In the global race for artificial intelligence (AI), the United States has decided to tighten its strategy with export controls to limit China’s access to advanced chips. However, authoritative voices in the tech sector warn that this tactic might be not only insufficient but even counterproductive. Among them is Sam Altman, CEO of OpenAI, who recently told CNBC that: “No matter how many barriers are erected, China has shown it can advance, even if slowly, but surely.”
Altman was clear in questioning the effectiveness of these measures: “My instinct is that it doesn’t work. You can control the export of one thing, but it might not be the right approach. People can build factories or find shortcuts.”
The Biden administration has been blocking the sale of advanced Nvidia and AMD GPUs, such as the H100, H20, and MI308, to China for over two years. The official goal was to hinder large-scale AI model training capabilities, protecting Silicon Valley’s competitive edge. However, as Altman pointed out, the results seem to be quite different.
The most evident counterexample is DeepSeek, a language model developed in China without access to cutting-edge chips. While it doesn’t reach GPT-4 or Claude level, it has shown remarkable abilities in programming and logical reasoning tasks. This progress is clear evidence that China finds alternative solutions via algorithm optimization, parallel use of less powerful chips, or new architectural designs. The metaphor of “closing the barn door after the horse has bolted” applies here: restrictions may slow progress but cannot stop it. China has proven this in telecom sectors like 5G and solar energy, where despite bans, it has become the global leader.
Nvidia’s CEO Jensen Huang aligns with Altman’s view, arguing that export controls are a strategic mistake. Preventing China from accessing advanced chips does not slow its development but compels it to invest more in building its own industry, accelerating technological independence. Additionally, the US loses economic influence by ceasing sales to such a large market. Smuggling also plays a role: in the last quarter, over $1 billion worth of banned GPUs entered China, with resellers even offering models like the B300, not officially released yet.
One of China’s key advantages is its energy capacity. Training AI models consumes enormous amounts of electricity. While the US grid shows signs of strain with bottlenecks in Texas and Virginia, China benefits from a colossal, diversified energy infrastructure including coal, renewables, hydro, and expanding nuclear power. This means they can compensate for less efficient chips with sheer power, deploying more chips, consuming more electricity, and providing more data for training.
China’s long-term strategy aims for complete technological self-sufficiency by 2030, not only in chips but across the entire ecosystem. From domestic EDA tools for semiconductor design to intelligent computing centers and national manufacturers like Huawei and Empyrean Technology, each step may seem gradual but is steady and cumulative. Altman emphasizes that it’s not just who leads at a given moment but which country builds a sustainable foundation for leadership in the long run.
Meanwhile, Europe risks being left behind in this power struggle. The European Chips Act aims to revive the continent’s semiconductor industry, but progress is slow compared to US and Chinese investments. European companies remain dependent on external suppliers, positioning the region as a mere consumer rather than a leader in the so-called “technological sovereignty.”
Historically, this situation echoes the Cold War, when the space race symbolized the US-USSR rivalry. Today, the race is for AI dominance. Sanctions might delay China’s progress, but they are unlikely to stop it. As history shows, perseverance often prevails.
Some frequently asked questions include:
Why does Sam Altman believe export controls don’t work? Because alternatives always exist—building local factories, black markets, or choosing different architectures. Altman believes Chinese innovation will find ways to advance regardless.
How has China responded to restrictions? Through projects like DeepSeek, which demonstrates language model training without access to top-tier chips, and by massively boosting domestic hardware and software production.
What role does energy play in this technological race? It’s crucial. China has an enormous, diversified energy infrastructure that can sustain the high power demands of AI training, unlike the US, which faces increasing limitations on its electrical grid.
What does Nvidia think about US policies? CEO Jensen Huang believes that controls weaken the US economically and push China toward technological independence more quickly.
What risks does Europe face in this context? If Europe doesn’t accelerate investment in semiconductors and computing, it risks becoming a secondary actor—dependent on the US and Asia for critical technologies.
What is China’s long-term goal? To achieve full technological self-sufficiency in AI and semiconductors by 2030, solidifying its position as a global leader in these fields.
Source: Noticias Inteligencia Artificial