[Photo/VCG]
Artificial intelligence is at the core of digital transformation and is set to drive the next phase of productivity growth. However, AI has been politicized ever since the administration of former US president Joe Biden introduced the first export controls on United States-designed semiconductors in 2022.
Restrictions on Chinese 5G network equipment were imposed under the first Donald Trump administration in the name of protecting national security. This stance has since been codified into guidelines and laws of the European Union.
The US' export controls on AI technology are primarily driven by a raw attempt to contain China.
Now, the dominant focus on semiconductors and scaling compute as the primary path to better models and higher intelligence, which makes US hyperscalers pay around $200 billion each year, has been disrupted by China's AI startup DeepSeek, which, like its Chinese peers, only has limited access to advanced Nvidia chips for model training and inference.
DeepSeek was able to optimize compute efficiency across what Jensen Huang, founder and CEO of US semiconductor company Nvidia, defines as the three scaling laws — pre-training scaling, post-training scaling and test-time scaling — without relying solely on sheer computing power for each of them.
Even Europe's tech community, paralyzed by the increasing gap with the US' AI industry, has been inspired by DeepSeek's compute efficiency-driven approach.
The idea has been reinforced that catching up is still possible by focusing on software engineering rather than pouring billions into compute infrastructure.
Yet, Europe's awakening is driven not just by DeepSeek, but primarily by the current US administration turning its back on Europe and former prime minister of Italy Mario Draghi's devastating assessment of the EU's decreasing competitiveness.
The success of DeepSeek also challenges the assumption that strict regulations hinder breakthrough innovation. China enforces stringent regulations to ensure AI safety and security. Yet, its AI ecosystem continues to flourish.
Immediately after his inauguration, Trump revoked his predecessor's executive order on AI as a means of rejecting "overregulation", ensuring the US "develop AI systems that are free from ideological bias" and sustaining US AI leadership for the benefit of American workers.
This shift may still lead to some relaxation in digital regulations in both China and Europe. Europe has withdrawn its long-announced AI Liability Directive and has pledged regulatory "simplification".
Meanwhile, China will continue to pursue an issue-based, sector-focused approach.
In March, the Cyberspace Administration of China announced a new regulation, to take effect in September, requiring that all AI-generated content be clearly labeled to combat misinformation and promote transparency in digital media.
Fundamental challenge
The fundamental challenge for improving AI remains compute and efficiency.
The current foundation models have already hit a scaling wall, where exponentially more compute is required to achieve some improvements in performance.
Empirical research suggests that doubling compute typically results in only a 10 to 20-percent reduction in loss functions — illustrating the rapidly diminishing returns of brute-force scaling. This scaling frontier makes innovations in model architecture and algorithms indispensable.
Conversely, DeepSeek or the focus on compute efficiency will not reduce the need for powerful hardware, nor should it justify the weaponization of chips as "chokehold technology".
The next generation of AI models — capable of humanlike reasoning, acting as autonomous agents and tackling complex applications — will require significantly more computational power.
Nvidia's Huang estimates that computational demand for advanced AI systems has already increased by a factor of 100 within a year.
Global AI computing demand is projected to reach 864 ZFLOPS (a unit for measuring the speed of a computer system) by 2030, a 4000-fold increase over 2000 levels, underscoring the need for continued investment in high-performance computing infrastructure.
Tech covergence
Now, the emphasis on compute infrastructure highlights another crucial dimension of AI: next-generation network infrastructure, which underpins all digital transformation.
5G networks — and with 6G already in development — not only provide the necessary connectivity and bandwidth, but network operations and management functions are increasingly becoming AI-centric.
Without upgrades, telecommunications operators will have difficulty in managing exponentially growing traffic volumes and entire economies will be constrained in their ability to utilize AI's transformative potential.
This upgrading or 5G-AI convergence is being driven by: the personalization of user experiences and high data consumption, autonomous collaboration in organizations and network optimization, and the integration and operation of intelligent cyber-physical systems across all economic sectors, including energy, industry, cities and infrastructure, mobility, and agriculture.
Regarding the latter, for example, AI-driven cyberphysical systems, like smart grids or intelligent traffic management, are being implemented to efficiently integrate and manage large electrified systems, which are also the prerequisite for decarbonization.
Current networks won't be able to keep up with this development. Some conventional 5G networks are already stretched by network traffic waste. It is not caused by those new use cases but by algorithmic features, such as autoplay, infinite scroll, and pre-fetching of content, which are incentivized by the business model of social media platforms, as pointed out in British multinational telecommunications company Vodafone's recent policy statement.
While "responsible use of networks" is vital for addressing energy efficiency, the reality is that data traffic will continue to grow exponentially. This makes a shift towards AI-centric network operations and management not just sensible but indispensable to manage rapidly growing data efficiently and enable highly flexible, real-time, high-bandwidth use cases.
5G-AI convergence offers telecom operators a unique opportunity to accelerate 5G adoption by unlocking new use cases across various sectors. However, this shift requires substantial investments in network infrastructure, which add to the already high investments in computing infrastructure and cloud services.
As a result, while the AI-driven digital transformation offers tremendous productivity potential, it is also far more capital expenditure-intensive than the initial internet revolution.
Shared leadership
While the US still leads in frontier AI, China is closely following but is ahead in 5G and shaping 6G standards. China is also making significant strides in 5G-AI integration.
China's tech giant Huawei's 5G-Advanced is an AI-centric network platform, already deployed by more than 60 mobile operators globally, enabling the uplink velocity and low latency required for industrial upgrades to power AI-driven robotics and manufacturing.
That Huawei pursues such an infrastructure-first approach to AI is no coincidence. It reflects China's broader development model, which has always prioritized infrastructure as the primary driver of progress.
Unfortunately, 5G is at the center of geopolitical tensions, which not only delays its deployment and upgrades, but also hinders efforts to make networks more cybersecure, as the proper approach requires implementing policies, international standards and transparent practices.
The US' latest measure is the formation of a National Security Council to further reduce supply chain dependencies and win the strategic competition with China over 5G and 6G.
Digital sovereignty is today's defining policy strategy, where investment and governance decisions in digital transformation should benefit society and promote shared prosperity. It is not only about independent policy choices but building capacity and enhancing technological capability, without resorting to protectionism and coercion.
China is on course to become a global AI innovation center by 2030. Should Chinese AI companies continue pursuing an open-source strategy, China won't be the next AI hegemon but would foster a more inclusive form of dominance.
Through its AI Development and Safety Network and Global AI Governance Initiative, China is actively seeking collaboration on the safety and security of AI, further emphasizing its approach to shared leadership.
The Global South countries can benefit from such a "Digital Westphalia", build upon their own sovereign strategies and bridge their digital divides.
This would stand in stark contrast not only to today but also to history, when a few Western countries maintained a telecommunications monopoly throughout much of the 20th century, making it nearly impossible for poorer countries to break through such asymmetry — except for China.
As Chinese Foreign Minister Wang Yi put it, "Where there is blockade, there is breakthrough; where there is suppression, there is innovation".
After the split with the US, Europe now faces the same "mountain" and is finally more determined to find and navigate its own "mighty river to carve a path through it" to narrow its 80 percent technology dependency gap.
For the US, AI is the driver of the next productivity miracle, which it desperately needs but is still out of reach. Such domestic pressure suggests that significant breakthroughs and innovation are still to come.
The author is a senior fellow and Europe director at the Beijing-based think tank Taihe Institute. He is also a visiting professor at the Aerospace Information Research Institute of the Chinese Academy of Sciences and a fellow at Hertie School's Centre for Digital Governance in Berlin. The views do not necessarily reflect those of China Daily.
