Former Trump Official: The U.S. Can Win the AI Race If It Gets Patent Policy Right

The conversation around American AI leadership tends to center on the same familiar items semiconductor supply chains, energy infrastructure for data centers, talent pipelines from universities. These are real, and the money flowing into them is real. Just look at Jeff Bezos’s reported $100 billion AI manufacturing fund or Elon Musk’s Terafab chip plant announcement in Austin both representing enormous private bets on the physical foundations of AI dominance.

But Laura Peter, the former Deputy Under Secretary of Commerce for Intellectual Property and Deputy Director of the United States Patent and Trademark Office under the Trump administration, published a pointed argument in Fortune on March 29, 2026: none of that infrastructure spending matters if the underlying legal system for protecting AI innovation remains unstable. In her view, patent policy isn’t a sideshow to the AI race it’s the foundation everything else is built on.

The Structural Gap Nobody Is Talking About

Peter’s core argument is straightforward, even if the legal machinery behind it is anything but. AI leadership requires compute, talent, and capital agreed. But it also requires that companies can actually protect the things they build. Right now, the United States has a significant problem: the legal rules governing which AI inventions can be patented are inconsistent, unpredictable, and, in many cases, an active deterrent to investment.

The specific flashpoint is Section 101 of the Patent Act the clause that defines what kinds of inventions are eligible for patent protection. Since the Supreme Court’s 2014 ruling in Alice Corp. v. CLS Bank International, software and AI-related patents have been caught in a legal gray zone. Courts have interpreted the ruling expansively, blocking patents on inventions that rely on abstract ideas even when those ideas are implemented in genuinely novel ways in the real world.

The result is a split that patent attorneys describe as a “forum gap”: the US Patent and Trademark Office, under new Director John A. Squires (sworn in September 2025), has moved to approve more AI patents and issued clearer examiner guidance. But federal courts particularly the Federal Circuit continue to strike patents down under the same Section 101 standards that have been applied since Alice. A patent that passes through the USPTO can still get invalidated in litigation, making it nearly worthless as a commercial or financing asset.

What China and Europe Are Already Doing

What Peter’s argument makes clear and what gets lost in most US AI policy coverage is that America’s major competitors aren’t leaving this to chance. China doesn’t treat intellectual property law as a separate legal technicality. It folds IP objectives directly into its national AI plans, pairing patent development strategies with the enforcement capacity to back them up. When Beijing announces an AI initiative, patent targets are built in from the start.

Europe has taken a different approach but arrived at a similar place. The European Patent Office has issued structured, detailed guidance on AI patentability explicitly designed to produce predictable outcomes for software-based inventions that demonstrate what the EPO calls “technical effect.” If your AI invention solves a real-world technical problem not just an abstract data manipulation Europe can tell you in advance what the rules are. US applicants often can’t get that clarity from their own system.

This is the part that should catch every venture capitalist and enterprise technology investor’s attention. Patent protection isn’t just a legal formality it’s a signal that determines where capital flows. If two jurisdictions are both developing comparable AI technology but one offers clear, enforceable IP rights and the other doesn’t, money moves. That’s not a theoretical risk; it’s a real arbitrage that’s been happening quietly for years.

The Forum Gap: When USPTO and the Courts Disagree

The tension between the USPTO’s current direction and what courts actually enforce is one of the more underreported stories in American technology policy. Under Director Squires, the patent office has been actively working to protect AI innovation. In August 2025, Deputy Commissioner Charles Kim issued a detailed memo to patent examiners, instructing them to stop reflexively rejecting AI claims as “mental processes” and instead analyze claims as a whole specifically looking for whether they represent a technological improvement.

That December 2025 precedential decision in Ex parte Desjardins extended this shift further, explicitly warning that categorically blocking AI innovations from patent protection jeopardizes American leadership in a critical technology area. The USPTO’s Appeals Review Panel stated plainly that using Section 101 as a catch-all gatekeeping tool against AI is wrong.

But the Federal Circuit, which is the court that actually hears patent appeals, has moved differently. In April 2025, the circuit court ruled in Recentive Analytics, Inc. v. Fox Corp. that patents applying known machine learning methods to new data environments without disclosing actual improvements to the ML models themselves are still ineligible under Section 101. Calling something “AI” or “machine learning” in a patent claim doesn’t make it patent-worthy. The improvement has to be real, technical, and demonstrable.

The problem isn’t that either of these standards is obviously wrong. The problem is that they’re in tension, and companies trying to finance AI development can’t underwrite that uncertainty.

Three Things Washington Must Get Right

Peter’s op-ed isn’t just a complaint it’s a three-point agenda. And each point maps to a real mechanism Washington can actually use.

First: maintain and refine the USPTO’s current direction on AI patent examination. The guidance from Director Squires and the new precedents from the PTAB represent genuine progress. But examiner training needs to keep pace, and the standards need to be applied consistently across all technology centers, not just the software-focused ones. Predictability at the front end of the patent process reduces legal friction before inventions ever reach the commercialization stage.

Second: pass legislation to clarify Section 101. This is the hard one. Congress has known about the Section 101 problem for years. The Patent Eligibility Restoration Act of 2025 (PERA), which had a Senate Judiciary Subcommittee hearing in October 2025, would narrow the judge-made exceptions to patentability and create clearer guardrails. Supporters include former USPTO Directors Andrei Iancu and David Kappos. Critics including the Electronic Frontier Foundation warn the bill could revive low-quality patents and open the door to business method abuse. The tension is real. But doing nothing is itself a choice, and right now that choice benefits China.

Third: align IP incentives with strategic sectors. Congress is already advancing legislation on domestic manufacturing, energy infrastructure, supply chain resilience, and defense all areas where AI-enabled systems are increasingly central. But those efforts are disconnected from IP policy. Peter argues they should be integrated: if the US wants companies to develop and manufacture transformative AI technologies domestically rather than routing operations through IP-friendlier jurisdictions, the patent system needs to make that worth doing.

Applied AI Is Where the Stakes Are Highest

One of the more useful distinctions in Peter’s argument is between frontier AI the large language models and foundation models that get most of the press coverage and applied AI. Applied AI is the stuff embedded in industrial processes, energy grid management, medical diagnostics, logistics optimization, and defense systems. It’s less glamorous than a new GPT release, but it’s where the largest capital pools flow and where enforceable IP protection most directly shapes investment decisions.

This is where you find the manufacturing AI that companies are deploying to transform operations, the healthcare AI being integrated into diagnostic pipelines, and the industrial automation software running in factories. These are the systems where a company will spend tens of millions of dollars in R&D and then need patent protection to recoup that investment. If those patents can’t survive challenge, the calculus changes and so does the geography of where that R&D gets done.

What’s notable here is the downstream effect on US manufacturing competitiveness. The whole political and economic argument for domestic AI investment the one underpinning everything from semiconductor subsidies to Bezos’s AI manufacturing fund depends on companies having a reason to develop and scale their AI tools in the United States. Weak IP protection erodes that reason.

The Investment Signal Problem

There’s a deeper issue embedded in Peter’s argument that’s worth pulling out. Patent policy isn’t just about protecting specific inventions after the fact it sends signals before any investment is made. Investors assess whether a startup’s IP portfolio is defensible before committing Series A money. Large enterprises evaluate patent strength in jurisdictions when deciding where to locate R&D operations. Global technology firms weigh enforcement regimes when choosing where to scale manufacturing.

A predictable patent system says: invest here, innovate here, your returns will be protected here. An unpredictable one where the USPTO says one thing and the Federal Circuit says another, and you can’t reliably know which standard your patent will be held to in litigation sends the opposite signal. It may not stop investment in US AI entirely, but it shifts the risk calculus, and in a competitive global environment, marginal shifts in risk calculus have outsized effects on where capital accumulates.

This is, ultimately, what makes Peter’s argument more than just a policy wonk’s complaint about Section 101 jurisprudence. The global AI race as she frames it isn’t just a hardware and compute race. It’s a race to build the most credible, predictable environment for AI innovation at scale. Right now, the US is ahead on many dimensions but leaving points on the table on this one.

What Needs to Happen Next

The path forward has two tracks running simultaneously. On the administrative track, Director Squires at the USPTO is already moving in the right direction his guidance reforms, the precedential PTAB decisions, and the examiner retraining all represent meaningful changes within the limits of what an executive agency can do without new legislation. That work should continue and accelerate.

On the legislative track, Congress needs to close the gap that judicial interpretation has opened under Section 101. Whether PERA is the right vehicle is genuinely debated and the EFF’s concerns about abuse deserve serious consideration. But the status quo, where patents can be issued and then struck down in litigation based on doctrine that even courts struggle to apply consistently, is not a stable foundation for a national AI strategy that depends on private capital doing most of the heavy lifting.