After Chinese AI company DeepSeek’s surprise success rattled financial markets in late January, tech investors have largely turned their attention to other issues such as tariffs. But Silicon Valley isn’t out of the woods just yet.
U.S. AI companies are investing more than their Chinese counterparts and are well-positioned to maintain their performance lead. Still, that edge can be misleading. China’s advances in other tech sectors demonstrate that its companies can outcompete their U.S. counterparts even if they do not make the “best” products.
Leading U.S. AI companies’ business models rest on a central assumption: that developing the most capable models will produce the revenue required to justify massive training costs. In December, OpenAI released ChatGPT Pro, a $200 monthly subscription that offers unlimited access to the company’s top models. Later that month, it unveiled o3, a powerful, not-yet-released model that can cost over $1,000 per query.
DeepSeek complicates that core assumption. Since its debut, the company has released its V3 and R1 models, both of which achieve near leading-edge performance and are accessible at a small fraction of the cost of OpenAI’s offerings.
DeepSeek’s challenge to established U.S. AI companies is reminiscent of Beijing’s tried and true tech playbook. Chinese firms aim to develop low-cost products that are “good enough” to outcompete rivals.
Huawei is a prime example. The company got its start by reverse engineering foreign information and communications technology products and building telecommunication networks in remote parts of China. In the late 1990s and early 2000s, Huawei expanded overseas by offering low-cost products tailored to the needs of less developed markets. And while the company might have originally succeeded by competing on cost, today, Huawei’s technologies are highly advanced and easily deployable, but still affordable. As a result, they have been widely adopted. In Africa, 380 million people rely on Huawei-built ICT networks, the company’s smartphone shipments grew by nearly 40% in 2024, and it has built data centers around the world.
We see similar trends in other advanced technology sectors. Chinese electric vehicles now outsell their Japanese, German, and U.S. competitors. Chinese drone-maker DJI controls over 90% of the global consumer drone market due to its products’ high quality and low prices. Today, the per-token cost of querying DeepSeek’s R1 model is less than 4% of the price of using OpenAI’s o1 model, despite scoring similarly on various AI benchmarks.
Since R1’s release, some analysts have argued that leading U.S. AI companies’ compute advantage—that is, their ability to use vast quantities of advanced semiconductors to train leading large language models—will allow them to race ahead of their chip-constrained Chinese challengers. Recently announced U.S. investments in AI infrastructure include up to shows that the performance of open-source models generally trails leading closed-source models by less than a year. Recently released Chinese open-source models, including DeepSeek’s R1 and Alibaba’s Qwen 2.5, trail their U.S. competitors by only a few months.
U.S. AI titans are betting that maintaining their lead in top performing AI models will be the decisive factor in the AI competition. They claim that successive generations of LLMs trained with ever-growing computational resources will ensure U.S. leadership in AI. If they are correct, Chinese competitors may not be as threatening as they now appear.
But if Chinese low-cost, easily deployable models are widely adopted, U.S. companies may find it more difficult to make money. Should Chinese AI labs continue to stay near the technological frontier by capitalizing on algorithmic improvements, smuggled AI chips, and model distillation techniques, then users may prefer their good enough, open models to U.S. companies’ expensive, closed ones. In short, AI adoption could depend more on cost than capabilities.
This presents a serious threat to the U.S. AI ecosystem, which has largely staked its future on expensive-to-train, proprietary, and highly capable models. If U.S. big tech companies cannot secure significant returns from their investments in LLMs, it is unclear how they will justify spending hundreds of billions on new AI infrastructure. Chinese companies, on the other hand, have long prioritized the adoption of their technologies over maximizing profits. They appear to be pursuing a similar strategy regarding LLMs.
To be sure, U.S. AI companies must prioritize technological innovation. But AI competition doesn’t exist only at the frontier. To stay competitive in this rapidly evolving environment, Washington and Silicon Valley must find ways to push AI adoption in order to guarantee U.S. technological leadership going forward.
About the authors: Sam Bresnick is a research fellow at Georgetown University’s Center for Security and Emerging Technology. Cole McFaul is a research analyst at CSET.
Guest commentaries like this one are written by authors outside the Barron’s newsroom. They reflect the perspective and opinions of the authors. Submit feedback and commentary pitches to ideas@barrons.com.
