Agencies
New York
The geopolitical competition between the United States and China over AI technologies is intense and multifaceted. Both nations are vying for technological supremacy, which has significant economic and military power implications. Both countries also navigate the ethical and regulatory challenges associated with AI development. Competition is not just about technological advancement but also about shaping the future of global order. It’s a complex race, and the competition is fierce.
The US has traditionally been a leader in AI research and development. It has a strong foundation in semiconductor design, with an 85 percent global share in this market. China is heavily investing in AI and aims for technological self-sufficiency. The US is focusing on fostering rapid innovation and maintaining its technological edge and so it has implemented export controls on high-end chips to limit China’s access to advanced computing resources.
Despite being the largest semiconductor importer, China is developing its own AI chips and has made significant progress in areas like photonic chips, which are reportedly much faster than existing commercial AI chips. China’s strategy includes stockpiling AI chips and seeking domestic alternatives to mitigate the impact of US export bans.
The restrictions on high-end chips, particularly from the US, have made it difficult for Chinese companies to access the advanced computing resources needed for training AI models. As a result, Chinese venture capitalists are rushing to invest in AI modelling technologies aimed at developing independent software and hardware to support AI development.
There’s indeed a growing sense of anxiety among Chinese AI companies. The rapid advancements in AI technologies, particularly by Western companies like OpenAI with their new text-to-video model Sora, have intensified the pressure on Chinese firms to keep up.
Chinese companies are racing to develop their own cutting-edge AI models and reduce dependency on foreign technologies. This urgency is driven by the desire to secure leadership in the domestic market and demonstrate global technological prowess
The Chinese government has approved over forty LLMs and related AI applications in the last two years. Several other locally developed LLMs are flooding the Chinese market. As Chat-GPT is officially unavailable on the mainland, several start-ups like Moonshot AI and Baichuan are touting themselves as more accurate alternatives to OpenAI.
Nevertheless, there are doubts about whether China’s rush to create its own LLMs makes many prospects to support its indigenisation drive. One of the major challenges is its large-scale reliance on the US for the AI technology stack- hardware, software, data, and talent. The success of American LLM projects comes from the amount of quality data, access to the best hardware, huge capital ventures, and lastly vast potential scope for commodification.
In response to the increasing export controls imposed by the US, which aim to limit China’s access to advanced semiconductor technologies, Chinese AI companies are stockpiling US tech hardware in anticipation of further restrictions and to ensure they have enough resources to continue their AI development.
There were reports that Chinese entities have obtained advanced Nvidia GPUs for AI and HPC applications despite US restrictions through an advanced underground smuggling network from third-party countries. The US government is unhappy about this, as it has imposed strict export controls on high-end GPUs, which are crucial for training advanced AI models, from intermediary countries.
Chinese entities mainly used certain Middle Eastern countries, Malaysia, Singapore, and Taiwan, to get restricted GPUs, including Nvidia’s latest H200 GPU. Although Saudi Arabia and the United Arab Emirates faced restrictions, other countries in the region did not, so some of them were used to re-export advanced Nvidia GPUs to China.
The new regulations aim to close loopholes that allow Chinese companies to obtain these GPUs through third-party countries. The proposed regulations include national quotas on GPU exports and a global licensing system with reporting requirements. Facing AI GPU shortages for their AI development, these countries will unlikely tolerate re-exporting this hardware to China. This move has significantly impacted China’s AI development efforts, forcing them to seek alternative solutions and innovate with their limited resources.
The Netherlands and Japan remain in possession of some of the most advanced equipment for developing semiconductors. Until now, China has been able to capitalise on weak import-export controls, including via these US-friendly countries, as well as loopholes in the type of equipment banned under restrictions, to continue apace with the development of its domestic chip industry. If these loopholes and restrictions are tightened, China will have a harder time making the breakthroughs it needs to circumvent the overall bans on AI chips.
The US has also been implementing a “chokepoint strategy” to restrict China’s access to advanced technologies, including the open-source RISC-V architecture.
RISC-V, which stands for “Reduced Instruction Set Computing V,” is an open standard, meaning that anyone may use it as a building block in their open or proprietary products and services. RISC-V has attracted global attention and support due to its relative simplicity, the low barrier to entry, and its cost competitiveness.
US policymakers are concerned that Chinese companies could use RISC-V to reduce their dependence on US-controlled technologies and potentially develop advanced military and surveillance capabilities. As a result, there have been calls to restrict US firms from participating in RISC-V to prevent China from leveraging this architecture for its geopolitical interests.
US industry experts think that China’s RISC-V implementations don’t perform particularly well, not due to the architecture, but because getting implementation right is a long and complicated process. In addition, without access to leading node foundry technology, China will not catch up on end-product performance even if it has a comparable implementation.
The US has a distinct edge over China and the global lead in terms of maturity and the number of organisations that have “fully implemented” genAI technologies. While China may lead in genAI adoption rates, higher adoption doesn’t necessarily equate to effective implementation or better returns.
The gap between genAI usage and implementation reflects the varying depths of organisational AI maturity and integration. That competition is heated, if not already exploding into an all-out trade war, with the US significantly banning the export of various technologies — for example, chips — and even the use of cloud services in China. Meanwhile, OpenAI has banned the use of its models in China.
