SHANGHAI — From July 17 to 20, the Shanghai World Expo Exhibition and Convention Center hosted the ninth edition of the World Artificial Intelligence Conference (WAIC). The four-day summit brought together more than 1,100 exhibiting companies and official delegations from over 100 countries, spanning 100,000 square meters of exhibition space across Pudong and Xuhui.

Organized under the official theme "Intelligent Partners, Co-Creating the Future," the event closed with an estimated $3.01 billion (20.36 billion yuan) in intended procurement deals. Yet beyond the headline transaction figures, the conference highlighted two major shifts reshaping the global industry: coordinated international efforts to establish AI safety guardrails, and the rapid migration of multimodal models into physical hardware and industrial automation workflows.

Shaping Policy and International Governance

Regulatory frameworks dominated discussions at the High-Level Meeting on Global AI Governance, held alongside the main exhibition. In his opening address, UN Secretary-General António Guterres called for multilateral mechanisms to address widening technological disparities.

"Technology that will shape the future of humanity must be shaped by all of humanity," Guterres stated during the opening ceremony, describing AI as "humanity's greatest opportunity in the 21st century—or one of its greatest risks." He emphasized that global efforts must ensure the technology bridges economic divides rather than deepening them, adding: "AI governance cannot be governed by a handful of countries or companies. Every nation needs a seat at the table.""AI governance cannot be governed by a handful of countries or companies. Every nation needs a seat at the table."

Addressing the assembly, Chinese President Xi Jinping outlined a policy framework centered on human-centric development and shared security.

"The development of artificial intelligence should not be a solo performance by any single country, but rather a symphony of global cooperation," Xi told attendees. He urged participating nations to "together oppose the practice of overstretching the concept of national security in the field of artificial intelligence.""together oppose the practice of overstretching the concept of national security in the field of artificial intelligence."

To support capacity-building across developing nations, conference organizers announced 5,000 research and training quotas alongside new regional cooperation initiatives.

The summit also yielded concrete steps toward international coordination. Representatives from 29 nations signed an agreement establishing the World Artificial Intelligence Cooperation Organization (WAICO). Headquartered in Shanghai, the intergovernmental body is tasked with setting shared safety benchmarks, facilitating joint research, and helping developing economies access foundational AI resources. Commenting on the agreement, Serbian Minister Nenad Popovic called the creation of WAICO "the beginning of building a new system of international cooperation based on dialogue, mutual trust, and shared responsibility.""the beginning of building a new system of international cooperation based on dialogue, mutual trust, and shared responsibility."

Participating in the policy forums, Turing Award winner John Edward Hopcroft urged regulators to keep pace with rapid software iteration.

"AI is going to constantly change," Hopcroft noted in an interview during the event. "You have to have a strategy which will be resilient to changes. Governments have to regulate technology so that the good parts of the technology are available."

Embodied Systems and Hardware Integration

While earlier editions of WAIC focused heavily on software benchmarks and large language model parameter counts, the 2026 exhibition floor marked a clear transition toward embodied intelligence—the integration of advanced AI models directly into robotic hardware.

A wide array of service robotics, autonomous quadrupeds, and industrial automation platforms filled demo zones across the venue. Commercial service robotics provider Pudu Robotics hosted the offline debut of its PUDU D7 semi-humanoid robot. Demonstrating multi-robot task routing, the machine navigated complex booth environments while directing a PUDU D5 quadruped through simulated obstacle courses mimicking power-substation and industrial park inspection routines.

Elsewhere on the floor, industrial automation vendors displayed specialized humanoid arrays designed for factory assembly lines. Rather than demonstrating basic locomotion, live showcases highlighted tactile manipulation, component sorting, and real-time error correction powered by vision-language-action (VLA) models.

Software developers complemented these hardware displays with localized enterprise solutions. Exhibitors presented edge-computing implementations of multi-trillion-parameter open-weight models, enabling corporate clients to run complex reasoning and code generation locally while maintaining strict data privacy compliance.

Operational Bottlenecks and Market Barriers

Despite strong commercial interest, industry panelists and analysts pointed to persistent friction points confronting the widespread deployment of frontier technologies. Market reports published during the summit noted that while commercial adoption of embodied systems is accelerating in structured environments like warehousing, retail, and hospitality, broader industrial rollout faces practical hurdles.

Hardware durability remains a major constraint. High-frequency physical operation accelerates mechanical joint wear, creating heavy demand for field maintenance and specialized spare-parts supply chains that many early-stage developers have yet to build.

Integration costs present another hurdle. Non-tech enterprises frequently face friction when connecting proprietary databases to generative AI tools, resulting in prolonged pilot phases and delayed commercial scaling.

Energy and compute overhead continue to strain balance sheets. Training and hosting multi-trillion-parameter systems require immense capital and power infrastructure, putting pressure on developers to demonstrate clear margins and viable unit economics.

Finally, talent supply gaps persist. Reports released during the conference estimate a domestic talent deficit exceeding 5 million professionals, particularly for hybrid roles that require expertise in both mechanical engineering and human-machine interaction.

Balanced Outlook and Unresolved Questions

The 2026 World Artificial Intelligence Conference reflected an industry transitioning from early technical hype into a pragmatic operational phase. The focus has visibly shifted from generating conversational text to building durable hardware, securing enterprise retention, and negotiating cross-border safety protocols.

Yet key questions remain unanswered for global markets. Whether newly formed international bodies like WAICO can successfully align regulatory frameworks amid broader geopolitical tensions remains uncertain. On the commercial side, long-term leadership will depend less on parameter counts or conference deal totals, and more on whether developers can deliver reliable unit economics, manage soaring compute costs, and build machines that withstand the demands of everyday industrial work.