Global AI governance "dual-track system" begins: industry implications of WAIC 2026 and WAICO

Global AI Governance's "Dual-Track" System Begins: The Industrial Implications of WAIC 2026 and WAICO

On July 16, 2026, one day before the opening of the World Artificial Intelligence Conference (WAIC) in Shanghai, 29 countries signed the founding agreement of the World Artificial Intelligence Cooperation Organization (WAICO). This China-led intergovernmental organization, headquartered in Shanghai, counts Russia, Cuba, Brazil, Venezuela, Pakistan, Indonesia, and others among its founding members—yet no major Western democracy is included. Chinese President Xi Jinping delivered his first speech at WAIC, calling WAICO "an important milestone in the history of AI development," and pledged 5,000 AI training opportunities for developing countries. Meanwhile, more than 300 products had their global debuts at the exhibition, with Huawei's Atlas 950 SuperPoD and Nubia's world-first AI Agent smartphone taking center stage.

For corporate decision-makers, investment institutions, and policy researchers, the signals stacked up by these events are clear: the global AI industry is forming two competing sets of rule systems—one centered on Brussels, Washington, and the Group of Seven (G7), and the other with Shanghai as its hub. Neither system currently makes technology access a condition for joining, but future standards interoperability and compliance costs will become real challenges that multinational companies must face.

Industry Background: From Proposal to Signing—Why WAICO Is Landing Now

WAICO is not an unexpected product of this year's conference. As early as WAIC 2025, Chinese Premier Li Qiang proposed the idea, after which China spent a year securing commitments from various countries. The founding agreement signed this time designates Shanghai as the permanent headquarters, with foreign ministers from 29 countries signing the charter, marking the organization's transition from an initiative to an entity. UN Secretary-General António Guterres attended the signing ceremony, granting the new body symbolic support within the UN system, though the UN has not formally joined.

Western critics argue that WAICO's membership structure deliberately excludes Western democracies, aiming to give China AI standards influence across the Global South that exceeds its actual standing. France has been the most outspoken, viewing it as an attempt to undermine the Hiroshima Process. Indian officials and analysts have also cautioned that WAICO could shape a new global AI order, and that democracies should be vigilant about its governance implications.

From an industrial perspective, however, WAICO's emergence reflects a genuine "representation deficit" in global AI governance. Current AI governance frameworks—such as the EU's Artificial Intelligence Act, OECD AI Principles, and the G7 Hiroshima Process—are mainly formulated by Western economies, leaving Global South countries with limited voice in rule-making. By offering an alternative through WAICO, China attracts countries unwilling to fully accept Western frameworks.

Market Impact: Dual-Track Compliance and Industrial CostsThe establishment of WAICO means that countries joining the framework will gradually form AI regulatory rules that are not fully compatible with the EU AI Act, OECD principles, and the Hiroshima Process. For any enterprise operating globally, this will directly lead to rising compliance costs: in the future, AI products may need to meet two sets of standards simultaneously and adopt differentiated compliance arrangements in different markets. For example, the two systems may make completely different provisions regarding cross-border data flows, model auditing, transparency requirements, and computing power export controls.

At the computing infrastructure level, the physical debut of Huawei's Atlas 950 SuperPoD is significant. The system consists of 8,192 Ascend 950DT processors, and through Huawei's self-developed UnifiedBus 2.0 protocol, it implements a unified memory pool, allowing all chips to work like a single logical machine. Huawei claims its FP8 computing power reaches 8 exaflops, with interconnect bandwidth of 16 PB/s, and asserts it provides 6.7 times the computing capacity of NVIDIA NVL144. It should be noted that these figures have not been independently verified by third parties. However, DeepSeek has confirmed that its models are running at production scale on Ascend clusters. This factual signal indicates that the gap between Huawei's software ecosystem (CANN) and NVIDIA CUDA is narrowing, though not yet fully closed.

On the consumer side, Nubia's "world's first AI Agent phone" runs StepFun's Agent OS, which uses a GUI agent to observe the screen and execute multi-step tasks across applications, such as automatically booking a flight and adding a calendar event. However, the previous generation product, the Nubia M153, was blocked by WeChat and Taobao from its automated UI interactions, because screen observation may capture content that platforms consider sensitive or proprietary, including messages displayed in decrypted form. Therefore, whether this product can obtain ecosystem access agreements from major application operators is key to its commercialization.

Competitive Landscape: Who Benefits, Who Feels Pressure, and Who Will Follow

From a governance perspective, the direct beneficiaries of WAICO are China and non-Western countries aligned with it. They gain a new platform for discourse in international AI rule-making and may use it to reduce their dependence on Western technical standards. Those under pressure are the EU and the G7—their dominant governance frameworks face the challenge of "another option," and in the future, they will need to be more attractive in engaging Global South countries.

In the computing power sector, the maturation of Huawei's Ascend ecosystem directly erodes NVIDIA's market share in China and the Global South. Although CUDA's developer ecosystem advantage remains enormous, DeepSeek's production-grade use on Ascend shows that alternatives are already viable for specific workloads. In addition, WAIC showcased more than ten domestic GPU architectures on site, including Sugon, Alibaba, and ZTE, marking the largest concentrated display of non-NVIDIA AI infrastructure to date.In the AI phone and Agent domain, Nubia is collaborating with ByteDance, attempting to change human-computer interaction through a system-level Agent. If successful, it will disrupt the existing app distribution model and weaken super apps' control over user entry points. However, the blocking actions by WeChat and Taobao show that platform players will not easily give up control over sensitive data. This contest will determine whether AI Agent phones become a new entry point or remain merely at the experimental stage.

Corporate Insights: Strategic Preparation for a Dual-Track World

For multinational corporations, the industry information conveyed at this WAIC must be factored into strategic considerations:

First, AI compliance must shift from a "single-track" to a "dual-track" mindset. When formulating global launch strategies for AI products, companies need to assess the regulatory differences between WAICO member states and member states of Western frameworks, set up modular compliance systems in advance, and avoid market access barriers caused by conflicts between standards.

Second, compute procurement strategies need to be diversified. Huawei's Atlas 950 demonstrates the ability to build large-scale AI clusters without relying on US components. Although performance data has yet to be independently verified, DeepSeek's actual deployment proves that this is not just an empty promise. Companies should evaluate the feasibility and cost of running critical workloads on non-US compute ecosystems.

Third, the interaction revolution brought by AI Agents will affect enterprise application development. If terminals such as phones directly control third-party apps through a system-level agent, companies will need to choose between being "proxied" and "integrating proactively." At the same time, managing relationships with platforms such as WeChat will become a key variable.

Future Outlook: 12 Months, 24 Months, 3 Years

In the next 12 months, WAICO will enter the phase of defining operational details, including the establishment of a secretariat, standard-setting processes, and the launch of the first batch of substantive working groups. More Global South countries may join in succession, while Western countries may attempt to consolidate their camp by strengthening the existing Hiroshima Process.

In the next 24 months, Huawei's Ascend ecosystem will continue to narrow the gap with CUDA. As leading models such as DeepSeek continue to be optimized on Ascend, more production-grade migration cases are expected to emerge. On the AI Agent phone front, whether more terminal manufacturers will follow suit or platform players will form open standards will become a key point to watch.

In the next 3 years, the global AI industry will most likely form two sets of technical standards and compliance systems that compete with each other but are not completely isolated. Companies need to build "dual-mode" capabilities that can adapt to both systems simultaneously. In the long run, this dual-track landscape may force international organizations (such as the United Nations) to attempt mediation, but at least in the medium term, fragmentation will dominate the evolution of the AI industry.

WAIC 2026 is not only a product exhibition, but also a milestone in the global AI industry landscape moving toward divergence. For decision-makers focused on AI industry trends, understanding the driving factors behind this divergence is far more strategically valuable than chasing certain product specifications.