How Washington responds, and whether its current posture matches the urgency its own intelligence community has outlined, came into sharp relief in May, as President Trump touched down in Beijing alongside Nvidia CEO Jensen Huang for a high-stakes summit that put AI and semiconductors at the center of United States-China diplomacy.
The three-month problem
According to some assessments, open-source Chinese AI models currently trail the most advanced American proprietary systems by roughly three to six months — DeepSeek’s own V4 release acknowledged that gap, which broader estimates from Epoch AI place closer to seven months. Yet that sliver of a lead is effectively meaningless from a military standpoint, because it takes not months but years for armed forces to absorb AI technology and translate it into a genuine battlefield advantage.
Carlos Perez, Director of Security Intelligence at TrustedSec, says the public numbers may not tell the full story.
“It is also important to recognize that China operates models that are not publicly available, so we have limited visibility into their true capabilities,” he tells The Cipher Brief. “Companies such as Alibaba and Tencent operate under significant government oversight and legal obligations tied to state investment and regulation. As a result, the actual capability gap may be smaller or larger than public comparisons suggest.”
China is not waiting for Washington to sort out its strategy.
The 2026 threat assessment describes Beijing’s approach as a coordinated, national-level strategy aimed at displacing the United States as the most influential AI power by 2030. Beijing is deploying autonomous drone programs at speed, integrating swarm intelligence into military doctrine, and leveraging a centralized governance architecture that allows civilian AI firms to be folded directly into People’s Liberation Army modernization efforts.
Alibaba and Baidu were both added to the Pentagon’s Chinese Military Companies list in February. DeepSeek has since become part of the People’s Liberation Army’s efforts to modernize its military healthcare infrastructure, according to analysis by the Foundation for Defense of Democracies. The House Select Committee on China concluded last year that the company was built, at least in part, on restricted American semiconductor chips and that its app functions as a direct pipeline for foreign intelligence collection on American users.
Aaron Estes, vice president of product management at Binary Defense and a former defense and intelligence official with 25 years of experience, tells The Cipher Brief the threat is more nuanced than a simple capability gap.
“The danger is that once these models are ‘good enough,’ the advantage shifts from model quality to speed of deployment, access to operational data, integration with command systems, and willingness to use AI in real-world workflows,” he explains. “A three-month gap in frontier model performance can disappear quickly if the other side is better at turning AI into operational tempo.”
Indeed, the real gap is not in the models. China can put its tech sector to work for its military tomorrow. The United States has to pass a bill first.
A policy vacuum in the middle of a race
On January 9, Defense Secretary Pete Hegseth signed the Department of War’s AI Acceleration Strategy — a document that declared speed wins, that the risks of moving too slowly now outweigh the risks of moving imperfectly, and that 2026 was the year the Pentagon would get serious about military AI dominance. Overlapping innovation offices were folded into a leaner CTO Action Group.
The document landed with fanfare. Yet, the budget did not follow. Washington keeps declaring AI a national security imperative while trimming the agencies and funding lines that would actually make it one — a contradiction the Bloomsbury Intelligence and Security Institute flagged in its recent read of the annual threat assessment.
Congress passed a more than 3,000-page National Defense Authorization Act that touched on AI across dozens of provisions — banning both the Defense Department and intelligence agencies from using DeepSeek and directing the Pentagon’s chief digital officer to build a department-wide AI assessment and procurement framework — yet defense analysts say it falls well short of the enforceable legal architecture the military needs.
Matthew Wein, a former policy advisor to the Department of Homeland Security, underscores that the White House’s March National Policy Framework for AI does not fill that gap.
“A strategy would be helpful to indicate where the government thinks we have an edge over competitors and how to maintain national security priorities as the technology landscape continues changing,” he tells The Cipher Brief. “It would also help the private sector understand how the government assesses China’s capabilities and give a blueprint for the frontier labs to use as a foundation to maintain their edge going forward.”
However, Leah Siskind, Director of Impact and AI Research Fellow at the Foundation for Defense of Democracies, argues that the threat itself is being mis-framed.
“There isn’t one AI threat, there’s a whole portfolio,” she tells The Cipher Brief. “Adversaries using AI for influence operations and cyber. Adversaries stealing or distilling American models. Adversaries racing us to military applications. A coherent strategy has to work all three at once, and right now Congress is moving faster than the executive branch on the first two.”
Siskind points to two bipartisan bills sitting on the floor — the AI OVERWATCH Act, which would codify congressional review of advanced chip sales to adversaries, and the Deterring American AI Model Theft Act, which creates a statutory framework to address Chinese model distillation attacks.
“The question is whether the White House, Commerce, and the AI czar’s office get behind them or fight them,” she continues.
The dispute between the Pentagon and Anthropic made that governance vacuum concrete. Anthropic was the first AI company to deploy frontier models on classified government networks, a position formalized by a $200 million DoD contract in July 2025. Talks broke down after the company refused to waive restrictions on autonomous weapons and mass domestic surveillance.
The Pentagon designated Anthropic a supply-chain risk—the first time that authority, historically reserved for foreign adversaries, had been applied to an American company—and Anthropic subsequently sued. As Siskind puts it: “Treating American frontier labs like Anthropic as national security supply chain risks is a strategic gift to Beijing.”
From experimentation to the edge
There is a basic problem that receives insufficient attention. Most military AI systems only work when connected to the internet. In a war, an enemy will cut that connection fast. OpenAI’s Pentagon contract actually reflects this limitation — its models can only be accessed through the cloud, meaning they cannot be built into weapons, sensors, or equipment troops use in the field.
The Army’s Project ARIA, Army Rapid Implementation of Artificial Intelligence, announced in March, explicitly aims to develop AI capabilities designed to function in denied, communications-degraded environments, but that effort remains in early development. Perez points out that AI is already deployed operationally on the intelligence and planning side, largely through platforms developed by companies such as Palantir, but stresses that the greater challenge is at the tactical level.
“Ukraine has demonstrated how AI can be integrated into autonomous drone operations and real-time battlefield adaptation, and this is an area where the U.S. is still learning and evolving rapidly,” he says.
Meanwhile, the Stanford AI Index, an annual comprehensive measurement of global AI progress published in April by Stanford University’s Institute for Human-Centered Artificial Intelligence, found that the United States ranks 24th globally in AI adoption at 28 percent. Singapore sits at 61. The UAE is at 54. Those figures reflect the civilian economy, not the military, but the pattern maps uncomfortably onto the Pentagon’s own trajectory.
Estes emphasizes that fixing it requires more than a strategy document.
“The Pentagon’s biggest obstacle is not a lack of AI technology. It is the gap between prototype and trusted operational use,” he continues. “Battlefield AI needs clean data, secure deployment environments, clear authorities, auditability, human accountability, and integration into existing systems that were never designed for this pace. That is where most AI programs get stuck.”
The Beijing question
Against that backdrop, President Trump’s arrival in Beijing in May was particularly important. He landed with a delegation that included Apple CEO Tim Cook, Tesla CEO Elon Musk, and Nvidia CEO Jensen Huang — the latter a last-minute addition whose presence signaled that semiconductors and AI were central to the agenda alongside tariffs, rare earths, Taiwan, and the Iran war.
Treasury Secretary Scott Bessent announced that Washington and Beijing would establish a protocol on AI best practices aimed specifically at preventing the most powerful models from falling into the hands of non-state actors. Pressed on why substantive dialogue with China was even possible, Bessent was candid.
“The reason we are able to have wholesome discussions with the Chinese on AI is because we are in the lead,” he responded. “I do not think we would be having the same discussions if they were this far ahead of us.”
How durable that lead is remains an open question. The Commerce Department has cleared roughly 10 Chinese companies — including Alibaba, Tencent, ByteDance, and JD.com — to purchase Nvidia’s H200 chips. However, no deliveries have been made: Beijing has quietly been steering its major tech platforms away from purchases to support domestic chipmakers instead.
China hawks on the Hill see the chip approvals as proof of exactly the problem they have been warning about — that Washington has no coherent line between competing commercially with Beijing and containing it militarily. The United States named AI the defining challenge of the era. It is now debating whether to sell China the chips that define it.
Wein says the path forward requires the government to be a better customer, not just a louder regulator.
“A more lax regulatory environment may be helpful for labs’ commercial growth, but a government strategy would also give them a useful framework from one of their largest customers, and position the United States as a standard setter in the space,” he asserts.
The cost of getting that wrong is not theoretical.
“The risk of relying too heavily on general-purpose commercial AI is that we end up with tools optimized for broad productivity, not contested military environments,” Estes adds. “Military AI has to work with incomplete data, deception, adversarial manipulation, degraded communications, classified context, and life-or-death accountability. What the United States actually needs are AI systems built from the ground up for the realities of warfare, not adapted from products designed for the consumer market.”
The Cipher Brief is committed to publishing a range of perspectives on national security issues submitted by deeply experienced national security professionals. Opinions expressed are those of the author and do not represent the views or opinions of The Cipher Brief.
Have a perspective to share based on your experience in the national security field? Send it to Editor@thecipherbrief.com for publication consideration.
Read more expert-driven national security insights, perspective and analysis in The Cipher Brief
