The 28 Pages That Could Save American Tech
And for those that hate reading, a summary of the article and the AI Action Plan provided by a Roman Legionary on the subway.
In Washington's theater of ritual gestures, where strategy documents typically arrive with fanfare and depart without consequence, this week marked a departure from the script. When President Trump brandished that leather-bound folder at the Hill & Valley Forum and declared America would "do whatever it takes to lead the world in artificial intelligence," the venture capitalists in the room erupted. They recognized what many in Washington still miss: this wasn't just another strategy document. It was a permission slip for the future.
The AI Action Plan reads like an interstate highway act for the digital age: 28 pages containing over ninety concrete directives that could reshape how America builds, deploys, and dominates in artificial intelligence. But beneath the technical specifications and bureaucratic restructuring lies a more profound shift: America is choosing acceleration over caution, builders over gatekeepers, speed over perfection.
To understand why this matters, consider what the plan actually does. It tears down the regulatory patchwork that has trapped AI development in a maze of conflicting state rules and federal guidelines. By establishing pre-emption authority and threatening to withhold federal AI funding from non-compliant jurisdictions, it creates something unprecedented: a single, nationally coherent standard for AI development.
It promises to fast-track permits for data centers and semiconductor fabs with the same urgency we once reserved for wartime production. Most critically, it embraces open-weight models—making the underlying parameters of cutting-edge systems freely available so agencies and companies can audit, customize, and deploy them inside secure or classified environments.
This is a recognition that in the AI race, the prize doesn't go to the most cautious but to the most capable. While we've been debating AI ethics in conference rooms, China has been pouring concrete for compute centers. While we've been drafting frameworks, they've been training models. The gap between American innovation and Chinese implementation has been narrowing at an alarming rate.
The plan's architecture reveals its ambition through three interlocking pillars. The first—accelerate innovation—fundamentally rewrites the relationship between federal and state regulation. NIST will establish a robust evaluation ecosystem for model safety and bias testing, while federal pre-emption ensures startups in Silicon Valley face the same rules as contractors in Virginia.
The second pillar—build infrastructure—acknowledges that breakthrough models are useless without the compute, silicon, and power to run them. It envisions secure "compute hubs" rising on federal land, powered by small modular reactors and geothermal plants, connected by hardened networks immune to foreign interference.
The third pillar—lead globally—pairs tougher export controls on advanced chips with an "AI-for-Allies" strategy that proactively shares full-stack American technology with trusted partners.
What makes this moment different is the convergence of political will and private sector enthusiasm. The tech industry, once allergic to defense work, now eagerly partners with the Pentagon. Meta reversed its prohibition on military use of Llama models. Anthropic announced a partnership with Palantir. OpenAI courts national security contracts. Venture capital, which used to flee from government contracts, poured $3.4 billion into defense tech last year alone. The same Silicon Valley that staged walkouts over Project Maven now races to build the next generation of national security tools.
This convergence reflects a growing recognition that the AI revolution won't wait for our bureaucratic comfort. When transformer architectures can be replicated as easily as stealing F-35 blueprints, and when the gap between GPT-4 and GPT-6 may be negligible due to asymptotic improvements, the old playbooks of technology control become not just outdated but dangerous.
The plan grapples with these uncomfortable truths head-on. It acknowledges that we can't lock down AI like nuclear technology—the math is already published, the talent is global, the genie has escaped the bottle. Instead of futile protectionism, it opts for aggressive competition: build faster, scale bigger, integrate deeper. If adversaries will inevitably access our innovations, our advantage must come from superior infrastructure, better integration, and the compound effects of our broader ecosystem.
Yet for all its ambition, the plan's success hinges on execution that has historically eluded Washington. Directives must become deployments within months, not years. The $32 billion in federal AI spending must flow to transformative projects, not legacy contractors. Agencies must actually adopt the unified framework rather than defending their regulatory turf. The infrastructure promised—from compute clusters to energy generation—must materialize before our competitors build their own.
The stakes extend beyond technological bragging rights. As AI becomes the substrate of economic and military power, second place means subordination. The leather-bound folder Trump waved might just be another Washington prop. Or it might be the document that future historians mark as the moment America stopped strangling its innovators with red tape and started giving them the green light they desperately need.
Ben Van Roo is the CEO and Co-Founder of Legion Intelligence.
Solid read. You write it Ben? Or did you prompt it?