At the White House on Tuesday, President Trump, Speaker Mike Johnson and leaders of America's major AI companies emerged with a voluntary accord calling for robust internal controls, independent auditing and safeguards against AI systems accessing technical systems in unintended ways. The agreement is limited and its implementation will matter. It also offers a useful example of government bringing fierce competitors together around a concrete problem while leaving competition intact.
That example matters because much of the federal machinery is already in motion. The White House AI Action Plan spans more than 90 federal actions across innovation, infrastructure and international policy. The American AI Exports Program is organizing full-stack offerings that include chips, cloud infrastructure, models, cybersecurity and applications. A June executive order created an AI cybersecurity clearinghouse and a voluntary framework for frontier-model security while expressly declining to authorize mandatory federal licensing, preclearance or permitting. This week, America.gov put advanced AI directly into federal service delivery, while the new U.S.- China Super Intelligence Dialogue created a bilateral channel for incidents related to advanced AI. Taken together, these efforts show how much activity is already underway.
The phrase “AI race” is too singular for the competition now unfolding. The United States and China are competing across at least seven fronts, and leadership in the most capable model does not guarantee leadership across the other six.
Frontier intelligence. The most obvious contest is over who develops the strongest systems in reasoning, coding, science, agents and robotics. Frontier capability still matters enormously because breakthroughs at the top eventually propagate through products, research and the wider economy.
Capacity. AI is digital, while its limits are increasingly physical. Chips, advanced packaging, networking, data centers, electricity generation, transmission, interconnection, cooling and skilled labor all determine how much intelligence can actually be trained and deployed. The administration already treats infrastructure as a core condition of AI leadership, and its June national-security directive also calls for next-generation high-security computing capacity. National-security AI directive.
Industrial AI. The strategic question reaches far beyond model benchmarks. It includes how rapidly intelligence moves into factories, robotics, logistics, healthcare, energy systems and supply chains. A country can extract enormous economic and strategic value from capable AI if it deploys that intelligence broadly and compounds the productivity gains across its industrial base.
Global diffusion. The technology stack that becomes easiest to buy, finance, integrate and operate can shape markets and dependencies for years. America’s export program recognizes this as a full stack competition. China is pursuing another form of diffusion through inexpensive and open models. The U.S.-China Economic and Security Review Commission reported this year that Alibaba’s Qwen ecosystem had produced more than 100,000 derivatives on Hugging Face, creating a feedback loop in which adoption drives iteration and iteration drives further adoption. USCC analysis.
National security and control. Advanced AI increasingly touches cybersecurity, intelligence, critical infrastructure, autonomous systems and the security of the models themselves. Tuesday’s White House accord sits squarely in this lane, as does the June security order. The emerging challenge is to build credible safeguards while preserving the competitive pressure and speed that drive technical progress.
Productive power at home. The country that invents the most capable AI and the country that absorbs AI most effectively throughout its economy may end up capturing very different amounts of value. Productivity depends on deployment across businesses, workers, science, healthcare and government. America.gov is a visible example of the federal government putting AI directly into service delivery, alongside private-sector adoption.
The information layer. AI is becoming an intermediary between people and knowledge. Increasingly, users will ask a model a question and receive an answer instead of navigating dozens of links. Accuracy, provenance, language coverage, resilience to manipulation and trust therefore carry strategic significance well beyond conventional software competition.
These fronts interact constantly, which is where the current architecture becomes harder to read. A decision about advanced-chip exports can affect national security, the economics of American semiconductor companies, the cost of frontier development, the attractiveness of the U.S. stack abroad and the incentive for other countries to build around Chinese alternatives. A rule written for frontier safety can also change fixed compliance costs and market structure. A transmission bottleneck can delay a data center, constrain compute, slow industrial adoption and alter the economics of exporting American AI. Each policy lane has its own logic, but the competitive effects travel across lanes.
The missing capability is a better operating picture across the machinery already in place. Three questions would make the existing architecture much more legible. The first is whether we have a real scoreboard across all seven fronts. GAO has already built a framework for assessing U.S. AI competitiveness across science and technology, human capital, governance and the economy. The next step is a current view of frontier capability, compute, energy, cost, industrial adoption, global deployment and productivity, so policymakers can see where America is actually gaining or losing ground.
The second question is what a major decision on one front does to the other six. That discipline would make tradeoffs visible before they become unintended consequences. Export controls, open-weight policy, safety standards, data-center rules and federal procurement can all improve one objective while changing incentives somewhere else in the system.
The third question is ownership when a strategic bottleneck crosses institutional lines. Global deployment can involve Commerce, State, financing agencies, cybersecurity authorities and private companies at the same time. Infrastructure can involve Energy, permitting agencies, states, utilities and capital providers. Many institutions can perform their assigned tasks well while an outcome that spans them remains nobody’s job to deliver.
Tuesday’s White House meeting offers one possible operating habit. A specific problem was defined, the companies with the relevant capability were brought together, responsibilities were made more explicit, and competition remained intact. The accord will ultimately be judged by what the companies implement. The method is worth applying more broadly when problems fall between established lanes.
China’s centralized system gives Beijing a natural view across infrastructure, industrial policy and national strategy. America is built around distributed power, private capital, competing firms, universities and experimentation, an architecture that has produced extraordinary technological dynamism. The work now is to add visibility across the system while preserving the forces that make it move quickly.
Seven fronts are moving at once, and every major decision changes the position on more than one of them. Seeing those interactions early would help the United States compound its strengths, expose tradeoffs before they become expensive, and clear bottlenecks while they are still solvable. America already has extraordinary companies, capital, universities, infrastructure programs and diplomatic tools. The advantage now is to see them as one strategic picture and move at AI speed.
