months.” Near-term power must come from gas and stopping existing plants from closing.
Wright nevertheless called nuclear his single largest workstream. Three next-generation Gen 4 reactors are expected to go critical at Idaho National Laboratory next summer; the government is supplying HALEU fuel to five developers, with plans to reach roughly a dozen, while tax credits help restart an industry he said government “smothered” for three decades.
His estimate for nuclear operating without support was “probably 10 years.” Small modular reactors first need repeated builds, supply-chain depth, and volume economics; meanwhile, physical AI and potentially hundreds of millions of robots only increase the value of producing energy equipment at scale.
5. AI factories will migrate toward fuel, equipment, and buildable land
Wright said the Energy Department offered 16 sites where it could rapidly permit data centers and adjacent generation, drawing 300 responses; the first four sites to be developed were to be announced “tomorrow.” The response count supported Burgum’s warning that power demand is probably underestimated rather than speculative.
Burgum urged the industry to stop saying “data centers.” Unlike systems processing shopping or healthcare claims, AI factories are “literally manufacturing every day over and over more intelligence”; the Energy Dominance Council is mapping supply-chain shortages, including turbines, pushing suppliers to expand, and considering tools including the Defense Production Act.
Burgum estimated $1 trillion to $1.5 trillion of capital may be trapped in two-to-four-year federal permitting, while citing $15 trillion of investment “coming back.” His siting shortcut is colocating off-grid AI factories beside stranded gas in the Marcellus, Permian, or Bakken, eliminating pipeline and transmission approvals while expanding trade jobs paying $120,000 to $150,000.
6. Lutnick converts tariff pressure into project capital and technology boundaries
Lutnick described Japan’s $550 billion commitment as a “signing bonus” funding projects important to the president and US infrastructure, including nuclear plants, fabs, pipelines, critical-mineral projects, or shipbuilding. Japan would pay for construction, assets would be net-leased to operators, and lease payments would split 90% to America and 10% to Japan — separate from a future sovereign wealth fund created only after addressing the deficit.
The structure took five months: Japan initially offered loans or guarantees, Trump rejected more borrowing, and the parties moved to committed capital. Lutnick said the deal prompted Korea to send officials to his office after seeing the price Japan paid.
Market opening remains the first demand — “open, open, open” — with bespoke tariffs where countries refuse. Lutnick described Vietnam and Indonesia as completely open and the Philippines as mostly open, with exceptions including Indonesia’s restrictions on pork and alcohol; on August 1, unsettled countries would move from 10% to rates such as 19%, 26%, 27%, or 31%, and could keep negotiating only while paying.
For China, Lutnick drew a line between baby clothes and soybeans below it, and H200s, H100s, hypersonics, and other strategic systems above it. His emerging export framework is cluster size and who controls the cluster — “ally or not” — rather than ally status alone, with trusted American operators and clouds part of the discussion; TikTok likewise requires American ownership and an American technology stack, or the alternative is shutdown.