The Battle for LLM Dominance

On 30 November 2022, the California‑based technology firm OpenAI announced the launch of a new chatbot. In a concise six‑sentence release, OpenAI stated that it had trained a new model “which interacts in a conversational way”.

The product received the name ChatGPT. The technology community reacted with immediate fascination.

“You could go on any sort of social network and there was just this flood of posts from people talking about all the different ways that they were using this new little text box that had appeared on the internet,” explains Bloomberg columnist Parmy Olson, author of Supremacy: AI, ChatGPT, and the Race That Will Change the World.

ChatGPT represented the first mainstream large language model, or LLM. An LLM processes massive quantities of existing online text and data, learning statistical patterns that govern how ideas are expressed.

Experts now broadly agree that, regarding AI “brains”, the United States enjoys the upper hand. OpenAI reports that more than 900 million people use ChatGPT each week, a reach that approaches one in eight people worldwide. Other United States technology companies—Anthropic, Google, and Perplexity—have invested billions of United States dollars to develop competing LLM systems.

The strategic calculus is clear: a successful LLM can automate many tasks traditionally performed by white‑collar professionals, turning technical superiority into substantial commercial revenue.

How the United States Played Its Chips

Washington also focuses on a parallel question: how will the AI contest influence the United States‑China struggle for global primacy?

A senior United States official, speaking to the Global Research and Economic Exchange (GREE), emphasized that the United States’ strategic advantage lies less in sophisticated algorithms and more in the hardware that powers massive computing clusters—specifically microchips.

Most high‑end, high‑performance computer chips used by Silicon Valley firms to train LLMs are controlled by the United States. The design of the majority of these chips originates from the California‑based company Nvidia. In October, Nvidia became the first company in the world to achieve a market valuation of five trillion United States dollars, a milestone that Stephen Witt, author of The Thinking Machine, suggests could make Nvidia the most valuable enterprise ever.

The United States enforces a strict network of export controls to prevent China from acquiring these powerful chips. The policy traces its roots to the 1950s, when the United States blocked advanced electronics exports to Soviet‑aligned states, and it was sharply reinforced in 2022 by President Joe Biden as the AI race intensified.

Although many of the chips are manufactured outside the United States—particularly in Taiwan by Taiwan Semiconductor Manufacturing Corporation (TSMC)—the United States leverages its “foreign direct product rule” to require foreign manufacturers to comply with United States regulations when their products contain United States‑origin components or derive from United States technology.

The Taiwanese semiconductor facilities sit conspicuously close to mainland China, rendering them an attractive target for Beijing.

Creating high‑end chips, however, demands specialized equipment such as extreme ultraviolet (EUV) lithography machines. Only a single company worldwide, ASML of the Netherlands, produces these machines. The United States applies the same “foreign direct product rule” to block ASML from exporting EUV systems to China.

This protectionist approach has, until recently, helped the United States preserve its lead in AI “brains”.

The DeepSeek Counter‑Attack

In January 2025, coinciding with the inauguration of Donald Trump for a second term, China introduced its own AI‑powered chatbot named DeepSeek.

From a user’s perspective, DeepSeek operates much like ChatGPT: it answers queries, writes code, and is offered free of charge.

Crucially, DeepSeek reportedly required only a fraction of the computational resources expended to develop United States LLMs such as ChatGPT and Claude.

The launch generated a seismic market reaction. On 27 January 2025, Nvidia experienced the largest single‑day loss in United States stock market history, with its market value dropping roughly six hundred billion United States dollars.

“It was hugely discombobulating for Washington,” remarks AI journalist Karen Hao. She suggests that United States export controls may have unintentionally spurred Chinese developers to innovate with limited hardware, accelerating China’s self‑reliance.

“The defining feature of DeepSeek is that it had similar capabilities, at the time, to the United States models such as OpenAI and Anthropic, but using a far smaller amount of computer chips for training that model,” observes Karen Hao.

In Beijing, optimism surged, according to Selina Xu, a researcher on China AI policy working in the office of former Google chief Eric Schmidt. “Everybody was trying to figure out, ‘How did DeepSeek do it?’” Selina Xu notes, adding that the event served as a powerful catalyst for the Chinese AI ecosystem.

The episode also highlighted a fundamental difference in development philosophies. United States AI firms fiercely guard intellectual property, whereas Chinese firms often adopt an “open source” stance, publishing code online to enable developers from other companies to build upon existing work.

“This means that tech companies in China, when they’re building a new AI model, don’t have to start from scratch,” explains Parmy Olson. “They can just take that model and build on top of it and make it better.”

Consequently, the clarity of the United States advantage in AI “brains” has blurred. While United States proprietary models may retain a performance edge, China has demonstrated the ability to produce comparable LLMs at dramatically lower cost.

“The United States closed‑proprietary models are probably better, but maybe just not by that much,” observes Selina Xu. “The Chinese model might be ninety percent as good, but it costs ten percent as much.”

China’s Advantage in the Robot Wars

When evaluating AI “bodies”—the realms of drones and robotics—China has historically possessed the lead.

Since the 2010s, the Chinese government has markedly increased support for robot development, providing research funding and billions of United States dollars in subsidies to robot manufacturers.

Current estimates suggest that China operates approximately two million industrial robots, surpassing the total number of robots deployed in the rest of the world combined.

Olson attributes much of this success to China’s status as a manufacturing powerhouse. “So you have all that expertise on building electronics and you capitalise on that and then you get incredible robotics start‑ups,” she says.

International visitors to Shenzhen or Shanghai frequently note the deep integration of robots into everyday activities, including drone‑based food delivery, according to Selina Xu.

China has especially excelled in “humanoid” robots—machines designed to resemble and move like humans.

The Center for Strategic and International Studies, a bipartisan United States think tank, has reported on a “dark factory” in Chongqing, southern China. The facility houses two thousand robots and autonomous vehicles that together can assemble a new automobile every minute. The term “dark factory” reflects its fully automated, human‑free operation.

Beijing recognizes the country’s rapidly ageing population. Selina Xu explains that the Chinese government envisions humanoid robots filling gaps left by retiring workers, particularly in elder‑care roles. By around 2035, the number of Chinese citizens aged sixty or older is projected to exceed the entire population of the United States.

China not only builds robots for its own massive domestic market; it also accounts for roughly ninety percent of global humanoid robot exports.

The Ghost in the Machine

Despite China’s supremacy in constructing robot bodies, each robot still requires an intelligent “brain”—software that controls its actions.

For robots performing simple, repetitive tasks—such as the automated assembly line in Chongqing—the necessary brain can be relatively modest and can be produced domestically by China.

However, robots tasked with diverse, complex operations need an advanced form of AI known as agentic AI. Agentic AI programs operate more like independent actors, handling multi‑step assignments and making decisions without constant human oversight.

In the realm of high‑performance robot brains, the United States continues to maintain the advantage.

“The United States is definitely still in the lead when it comes to robot brains,” asserts Nick Wright, the cognitive neuroscience scholar at University College London. “That’s the chips and the AI software that helps the robot do actual tasks. And what you do need to bear in mind is that about eighty percent of the value of a robot is in its brain.”

Of Robot Dogs and Drones

Both the United States and China are now racing to integrate robots with agentic AI. A United States firm has already demonstrated that it no longer holds a monopoly on delivering successful autonomous machines.

Boston Dynamics, a United States engineering company, employs agentic AI in its quadruped robot Spot. Spot has become an internet sensation, amassing millions of YouTube views. The robot incorporates advanced vision (“eyes”) and acoustic sensing (“ears”), enabling it to conduct inspections inside warehouses, detecting overheating equipment, gas leaks, or spills.

Spot’s sensor data is transmitted to industrial AI software provider IFS, where AI analyzes the findings and may autonomously execute corrective actions without human intervention.

On a more ominous front, Wright points to battlefield drones as an existing illustration of robotics merged with agentic AI. Last summer, Ukraine deployed the Gogol‑M aerial “mothership” drone, capable of traveling hundreds of kilometres into Russian territory before releasing two smaller attack drones. Those attack drones, operating without direct human control, employed AI to scan terrain, identify targets, and navigate toward them before detonating.

Who Will Triumph?

Predicting the victor of this contest is challenging because the finish line remains undefined, notes Greg Slabaugh, professor of computer vision and AI at Queen Mary University of London.

“Victory is unlikely to be a singular moment, like landing on the Moon,” he explains. “Instead, what matters is sustained advantage: who leads in capability, who embeds AI most effectively across their economy, and who sets global standards.”

Historically, technologies such as electricity and computing proved less about who built the first system and more about who deployed it most widely and efficiently across the economy. Prof Slabaugh argues that the same dynamic may apply to AI.

The United States’ major tech corporations appear eager to rush into an uncertain future with minimal regulatory guardrails, while the Communist Party of China emphasizes state oversight of AI research.

One vision foresees a hyper‑consumerist capitalism driven by United States firms; the alternative foresees a state‑directed model where the government determines permissible uses of the technology.

“Each side is best placed to prevail in its own game,” states Mari Sako of the University of Oxford’s Said Business School. “When two players fight with different rules of the game, I suspect the player that courts the wider audience—users, adopters, etc.—is likely to prevail.”

The stakes are immense. It remains unclear which nation will emerge more powerful in the twenty‑first century, and the AI race could become the decisive factor.