From October 14th to 16th, the 2026 Artificial Intelligence Industry Conference was held at Jinan Yellow River International Convention and Exhibition Center. After being held in Beijing for two consecutive years, this national-level industrial event left Beijing for the first time and settled in Shandong.
Why was Shandong chosen for the national AI conference?
When AI accelerates its entry into the real economy, what lies ahead is no longer a simple technical problem, but a complex system problem.
Real scenarios provide problems, high-quality industrial data forms the basis, a deep understanding of industry mechanisms calibrates the direction, solid engineering verification maintains the bottom line, and efficient ecological collaboration promotes the diffusion of capabilities. Only when the conditions of all parties are connected and coordinated can technological advantages be transformed into industrial capabilities.
Shandong, which has all 41 major industrial categories and more than 40,000 regulated industrial enterprises, provides a "full subject examination room" with rich question types and strict requirements. Here, industries such as steel, chemicals, equipment, home appliances, and energy are densely distributed, providing real and complex application requirements.
The first question is how to turn massive data into usable industrial corpus.
Industrial sites generate data every day. Equipment parameters, process records, operation videos, fault logs... are huge in number, but often in different formats, standards, and quality. Some of them involve trade secrets and production safety. Only after collection, cleaning, alignment, annotation, desensitization, confirmation and verification can it become high-quality corpus that the model can truly "eat".
At Inspur Hairuociyuan Factory, a petrochemical chain owner submitted more than 50,000 pieces of data, hoping to develop an anti-corrosion model for pipeline equipment. After security review, information desensitization and feature analysis, technical staff found that the proportion of equipment information was too high and the samples were unevenly distributed. If used directly for training, the model can easily become "biased".
As a result, the team supplemented industry data and generated auxiliary samples, forming 580,000 training materials. First, 6,000 items were carefully annotated manually, and then multiple models were used to annotate the remaining data respectively; if the results were consistent, the results would be passed, and if there were differences, they would be handed over to manual adjudication. 11 processes and 60 sets of tools were completed, and the data preparation was compressed to less than two weeks.
In 2025, Shandong will launch the "unveiling the list" action of key industry corpora, allowing "experts" on the front line of the industry to take the lead in organizing the industrial data sleeping in the workshop into industry corpora that can be used in large models. So far, Shandong has selected and built a total of 35 key industry corpora, covering 31 manufacturing categories such as steel, chemicals, automobiles, medicine, and equipment. It is the only province in the country whose corpus covers all manufacturing categories.
Completion is not the key, using it is the key. In the field of steel, the steel industry corpus led by China Mobile Shandong Company integrates multi-source data such as process manuals, equipment files, maintenance cases, industry standards, and expert experience into an industry-specific knowledge map. The knowledge retrieval efficiency is increased by 30%, allowing the implicit experience of old masters to be precipitated, reused, and passed on. On this basis, the Rizhao Steel Intelligent Engine model is hatched.
In the field of power transmission and transformation equipment, Shandong Cloud Electric Chain Industry Development Co., Ltd. adopts the technical architecture of "distributed storage + intelligent governance", integrating natural language processing, computer vision and large model technology to perform automated analysis, deep cleaning, intelligent annotation and high-quality denoising of multi-modal data. At present, the library has gathered 82,000 texts, 35,000 images, and 520 hours of audio and video, with multi-modal associated data accounting for 32%, providing a solid data base for scenarios such as equipment predictive maintenance and intelligent scheduling.
The second question is how credible the model is.
A deep contradiction faced by large industrial models is that the model output is probabilistic, while industrial production emphasizes certainty. If a question is answered incorrectly in the chat box, the user can ask the question again; if a parameter is miscalculated next to the chemical plant, the entire production line may be shut down; if a danger signal is missed deep in the mine, the price will be even more unbearable.
Therefore, the credibility of industrial AI is guaranteed by a complete set of engineering mechanisms: the mechanism model demarcates the boundaries, the knowledge map supplements industry relationships, professional small models undertake real-time perception, digital twins simulate and verify in advance, and security rules, authority systems, and manual confirmation control the last hurdle.
In the steel industry, "looking at the color of the fire" was once an important skill for furnace directors to judge the smelting status. Shandong Iron and Steel's Shangzhi Metallurgical Large Model associates flue gas, video, audio and process data, allowing different forms of on-site information to participate in judgment. The end-point temperature hit rate is stable at over 95%, the single-furnace smelting cycle is shortened by two to three minutes, the cost per ton of steel is reduced by 3 to 6 yuan, and all scenarios create profits of more than 100 million yuan a year.
The ability to generate benefits of hundreds of millions of yuan relies on Shandong Iron and Steel's more than 40 years of experience. The development team sorted out the master's fire skills into a knowledge base step by step, and built a metallurgical industry knowledge map to provide process basis. There are also professional small models responsible for real-time perception, safety rules limit the operational boundaries, and key links are still subject to manual review.
This is where the value of the vertical domain model lies: its value does not depend on the parameter scale, but on how deeply it understands an industry.
When developers fully understand the equipment, process, and responsibility boundaries, and then use algorithms to reconstruct this knowledge, there will be soil for trust to grow. Only when companies dare to hand over production lines to it and workers are willing to follow its instructions can AI truly get its "employment certificate."
The third question is how the industry collaborates.
Industrial companies hold data, but it is difficult to bear all the costs of basic models, data governance, and security verification on their own; AI companies hold algorithms, but cannot obtain complete process data; small and medium-sized enterprises need public capabilities the most, but often cannot afford the cost of trial and error.
Each of the three has its own strengths and difficulties. For large-scale implementation, these capabilities need to be organized.
On May 15 this year, at the first event of the "Qilu Manufacturing Living Room", three companies, Inspur, Kaos, and Guogong Intelligent, hit it off. On July 24, the Shandong petrochemical industry general model based on the Inspur Hairuo Petrochemical general model was officially released. The "Huayu Tiangong" artificial intelligence innovation ecological co-construction plan gathered 100 partners in only 70 days.
The significance of this matter is not just to release one more model. It is equivalent to building a set of intelligent public infrastructure for the trillion-level Shandong chemical industry: safety standards, common processes, and common algorithms are deposited in the public layer, and the company's formulas and business data of key parameters remain in the private layer. The cost of repeated investment by each company in the past has become shared by the entire industry.
The more solid public capacity is built, the more support companies will have for personalized innovation.
The petrochemical industry is just one example. Shandong's "one industry, one model" initiative has rolled out 122 large industry models, achieving full coverage of 41 major industrial categories; it has selected 100 large vertical field models for key promotion, covering 13 key areas such as industry, people's livelihood, government affairs, and scientific research. The idea behind it is clear: not only should leading companies be the first to make breakthroughs, but also the capabilities accumulated by leading companies should be transformed into tools that can be shared by the industry.
This also highlights another important role of Shandong in developing artificial intelligence: the organizer of industrial collaboration.
Open application scenarios, connect supply and demand, promote standard construction, support pilot verification, and allow industrial enterprises, technology enterprises, scientific research institutions and service providers to find cooperation interfaces. Only when the government "sets up the stage" and "sets up the game" and enterprises can show their strengths and achieve mutual benefit and win-win results can the threshold for the implementation of industrial AI be lowered.
So far, the "Artificial Intelligence Empowering New Industrialization Qilu Tour" has been held 11 times, more than 1,200 artificial intelligence application needs have been explored, and on-site signings of 116 intelligent projects have been promoted, with a contract value of 1.456 billion yuan. In the past two years, Shandong has integrated 125 million yuan in fiscal funds and implemented reward and subsidy policies such as "model coupons", "corpus coupons" and "scenario coupons"; in the next three years, it will guide more than 10 billion yuan in investment in the field of artificial intelligence-empowered manufacturing.
According to the plan, by 2030, the output value of Shandong's core artificial intelligence industry will exceed 300 billion yuan, and the penetration rate of artificial intelligence application in manufacturing enterprises above designated size will reach 70%.
Achieving this goal means that AI must enter more companies with weak foundations, smaller scales, and different needs. This process requires lightweight products and reliable services, as well as skills training and job transformation support, so that front-line workers have the ability to participate in technological progress and share development results.
The success of a factory is a breakthrough in a problem; if more companies can afford it, use it well, and use it for a long time, it will be a more weighty report card.
Looking at the whole country, the value of Shandong’s exploration lies here.
The development of artificial intelligence requires not only continuous breakthroughs in high-end chips, basic software, and basic models, but also continuous verification and improvement in vast industrial sites. Shandong has all 41 major industrial categories in the country. Solutions that work here can be more easily copied to other regions; the problems exposed here can also provide valuable reference for other places.
The conference chose Shandong because of its strong industrial base and the possibility of transforming the depth of industry into depth of artificial intelligence applications.
"Kaogong Ji" says: "The sky has its timing, the earth has its Qi, the materials have beauty, and the workmanship has skill. If these four are combined, then it can be good."
The weather is the historical juncture for AI to move out of the dialog box and into the industry, and it is the comprehensive rollout of the national "artificial intelligence +" action; the land atmosphere is the industrial soil supported by more than 40,000 domestic industrial enterprises in 41 major industrial categories; the beauty of materials is the awakening of data sleeping in the machines; and the craftsmanship is the dedicated effort of many vertical large-scale models in Qilu to delve into the production site.
But the most difficult and critical thing is the last word "合".
Scenarios, data, mechanisms, engineering and ecology are not simply stacked together, but must be embedded in the same set of production logic; technology, industry, talent, capital and systems can only form true industrial intelligence by connecting and achieving each other.
