Inside America’s AI Strategy: Infrastructure, Regulation, and Global Competition
These notes were generated by AI and may contain inaccuracies.
President Trump gave a major AI policy speech in July, declaring the US must win the AI race, reminiscent of President Kennedy's space race declaration. American companies have innovated more since, releasing incredible products. American AI models, chips, and data centers keep improving. US has competent competitors like China with many smart people, but Silicon Valley innovation is incredible.
Huge infrastructure buildout due to demand. Unlike late '90s fiber buildout leading to dark fiber and dot-com crash, every GPU is used now to generate tokens powering AI chatbots and coding assistants. Recent coding releases are mind-blowing, revolutionizing software development. Demand for tokens drives data center buildout. Last year, it added about 2% to GDP growth, propelling 4-5% growth rate; similar expected this year.
Three Pillars of US AI Plan
- Out-innovate competitors.
- Drive infrastructure buildout for AI revolution.
- Export American technology.
Federal government has taken actions on all three, making good progress.
Core insight: regulatory environment must allow technology development and commercialization. US excels compared to world but can improve. Patchwork of state regulations detrimental, especially to early-stage companies and entrepreneurs building on frontier models—big companies succeed best. Working on legislative proposal for sensible national framework.
President's December executive order directed work on this, listing areas states can pursue: child safety, data center permitting.
Over 200 bills in state legislatures now—knee-jerk reactions to hypothetical concerns. Better to study actual AI uses and risks before overregulating. President consistent on need for single federal rulebook, lightweight standard. Patchwork worsens over time. Only Congress can preempt states (needs 60 Senate votes, bipartisan). Early stages of building consensus this year.
Pushback in Congress to preemption without federal standard, but interest in lightweight federal standard in House and Senate.
Bernie Sanders letter called to stop all data center development—would lose AI race. China builds new nuclear or coal plant weekly to power data centers. President Trump clear: consumers should not pay higher electricity rates due to data centers. Microsoft pledged last week its data centers won't increase residential rates; expect others to follow. Hyperscalers and AI companies plan own power generation, not drawing from grid.
Secretary Wright reforming regulations to ease behind-the-meter power for data centers. President's vision: let AI companies become power companies, building power alongside data centers—gets infrastructure without raising residential rates.
Race now power race. Buildout net positive for ratepayers long-term, lowering rates. Data centers can sell excess power back, amortizing fixed costs over greater supply for economies of scale, reducing prices. Must ensure they contribute back, not just plug in.
Biden administration policy blocked behind-the-meter generation; now changed by Secretary Wright and FERC to allow it, benefiting ratepayers with greater scale.
Concerns banks left holding bag, but savvy investors like Oracle, Blackstone, real estate companies investing due to seen ROI.
Evolution: started with chatbots like ChatGPT (better web search), added chain of thought for reasoning, then coding assistance—major recent breakthroughs impressing developers.
Next: tools for knowledge workers generating Excel models, PowerPoints, websites, etc. Expect productivity boom in 2026.
Industry verticals:
- Healthcare: reduce administrative bureaucracy, paperwork; AI in research for cures; users/doctors input records for remarkable diagnoses.
AI for Science: Early models trained on internet scrapes; coding on existing code. Science data fragmented across disciplines (chemistry, math, material science). Launched Genesis mission to leap in AI scientific discovery using national labs' 50-60 years of DOE data. Accelerate experiment selection, execution, iteration. Dream: double R&D output in 10 years via AI labs running hypotheses/experiments.
Expected breakthroughs:
- Faster fusion simulations to advance timelines.
- Material science for space (lunar base, Mars, nuclear in space).
- Healthcare therapeutics: identify molecules, iterate to clinical trials.
Auto sector beneficiary, especially self-driving—inflection point with robo-taxis (Waymo, Tesla).
China uses AI assistants for bills, cleaning, gifts. Expect commonplace this year. Latest Claude (powered by Anthropic's Opus 4.5)—breakthrough in coding. New "co-work" tab for non-coders: spreadsheets, PowerPoints; points to file drive/email to emulate style/format. Task-based now, but one abstraction layer away from personal digital assistant connected to data sources.
Voice interface like movie Her possible in 2026—tools evolve from coding to personal assistants.
Easy to see AI as sophisticated chatbot, but long-term impacts across industries via faster scientific discovery velocity.
US ahead of China overall. Layers of stack: models, chips, chip equipment—deeper in
The deeper in the stack, the greater the American advantage. US models are about 6 months ahead of Chinese models. US chips are 2 years ahead. Semiconductor manufacturing equipment is 5 years ahead. US has significant advantages there. Only a couple areas where China has an advantage.
China's grid has roughly doubled in the last 10 years, while US grid grew by 2-3%. Energy production in US was sleepy before AI due to regulations and previous administration's antipathy. President Trump understood energy growth as precondition for economic and AI infrastructure growth, advocating "drill baby drill" 10 years ago. US needs to expand energy production to catch up.
China has edge in AI optimism. Stanford polling: 83% of Chinese feel AI benefits outweigh harms; US only 39%. Asian countries high, Western countries low. Explanations: media focus on doom and gloom; Hollywood dystopian portrayals like Terminator or 2001; tech leaders not describing benefits well, talking about AI eliminating 50% of knowledge workers, fueling pessimism.
AI pessimism feeds regulatory frenzy: 1,200 bills at state level. US ahead in key dimensions but could overregulate and lose the race.
Overemphasis on leaderboards; frontier models neck-and-neck, US 6-12 months ahead. History shows you don't need the best tech to dominate globally, e.g., Huawei in telecom wars—not best vs. Ericsson/Nokia but good enough and subsidized, became default in much of world.
China ambitions to export models to global south. President launched American AI Export Program to share US tech with partners/allies: ensure developers worldwide fine-tune American models on American chips. US dominant in stack: best models, applications, chips. Launched late last year, big push this year.
China developing own models; DeepSeek release ~1 year ago put Chinese AI on map, reduced Western complacency. Previously, no talk of global competition or regulation slowing US vs. China, e.g., Biden's 100-page AI executive order.
China blocking Nvidia chips to indigenize, stand up Huawei as national champion via market subsidy. Plan: dominate China first, then scale globally—chip production is scale business.
Closed Commerce Department request for information late last year on exporting American AI stack. Now ingesting info; request for proposals soon for companies to form consortia on packages.
Global buyers vary in sophistication. US Fortune 50 companies have sophisticated CIO/CTO evaluating clouds, models, fine-tuning. Many countries lack billion-dollar IT budgets, need simple tools for healthcare, tax collection.
Focus on turnkey solutions: manageable-sized data centers for inference (not large training like Colossus). Partner with Development Finance Corporation, Export-Import Bank for financing. Few countries aspire to frontier model training; most need inference chips.
Upcoming: India AI Impact Summit next month to share progress.
Win if American chips/models everywhere in 5 years; lose if Huawei chips/DeepSeek models. Proliferation of US tech critical.
Silicon Valley lesson: biggest ecosystem wins—most apps, developers on API, platform companies. Create value for partners; not all countries develop chips/models but can apply US tools to economies.
Partner mindset common in Silicon Valley, alien to Washington's command-and-control bureaucracy.
Inherited Biden regulations: 100-page AI executive order, 200-page Biden diffusion rule on semiconductor exports—turning AI into highly regulated industry.
President Trump rescinded them first week in office. Silicon Valley built on permissionless innovation: founders start in garage/dorm without Washington permission—Hewlett-Packard 85 years ago. Biden's 300 pages would end that.
Trump's AI Action Plan: pro-innovation, pro-infrastructure, pro-energy, pro-export. Results visible in past year.
Export US approach: create regulatory environment for success. Europe has roadblocks despite Draghi report.
US mindset: whiteboard rules to unlock innovation, remove barriers—not precautionary principle of listing hypotheticals to prevent.
AB test: US vs. Europe last 20 years shows recipe. Draghi report: Novo Nordisk $350-400B; US has trillion-dollar companies, Nvidia $5T.
Innovation from private sector entrepreneurs; government enables, sets rules/guardrails, lets them cook.
EU AI leadership: regulators as main characters (main character syndrome)—wrong; regulators supporting, entrepreneurs main.
EU AI Act passed before ChatGPT, detrimental, now needs edits as innovation outpaces.
Concerned about Orwellian scenarios: government misuse for surveillance, censorship, brainwashing—not Terminator.
Woke AI: political bias built in, subtle impact on learning/thinking, especially children. Biden EO had 20 pages on DEI, promoting DEI layer in models—led to Gemini's black George Washington rewriting history.
Trust/safety apparatus like social media censorship could embed in AI to control discourse.
Trump rescinded Biden EO first week; signed EO that federal government won't procure politically biased AI. Companies have 1st Amendment right to bias but government won't buy it. Secure for next 3 years, but future regimes could pressure for bias—serious threat to freedoms.
Elon Musk says AI does it all, no need to work—leads to abundance, no money needed (like Star Trek, not just job loss). People report only job loss part. Elon is a friend, slight disagreement, but
Directionally correct about the future: heading toward a world of much greater abundance, rising living standards for everybody, greater productivity. This will lead to rising wages. Will not put everyone out of work.
Timelines matter a lot. Getting to a world with no money is not going to happen in the next 5 years.
And of course, Michael, this is helping us in terms of longevity and living longer, right? In terms of the impact on science.
Totally. The abundance story extends itself well into health care and everywhere else, and just quality of life. Good things ahead.
We'll leave it there. Michael Katzios and David Sacks, thanks so much.
Thank you.
Keep All-In Podcast in your library
Save the videos and channels worth coming back to, and find them again in one place.
