Related articles
Image source: JD.com
AsianFin — JD.com's industrial arm is making a bold push into AI-powered manufacturing with the launch of Joy Industrial, the industry's first supply chain-focused industrial large model.
The move comes as China escalates its "AI+" national strategy to integrate artificial intelligence more deeply into industrial production and infrastructure.
At a launch event in Shanghai, JD Industrial showcased Joy Industrial as a full-stack AI solution designed to tackle inefficiencies and bottlenecks across the industrial supply chain — from demand forecasting and procurement to compliance and fulfillment. The system is built atop JD's proprietary large language models, drawing on more than a decade of supply chain data and industry know-how.
"Every supply chain deserves to be reimagined with AI," said Gu Yingkun, Vice President of JD Industrial. "This is about moving from experience-driven decision-making to data-driven intelligence."
While general-purpose large models like OpenAI's GPT and China's own foundational models have seen rapid adoption in consumer-facing applications, they often fall short in vertical industrial settings. A recent report by the China Industrial Internet Research Institute found that the average accuracy of these models in industrial applications remains below 60%, underscoring the need for more specialized solutions.
Joy Industrial takes aim at that gap. Built on JD's 750-billion-parameter foundation model and enhanced with domain-specific data from sectors such as automotive, energy, and manufacturing, the system boasts a context window of 1.28 million tokens — allowing it to reason over extensive technical documentation and complex supply scenarios.
JD has already integrated over 14,000 AI agents across its operations. These agents now handle more than 18% of all work tasks in areas spanning retail, logistics, healthcare, and industrial sourcing.
Despite China possessing the world's most comprehensive industrial system, it faces persistent supply chain challenges: fragmented standards, inefficient collaboration, siloed data, and complex workflows.
Gu cited the example of JD managing over 57 million SKUs, where small parameter variations across similar products can complicate price comparison and regulatory compliance. "Manual coordination simply can't scale to this level of complexity," he said.
In one notable case, JD helped a nuclear power plant complete an emergency procurement within 72 hours, avoiding millions in potential losses.
JD's ecosystem approach includes partnerships with giants like JinkoSolar, State Grid, and PetroChina. JinkoSolar, for example, reduced distributed power station maintenance costs by 30% through AI-enabled resource optimization.
JD is also pushing for standardization. Its AI-generated Mercator Standard Product Database now covers data from hundreds of thousands of industrial SKUs, including those from Schneider Electric and Jinbei Electric. The goal is to establish "JD Standards" as a national framework for product classification, reducing friction in B2B supply chains.
He Xiaodong, Deputy Director of JD's Exploration and Research Institute, revealed the company is now investing in embodied intelligence — aiming to bring general AI capabilities into robots and warehouse equipment. JD's Joy Inside robotic dog, powered by its large model, is already in development for use in logistics and industrial inspections.
Despite the promise, hurdles remain. Industrial customers are wary of data privacy and the trustworthiness of AI decisions. Even with a 95% accuracy rate, the remaining 5% can be costly in high-stakes environments. Moreover, implicit human knowledge — such as equipment repair expertise — remains difficult to digitize.
To address these issues, JD is investing in model distillation to lower training costs by 70% and boost inference efficiency by 30%. The company is also building simulation environments to capture experiential data and further train its models.
As industrial AI shifts from foundational models to verticalized and even self-refining systems, JD is positioning itself at the forefront of this transformation. Its efforts align closely with China's broader industrial digitalization agenda, which has designated manufacturing as a key area for AI empowerment.
"The release of Joy Industrial represents a pivotal moment," Gu said. "We're not just chasing efficiency — we're rewriting the logic of the industrial supply chain."
With deeper industry integration, scalable AI agents, and a growing portfolio of real-world applications, JD's latest AI leap could help usher in a new era of industrial productivity — and set a benchmark for what AI + manufacturing really means.
