From June 23 to 25, Annual Meeting of the New Champions 2026, also known as the Summer Davos 2026, opened in Dalian. More than 1,700 participants from over 90 countries and regions gathered in the coastal city for a series of high-level exchanges.

Compared with previous editions, the hottest keyword running through this year's forum was undoubtedly AI, or artificial intelligence. On the opening day alone, more than 10 sessions were directly related to AI. However, National Business Daily reporters (hereinafter referred to as NBD) at the venue observed that the focus of discussion had clearly shifted toward the practical difficulties and real-world challenges of AI implementation.

In the past, people marveled at leaps in AI computing power and breakthroughs in quantum computing. Today, as foundation models are no longer the main barrier and AI is penetrating manufacturing, transportation, healthcare and many aspects of daily life, new questions are being raised more frequently: How can AI be deployed at scale? How should AI evolve and be governed?

"The evolution of AI is not taking place in just one place. It needs to evolve in a coordinated way across regions and dimensions globally. At the same time, the biggest constraint on AI development is infrastructure, not intelligence," Roli Agrawal, Chief Strategy Officer of NTT DATA Inc., a Japanese IT services giant and one of the five core companies under Fortune Global 500 NTT Group, told NBD in an exclusive interview on June 23.

Summer Davos 2026 (Photo/Zhang Hong)

Look Beyond Large Models: "The Constraint on AI Development Is Infrastructure, Rather Than Intelligence"

On June 23, during a forum session on "AI Everywhere, Not at Once", speakers from around the world discussed the large-scale deployment, evolution and governance of AI from a macro perspective.

When asked about AI evolution, Agrawal said, "We should not be focusing on just one issue. We need to address several challenges simultaneously."

In her view, making AI work at scale involves far more than investing in AI itself. Data readiness, change management, infrastructure, AI sovereignty and privacy, as well as proper governance, are all issues that need to be addressed during AI's evolution. She emphasized the need to ensure that AI can evolve in a coordinated manner across the world.

"Think about your infrastructure, because the infrastructure that we have currently is built for digital era, not for AI," Agrawal told NBD, "It's like building a Formula One car, but when you get it out on the road, you realize that your roads are from 1880s. So how do you drive?"

Think about your infrastructure, because it will need a significant upgrade before AI can be deployed at scale. Small pilots are manageable with today's infrastructure, but scaling AI requires a fundamental infrastructure upgrade. When building AI-related infrastructure, the most important resources are computing and networks, Agrawal said.

On the computing side, Agrawal said it is necessary to assess the capacity and capability of data centers, and to clarify the distribution strategy between edge computing and cloud computing, including what should be deployed on-site at factories and what should be placed in the cloud. On the network side, data movement depends on network strength, which must feature low latency and high bandwidth. This is especially critical for autonomous AI.

"If a sensor detects an abnormality and a production line needs to be stopped immediately, extremely low network latency is indispensable," she said, citing a factory scenario. "New technologies such as end-to-end all-photonic networks, have already been deployed in multiple NTT DATA data centers. Therefore, infrastructure upgrades should cover both computing, including edge, cloud and local data centers, and networks."

Agrawal also noted that AI sovereignty and privacy are major issues, and that sovereignty and privacy boundaries need to be designed into the infrastructure from the very begining. In addition, governance and trust are also key factors affecting AI's adoption and scaling. Access rights and accountability, among other issues, are important, and AI governance needs to be coordinated globally.

"Innovation creates potential, execution delivers impact, and governance is what truly scales that impact," Agrawal said.

AI Becomes a New Arena of Major-Power Competition: Cost and Openness May Become Future Differentiators

Photo/NBD Media Asset Library

Artificial intelligence has become a key arena of major-power competition. Major economies such as the United States, Japan and the United Kingdom have all identified AI as a major strategic priority for enhancing national competitiveness and safeguarding national security. Global competition has moved beyond a singular race in computing power and model capability toward a broader contest involving ecosystems and applications.

In 2026, the AI race between China and the United States continues to intensify. Competition between the two countries has expanded from a single technological dimension to ecosystem building and global application.

"Both countries' AI capabilities are advancing rapidly. But I think, in technology, how these technologies are adopted globally will be governed and defined by sovereignty rules and geopolitics," Agrawal told NBD, drawing on her years of industry experience.

According to Stanford University's AI Index Report 2026, as of March 2026, the performance gap between top AI models in China and the United States had narrowed to about 2.7%, with the lead having changed hands several times since early 2025. Since 2025, China has gained advantages in several indicators, including the number of published papers, citation frequency, total patent output and industrial robot installations.

Beyond technical indicators, however, cost and openness are also becoming key variables reshaping the global AI landscape.

"For example, lower-cost models are particularly valuable for organizations and countries with budget constraints. In addition, open-source model can benefit the broader AI ecosystem by making innovation more accessible. Cost and openness could become important differentiators in the future," she said.

A research report by Orient Securities also noted that several Chinese model companies rank highly on global model performance leaderboards. Most of them remain open source, while their API, or application programming interface, usage costs are relatively low. This has enabled Chinese models to take leading positions on token distribution platforms such as OpenRouter. Microsoft is reportedly considering introducing DeepSeek V4 into its enterprise AI tool Copilot Cowork because DeepSeek V4 is open source, and its API usage cost is less than one twenty-fifth that of Anthropic Opus 4.8.

How Can AI Truly Benefit Enterprises on the Bottom Line? "Scaling AI Requires a 1-2-3-4 Rule"

Industries across the board are racing to embed AI into their operations. Yet in February this year, the U.S. National Bureau of Economic Research released a report based on a survey of more than 6,000 CEOs and executives in the United States, the United Kingdom, Germany and Australia. The report found that more than 80% of companies said AI had not produced a measurable impact on their development over the past three years.

How, then, can AI truly deliver benefits to enterprises on the bottom line?

Agrawal told NBD that companies should identify up to three core areas with real economic leverage, rather than attempting hundreds of use cases.

"For example, in the insurance industry, these areas could be claims and underwriting. In banking, they could be transaction monitoring and fraud detection. Companies should first identify these key areas, and then consider how to fundamentally re-architect them with AI in order to drive real value for business," she said.

In addition, companies need to estimate cost allocation across different stages.

"Scaling AI requires a 1-2-3-4 rule," Agrawal said, "If one dollar is invested in AI technology itself, companies should plan to invest two dollars for change management; three dollars for architecture design, including governance, token optimization, security guardrails and multiagent coordination; and four dollars for data readiness, to ensure that data is ready before AI applications are built."