On May 13, the National Artificial Intelligence Industry Development Conference opened in Beijing. The theme is eight characters: “AI empowers new quality productive forces, all-domain integration into a new ecosystem.” This conference is one of the highest-level events in the domestic AI field this year, with institutions across the upstream and downstream of the industrial chain almost all present.
On the same day, the Ministry of Industry and Information Technology published on its official website the Special Action Implementation Plan for "AI + Manufacturing," jointly issued by eight departments. These two documents and one meeting basically set the tone for enterprise AI implementation in the second half of 2026.
But outside the venue, the real situation facing enterprises is far more complicated.
The meeting outlined three main topics
Based on public information and brokerage interpretations, the signals released by this meeting can be summarized into three main lines:
Article 1: Policy shifts from “directional encouragement” to “execution-driven advancement.” The meeting emphasized three key areas: technological breakthroughs, industrial upgrading, and improvements in people's livelihoods; the core goal is to clear the “last mile” of AI implementation. Localities simultaneously issued supporting policies. This is essentially telling local governments and enterprises: stop waiting and watching—it is time to act.
Article 2: Computing power infrastructure is moving from “adequate” to “easy to use.” The conference showcased high-density computing clusters, energy-efficient computing solutions, and a new generation of AI hardware. The direction is clear: high performance, low power consumption, and intensification. Behind this lies the continued expansion of demand for large-model training and inference, as well as the dual pressure of global green and low-carbon development.
Article 3: Vertical industry integration is moving from single-point pilots to full-process penetration. Exhibition areas were set up specifically for industrial manufacturing, healthcare, the low-altitude economy, and smart cities. Unlike a year ago, this year’s discussions in the exhibition areas are no longer about “whether AI can be used,” but about “how to embed AI into end-to-end business processes.”
At the meeting, several securities firms gave an almost identical assessment: the AI industry is transitioning from “technology R&D” to “value creation,” among which “AI+” integrated applications, computing power upgrades, and vertical deployment are the three core investment tracks.
Manufacturing: "Accelerating both ends" is a reality, not a slogan
While policy signals are being sent, the AI implementation in manufacturing is indeed accelerating. However, the way it accelerates is different from what many people imagine.
The industry has summarized a rule of "acceleration at both ends":
Leading Enterprise
Leveraging advantages in capital and data, drive the deep application of artificial intelligence throughout the entire process. For example, automotive OEMs have already incorporated suppliers' level of intelligentization into their admission evaluation system—if your factory has insufficient AI application, you won't get orders. This is not a trend in a PPT, but a procurement standard currently being implemented.
Small and medium-sized enterprises
Rely on standardized AI products and lightweight solutions for foundational transformation. Products such as Odoo, Kingdee Jingdou Cloud, and Yonyou Chanjet are integrating AI modules—highly similar to the past “cloud migration” path: start using first, then gradually upgrade.
Beijing's maximum subsidy for a single embodied intelligence demonstration project is 30 million yuan.
Beijing is promoting "embodied intelligence factory demonstration benchmarks," while Shenzhen has released the "Embodied Intelligent Robot Industry Development Action Plan," aiming to establish multi-domain robot industrial clusters by 2027. Guangdong and Zhejiang are also introducing special policies for the digital transformation of manufacturing.
Subsidies are real money, but receiving a subsidy and actually using AI are two different things.
After two years of consolidation, the industrial robot market saw modest growth in 2025 and is expected to rebound more strongly in 2026. The driving factors are not high-tech ideals, but continuously rising labor costs and the growing difficulty of recruiting skilled workers—"replacing people with machines" has shifted from an option to a necessity for survival. The three areas with the most obvious demand recovery are: automotive parts, 3C electronics, and new energy.
Large AI models are also beginning to permeate the entire manufacturing process: AI rapidly generates options in product design, automatically adjusts parameters in process planning to improve yield, and in quality inspection, machine vision accuracy has surpassed human accuracy in some scenarios.
What truly kills everything is neither policy nor technology
If one only reads policy documents and meeting news, it is easy to get the illusion that all conditions for implementing artificial intelligence are already in place.
In reality, the true situation of enterprises is another matter. In summary, there are three "bottleneck" problems:
- Talent. For manufacturing companies to adopt AI, they need talent that understands both production processes and data. Such talent is not cheap and is difficult to recruit. Most small and medium-sized enterprises do not even have a decent IT team, let alone AI engineers.
- Data. The production data of a large number of manufacturing enterprises has never been collected at all. Sensors have been installed, but the data lies in local machines, uncleaned, unlabeled, and unconnected. If an AI model needs to be trained, what will feed it?
- Technical generalization. Industrial scenarios vary greatly. A quality inspection model that performs well in an auto parts factory may be completely ineffective in an electronics component factory. Every time the scenario changes, time must be spent on parameter tuning, annotation, and validation, which is not cheap.
China's industrial robots account for 40% of the global total, but the overall integration depth between robots and AI quality inspection remains at an early stage. An IT manager at a manufacturing enterprise said: "We do not lack robots; what we lack is people who can make robots smarter."
What will the second half of this year look like?
Growth momentum is strengthening, but headwinds are also intensifying. In the second half of the year, several key points need attention:
- whether the subsidies can truly reach the hands of those in need. Although the NT$30 million special subsidy for embodied intelligence is tempting, if the selection criteria favor "companies that already have AI teams," small and medium-sized enterprises simply cannot meet the threshold. "Accelerating both ends" will become "one end sprinting ahead, the other falling behind."
- Can lightweight AI products meet the needs of SMEs?The delivery quality of products such as Odoo's AI record management and Kingdee Lingji's standardized AI modules will directly determine the speed at which SMEs adopt AI.
- Pace of AI Standards Implementation.The national AI agent evaluation standard has been submitted for review. Once released, enterprises will have a hard reference when selecting AI products. This will eliminate a group of AI vendors that only know how to make PPTs.
Policies have paved the way, but whether they work depends on companies' own data foundations and talent reserves. This is the real watershed for enterprise AI implementation in the second half of 2026.
National Artificial Intelligence Industry Development Conference (May 13, 2026), MIIT's "Implementation Opinions on the Special Action for 'AI + Manufacturing'", NetEase Hao analysis of "AI + Manufacturing Accelerates Implementation" (May 20, 2026), Guolian Minsheng Securities meeting interpretation, Sina Finance
