At Apsara Conference 2025, Alibaba Cloud positioned itself as an AI-native cloud provider, combining infrastructure, foundation models, and platform services into a unified AI stack. One year later, the company’s ambition has expanded significantly. At Apsara Conference 2026, Alibaba outlined a full-stack AI strategy spanning chips, infrastructure, data, models, agents, and global operations. More importantly, Alibaba is signaling that the next phase of enterprise AI competition will not be defined by model superiority alone, but by who can successfully operationalize intelligent agents at scale and continuously improve them over time. 

Agent-Native Cloud Becomes Alibaba’s New Strategic Direction 

The clearest message from this year’s event was that Alibaba sees the industry moving beyond AI-assisted productivity toward agent-driven execution. Across multiple keynotes and product announcements, the company positioned Agent-Native Cloud as the next evolution of AI-native infrastructure. This represents a shift from delivering AI capabilities to enabling AI systems that can plan, act, collaborate, learn, and improve as part of enterprise operations. Alibaba Cloud is: 

  • Redefining cloud architecture around agents rather than models. The company repeatedly described enterprise AI through the lens of Model, Harness, and Context, while introducing Agent Sandbox, AgentCore, Agent Identity, Agent Memory, Agent Security Center, and AgentLoop. Taken together, these announcements suggest Alibaba increasingly views agents as a new computing abstraction that requires dedicated runtimes, orchestration, governance, observability, and lifecycle management rather than simply another application workload. 
  • Building operational infrastructure for enterprise-scale agent deployment. Beyond model hosting, the company emphasized capabilities such as agent runtimes, sandbox environments, identity management, memory services, tool integration, observability, and governance controls. These capabilities address the practical challenges enterprises face when moving from isolated AI pilots to thousands of agents interacting with users, data sources, applications, and business processes. 
  • Preparing for an agent economy inside enterprises. Alibaba executives described a future where agents operate as digital employees that possess memory, context, governance, and organizational awareness. Products such as Qoder Wake, Qwen Office digital employees, and enterprise agent platforms indicate that Alibaba expects organizations to manage agent populations in ways that increasingly resemble workforce management. 

Context Is Emerging As A Strategic Control Point 

Alibaba devoted strategic attention to context. Across analyst sessions and executive presentations, context appeared repeatedly as a critical pillar of enterprise AI. This aligns with a growing realization that as models become more capable and more accessible, differentiation will increasingly come from how organizations provide proprietary knowledge and business awareness to AI systems. Alibaba Cloud is: 

  • Elevating context engineering to a first-class infrastructure layer. Announcements such as AgentContext, enterprise memory services, multimodal knowledge platforms, semantic-layer capabilities, vector services, and context management frameworks suggest that Alibaba no longer views context as an application feature. Instead, it is treating context as a foundational platform capability that underpins search, reasoning, memory, orchestration, and decision-making across enterprise agents. 
  • Addressing the challenge of enterprise-scale contextual awareness. The company highlighted requirements such as multimodal data access, long-term memory, dynamic knowledge updates, sub-second freshness, and cross-agent context sharing. These capabilities are designed to help agents operate with a continuously evolving understanding of enterprise information rather than relying on static prompts or isolated knowledge repositories. 
  • Redesigning data architectures around context consumption rather than data storage. New initiatives such as AgentContext, context-aware databases, vector-native services, semantic capabilities, and knowledge-layer services demonstrate a shift away from traditional data platform thinking. The objective is increasingly to transform enterprise data into consumable context that agents can understand, retrieve, reason over, and operationalize. 

Alibaba Is Building A Full-Stack AI Platform For The Agent Era 

Unlike previous years, Alibaba used Apsara Conference 2026 to demonstrate coordinated progress across nearly every layer of the AI stack. Rather than focusing on a single flagship model or product, the company presented a comprehensive platform vision that spans silicon, infrastructure, data, models, agent frameworks, and enterprise applications. Alibaba Cloud is: 

  • Investing aggressively in AI infrastructure for the agent era. Highlights included the launch of the Yitian 730 processor and Zhenwu V900 with great performance gains and solid roadmap by Alibaba’s T-Head business unit, the next generation of Lingjun AI infrastructure, and upgrades across networking and storage architectures. Alibaba framed these innovations around the industry’s transition from pretraining to agentic reinforcement learning and eventually recursive self-improvement (RSI). This infrastructure strategy aims to provide the foundation for increasingly autonomous AI systems that require massive inference, memory, coordination, and continuous learning at scale. 
  • Expanding Qwen into a comprehensive multimodal AI portfolio. The company outlined an ambitious roadmap across foundation models, multimodal reasoning, audio understanding, image generation, video generation, world models, and software engineering capabilities with Qwen 4 on the way. Rather than competing solely on benchmark performance, Alibaba is positioning Qwen as a family of models optimized for diverse enterprise workloads. The strategy acknowledges that enterprises increasingly require a portfolio of models capable of supporting coding, productivity, customer engagement, content creation, reasoning, and agent orchestration scenarios. 
  • Modernizing the data layer for context-driven AI. Announcements such as AgentContext, ApsaraLakebase, semantic-layer capabilities, vector-native services, AI-native databases, and data agents signal a broader architectural shift. Alibaba is increasingly focused on helping enterprises transform raw data into consumable context that can be discovered, shared, governed, and operationalized by agents. This evolution moves beyond traditional data management and aligns closely with the emerging discipline of context engineering, which is becoming increasingly critical for enterprise AI success. 
  • Operationalizing agents within enterprise workflows and business processes. Products including Qoder, Qwen Office, Accio, Lingque, AI-native productivity capabilities, and industry-specific agent offerings demonstrate Alibaba’s effort to move AI from standalone tools into day-to-day operations. Rather than requiring employees to visit separate AI applications, Alibaba increasingly envisions agents operating within software development, office productivity, customer service, content generation, and operational workflows. This reflects the broader market transition from AI-assisted work toward AI-executed work. 
  • Integrating trust, governance, and sovereignty across the AI stack. The company highlighted capabilities such as Agent Identity, Agent Security Center, agent governance in AgentCore, secure model access, isolated runtime environments, model governance, observability, auditability, and support for private and regulated deployments. Combined with continued investment in global infrastructure and regional cloud expansion, these announcements recognize that enterprise AI adoption increasingly depends on security, compliance, operational control, and data sovereignty. As intelligent agents gain access to enterprise systems and business processes, trust maywill become as important a competitive differentiator as intelligence itself. 

What Enterprise Leaders Should Do Next 

While Alibaba’s announcements are competitive and its roadmap is ambitious, the broader lessons extend well beyond a single vendor. Enterprise leaders should focus less on headline model announcements and more on building the foundations required to operationalize AI at scale. 

  • Start with use cases, not models. The organizations generating measurable AI value begin with business workflows, operational bottlenecks, and desired outcomes. Model selection should support business objectives rather than become the objective itself. Organizations that struggle to prioritize opportunities should leverage Forrester’s AI use case catalogs and value frameworks to identify, prioritize, and sequence AI investments based on business impact. 
  • Take a holistic view of AI-native cloud capabilities. As AI workloads become specialized, organizations should assess more than GPUs and foundation models. Evaluate capabilities across agent platforms, context management, data services, observability, governance, and developer tooling. The strongest AI-native cloud platforms will increasingly combine neocloud strengths with neoPaaS capabilities to accelerate enterprise AI adoption at scale. 
  • Develop a sovereign AI strategy before regulations force one. Geopolitical fragmentation, data sovereignty requirements, regulatory pressures, and board-level risk concerns are making deployment flexibility increasingly important. Enterprise architecture leaders should evaluate how sovereign cloud and sovereign AI fit into their long-term operating model. The goal is not simply compliance, but ensuring that AI innovation can scale across jurisdictions, business units, and risk profiles without costly architectural rework. 
  • Invest in context and harness engineering capabilities now. At Forrester AI Forums in Singapore and Sydney earlier this year, I argued that enterprises are entering a new era where “Context Is All You Need”, which was also used by Alibaba Cloud in its keynote. While models are advancing rapidly and becoming increasingly accessible, context remains proprietary, difficult to govern, and uniquely tied to enterprise value creation. Forrester’s AEGIS framework provides a comprehensive governance framework for agentic AI governance, and we will soon publish new research on context engineering and ContextOps, please stay tuned. 
  • Embrace open ecosystems while maintaining architectural control. Open-weight models, agent frameworks, vector databases, semantic technologies, and infrastructure tooling continue to accelerate AI innovation. Organizations should take advantage of the speed and flexibility offered by open-source AI ecosystems while maintaining control over data, context, governance, and integration standards. This balance allows enterprises to avoid unnecessary lock-in while preserving the ability to adopt emerging technologies as the market evolves. 

If you’d like to dive deeper set up an inquiry or guidance session with Charlie Dai (AI-native cloud, agentic AI, context engineering, harness engineering, and humanoids), and Meng Liu (enterprise fraud management, identity verification, fintech, security and risk in financial services) for a conversation. 

Share