America Must Win the World of Atoms, Not Just Bits
The path to winning in AI includes autonomous systems.
Last week, the Trump administration released Winning the AI Race: America’s AI Action Plan, a major step in aligning U.S. policy. The plan emphasizes deregulation, support for open-source models, expanded energy infrastructure, and represents a shift away from the risk-averse “safetyism” of the Biden era. That same day, at a summit in Washington, Trump left no ambiguity about his priority: it’s time to “build, baby, build.”
Among the plan’s highlights:
Rescind Executive Order 14100, which imposed “bias” guardrails on AI
Direct the OMB and FCC to consider the state AI regulatory climate when awarding grants and permits
Create a market for compute to reduce friction from long-term contracts
Increase federal support for open-source and open-weight models
Fast-track AI-related energy infrastructure, including nuclear, with broad NEPA exemptions for data-center-related activity
Removing onerous regulatory burdens while maintaining a minimal regulatory environment is the right approach— and fully aligned with Boyd’s vision of technorealism. His Action Plan is therefore a significant and positive development.
But it’s not enough. Trump’s Action Plan is overly focused on only one aspect of AI—LLMs. It has a major blind spot: physically instantiated AI. Or as we put it, atoms. America could win the generative-AI race and dominate the “world of bits” only to cede the future, in its infinite horizon, to Beijing by neglecting the “world of atoms.”
Given that the stakes are existential—any war will be decided in the physical world—the Trump administration should focus on a simple maxim:
Winning in AI means leading in autonomous systems.
Yes, this poses risks that should not be downplayed. But we believe the opportunities outweigh the risks—and that winning in autonomy is a requirement for geopolitical and commercial competitiveness. We must lean into the future and build.
The State of the Race
The Trump plan’s blind spot on the “world of atoms” reflects a broader societal focus on the “world of bits.” For decades, much of our technical and business brilliance has been poured into the digital, rather than real, world. With the partial exception of Elon Musk, few have focused on applying AI to the physical world with the same level of attention, resources, and ambitions as, say, software. The result is that we’re falling behind.
For example, look at robotics: In 2022, the U.S. held only 9 percent of the global market share, ranking fourth; by contrast, China held almost 40 percent. All evidence indicates that this lead has only grown over the past three years.
In drones, the gap is even wider: In 2023, China and Hong Kong represented 57 percent of the drone export market compared to America’s 5.4 percent—below Turkey’s 5.9 percent. Despite this, America remains a substantial market for drones even if we do not produce enough ourselves—we were the second-largest importer of drones after Ukraine.
As a result, America faces the dual threat of being unable to match a near-peer like China’s productive capacity while simultaneously having our AI cloistered in a digital silo.
Without a national push for autonomous systems—drones that move goods, robots that build, machines that detect and defend—we’ll continue to fall behind.
What a Real Autonomy Strategy Would Look Like
Pete Hegseth’s newly announced drone initiative offers serious potential. It shows an Autonomy Initiative that puts hardware, manufacturing, and defense back at the center of U.S. competitiveness is still possible. This is just the beginning though. We also recommend the following policies:
1. Mobilize Supply Chains
Direct the Commerce Department to convene industry and government leaders to map the entire supply chain for U.S. robotics, drones, and autonomous-system hardware.
Publish the top ten bottlenecks and release a corrective-action plan with deadlines and responsible entities. Track and report progress quarterly.
2. Build Testing Infrastructure
Fast-track a DoD AI & Autonomous Systems Virtual Proving Ground
Fold it into a Frontier Autonomy R&D Program that treats large-scale simulation data as a national strategic asset
Grant vetted commercial firms access so prototypes move from lab to field
3. Modernize Regulation
Reform FAA rules to allow drones beyond visual line of sight
Create a national framework for autonomous maritime transport
Consolidate local drone laws into one federal standard
Apply FCC-style labeling and modularity to reduce risk and vendor lock-in
4. Prime the Fiscal Pump
Designate 500 sq mi of federal desert land as an Autonomy Freeport; waive most FAA/FCC/OSHA/EPA rules and grant 10-year tax holidays to firms that hit safety and job-creation benchmarks
Invoke the Defense Production Act to guarantee procurement of at least: 5,000 cargo drones, 1,000 robotic construction units, and 100 autonomous ship kits by 2029
Seed a National Autonomy Bank with $30B in Treasury equity and allow 10× leverage to provide $300B in low-interest loans to domestic robot and drone factories
Launch an American Autonomy Corps to pay 50,000 recent graduates a two-year stipend to build and maintain autonomous systems on public works projects, convertible to technical-school credits
Incentivize domestic manufacturing with tax credits, stockpiles, and federal demand mandates
The benefits extend far beyond defense. We should pursue autonomy use cases that deliver tangible benefits to Americans: firefighting bots, automated farming, drones for border monitoring, search and rescue, and even space debris cleanup.
Conclusion: Real AI Means Autonomy
We’re hitting the limits of what can be done in the digital world. The next chapter of AI won’t be written by LLMs alone. It will be driven by autonomous systems that operate in reality.
Trump’s Action Plan sets the right foundation, but now we must build on this momentum. An American Autonomy Initiative is key to American competitiveness and a way to extend American success to the world of atoms, not just bits.
Good stuff. "Publish the top ten bottlenecks..." is solid Theory of Constraints process.
This is well researched and well reasoned. I would recommend reaching out to as many ai-talking-heads as possible and asking them their thoughts or for a review.
Likely you’ve already done that, but this article is too good to ignore. I imagine there are even people in the administration that would genuinely value reading something like this.