中国平安携手阿里巴巴：日均消耗超3000亿Token 产业 AI加速创造业务价值

This article covers key updates on Ping An’s industrial AI deployment in partnership with Alibaba, unveiled at the 2026 Apsara Conference, including practical updates relevant to everyday consumer services.

1. Efficiency gains from AI rollout: In under a year, with full-stack technical support from Alibaba Cloud, Ping An’s average daily AI token processing volume grew from roughly 30 billion to over 300 billion, while computing power investment increased by less than 2x. Per-token costs fell by 76%, marking that consumer-facing AI services have moved past the early high-cost phase.

2. Accessible everyday AI services: Ping An has launched AI-powered quick services for its over 250 million individual customers, with cumulative usage exceeding 100 million within six months of launch. In healthcare, 25% of in-hospital services and 70% of at-home services are initially triaged and routed by AI; the medical large language model (LLM) has a hallucination rate below 0.3% in consultation scenarios, reflecting strong service reliability.

3. Pragmatic approach to enterprise AI adoption: Rather than blindly scaling up computing power, the focus is on improving token conversion efficiency and adapting general AI capabilities to specialized use cases. This will continue to improve the practical utility of AI services available to the general public going forward.

Ping An’s industrial AI deployment in collaboration with Alibaba offers a benchmark reference for brands looking to advance AI adoption, improve operational efficiency, and upgrade customer service.

1. Cost benchmark for AI investment: Brands do not need to overinvest in computing power to scale AI. Full-stack optimization across models, inference, and computing scheduling can deliver very high return on investment. Drawing on Ping An’s practice, per-unit computing power can deliver a 368% increase in token output, with per-token costs falling 76%, enabling scaled AI service coverage at controllable costs.

2. Direction for customer service upgrades: Ping An’s AI quick services for 250 million customers surpassed 100 million uses within six months of launch, sharply improving response efficiency and reducing labor costs. This model offers a viable path for brands to optimize customer touchpoints and elevate service experience.

3. Approach to building product capabilities: Combining general-purpose LLMs with in-house professional data, expert knowledge, and real-world business feedback to build vertical-domain AI capabilities has proven effective. For example, Ping An’s financial and medical LLMs both ranked first in industry evaluations; this model helps brands build differentiated product competitiveness aligned with users’ core demand for professional services.

Ping An’s industrial AI deployment with Alibaba Cloud provides actionable methods and opportunity signals for businesses of all sizes to unlock value from AI use cases and identify paths to reduce costs and raise efficiency.

1. Replicable AI deployment path: Businesses do not need to bear high computing costs to adopt AI. Through model optimization, inference optimization, and smart computing scheduling, limited investment can support a 10x increase in business processing volume, significantly lowering the barrier to using AI tools.

2. Efficiency upgrade opportunities in service operations: The model of AI-led initial intake and routing has been validated in customer service and local lifestyle services. Ping An’s data shows 70% of at-home services are initially routed by AI; sellers can adopt this model to reduce service labor costs and speed up response to customer inquiries.

3. Direction for productivity upgrades: In R&D and production, Ping An now generates 80% of new code with AI and completes 67% of testing via AI. Sellers can leverage mature general-purpose LLM capabilities, combined with their own business data, to build AI tools tailored to their operations, unlocking new efficiency gains in marketing, customer service, quality control, and other areas.

Ping An’s scaled AI deployment with Alibaba Cloud offers clear reference points for factories pursuing digital transformation, unlocking AI value in production, and aligning with new industry demand.

1. Pragmatic approach to digital transformation: Factories do not need to make large upfront investments in hardware computing power to adopt AI. A better starting point is efficiency optimization—improving actual output per unit of computing power through model adaptation and process scheduling optimization. Drawing on Ping An’s experience, where per-unit computing output rose 368% and per-token costs fell 76%, controllable investment can deliver step-by-step upgrades to digital capabilities.

2. Efficiency upgrade direction for production: AI can be deeply integrated into core R&D and quality control processes. Mirroring Ping An’s model—where 80% of new code and 67% of testing in R&D are AI-powered—factories can embed AI into product design, production quality control, and process scheduling to reduce repetitive labor and improve production precision and efficiency.

3. Potential commercial opportunities: As scaled AI adoption accelerates across industries, demand will grow for hardware and services related to industrial scenario adaptation, computing infrastructure, and specialized model deployment. Factories can leverage their production scenario strengths to tap into related supply chain demand and open new growth avenues.

Ping An and Alibaba’s AI deployment clearly illustrates current industrial AI trends, the core pain points of large enterprise customers, and validated, viable solution directions.

1. Industry trend assessment: Industrial AI has moved beyond early pilot stages into a phase of creating tangible business value at scale. Ping An’s average daily token consumption grew 10x to over 300 billion in less than a year, indicating fast-growing enterprise AI demand and a rapidly expanding service market.

2. Core pain points for enterprise customers: Large enterprises do not want to scale AI purely by adding computing power. Their priority needs are improving token conversion efficiency, reducing per-unit AI usage costs, translating general LLM capabilities into business-specific professional capabilities, and solving reliability issues in specialized scenarios—for example, controlling model hallucinations in healthcare settings.

3. Actionable solution directions: Service providers can offer full-stack technical services spanning model optimization, inference optimization, and computing scheduling to help clients boost computing output and lower costs. They can also support vertical model fine-tuning using clients’ professional data, expert knowledge, and real business feedback to meet the high-reliability requirements of highly regulated sectors such as finance and healthcare.

The outcomes of Ping An’s AI partnership with Alibaba Cloud, unveiled at the 2026 Apsara Conference, highlight enterprise customers’ core needs for platform-based technology services and offer direction for platforms to optimize their service systems and expand client partnerships.

1. Core enterprise demand for platforms: When selecting AI technology partners, enterprises prioritize more than just a platform’s computing scale. They also need full-stack optimization capabilities across models, inference, and computing scheduling to help lower per-unit AI costs and improve token output efficiency, as well as mature general-purpose foundation models that support enterprises in building vertical-domain capabilities using their own data.

2. Validated platform collaboration model: By providing full-stack AI technical support and opening its Tongyi series of foundation models, Alibaba Cloud supported Ping An in achieving 10x token volume growth, and helped Ping An’s financial LLM top the CNFinBench ranking and its medical LLM score highest on HealthBench Hard. This “technology foundation + joint refinement” model helps platforms build high-value industry benchmark cases.

3. Platform operations and risk mitigation direction: Platforms can build an end-to-end service system covering computing scheduling, model fine-tuning, and LLM application deployment to meet the full lifecycle needs of scaled enterprise AI rollouts, attracting more large enterprise clients. At the same time, they should prioritize AI reliability in professional scenarios, such as reducing model hallucination rates through technical optimization, to mitigate service risks.

Ping An and Alibaba’s AI collaboration, unveiled at the 2026 Apsara Conference, provides a latest empirical case for observing China’s industrial AI deployment progress, summarizing patterns in scaled application, and studying new industry collaboration business models.

1. Latest trends in industrial AI development: China’s industrial AI has entered a value realization phase of scaled deployment. Ping An’s average daily token usage grew from 30 billion to over 300 billion in under a year, supporting 10x business scale with less than 2x growth in computing power, while per-token costs fell 76%. This indicates that efficiency gains from full-stack technical optimization, rather than brute-force computing expansion, have become the core driver of scaled AI development.

2. Real-world challenges and solutions in industrial AI deployment: As AI enters professional sectors such as finance and healthcare, the core challenges are translating general capabilities into domain-specific expertise and ensuring service reliability in professional scenarios. For example, Ping An’s financial LLM topped the public CNFinBench ranking, its medical LLM set a new record on the HealthBench Hard benchmark, and consultation hallucination rates were kept below 0.3%—proving that collaboration between foundation model providers and industry players, combining professional data, expert knowledge, and business feedback to refine vertical models, is a viable path.

3. New industry collaboration model: The model in which technology platforms provide full-stack AI technology and foundation models, while industry players open up business scenarios and professional data for joint model optimization and efficiency improvements, is emerging as the mainstream commercial collaboration model for industrial AI deployment.