Alibaba foresees stronger second half of 2026 after expanding AI strategy

Alibaba has reorganized its AI operations under the Alibaba Token Hub while continuing to release new Qwen models and AI products.

Alibaba has reorganized parts of its AI business and expanded its cloud infrastructure as the company continues to build out its AI-related products, models, and services.

In March 2026, the company created the Alibaba Token Hub Business Group under Chief Executive Officer Eddie Wu. The unit brings together Tongyi Laboratory, the Model-as-a-Service business line, the Qwen Business Unit, the Wukong Business Unit, and the AI Innovation Business Unit.

The move consolidates Alibaba's AI model, platform, and application teams under one group. Alibaba describes the unit's focus as creating, delivering, and applying tokens across its AI systems.

Alibaba released several AI models in the first half of 2026. In May, it introduced Qwen3.7-Max, a large language model designed for agentic coding, complex reasoning, and longer task execution.

Alibaba said Qwen3.7-Max outperforms leading Chinese models and is comparable with top global systems. The company did not provide detailed benchmark categories or a complete comparison table in the source material.

Alibaba has released some Qwen models as open models while keeping other systems proprietary. The broader Qwen3 family includes dense and mixture-of-experts models ranging from 0.6 billion to 235 billion parameters, according to Alibaba's Qwen3 technical report.

In June, Alibaba released HappyHorse 1.1, an updated video generation model. The company said the model improves motion realism, consistency, and visual quality compared with its earlier version.

HappyHorse 1.0 was introduced in April. Alibaba said the model has been used in short-form content, advertising, brand marketing, and gaming cinematics.

Alibaba also launched HappyOyster 1.0 in June. The model is designed for interactive visual production, with features such as environmental interaction, expanded controls, and rewindable storylines.

The company has also added AI features to products used by consumers and enterprises. In January, Alibaba upgraded the Qwen App, its consumer AI assistant, to connect services across Taobao, Alipay, Fliggy, and Amap through a conversational interface.

The upgraded app is designed to support tasks across shopping, payments, travel, and navigation. Alibaba did not provide usage figures in the source material.

Alibaba also unveiled Qwen Glasses at MWC Barcelona. The device includes real-time translation, HD capture, transcription, visual recognition, and payment functions.

In China, Qwen Glasses can also be used to order food and hail rides through voice commands, according to the company.

In March, Alibaba introduced the Wukong Platform, an enterprise agentic platform for multi-step workflows. The company said the platform supports complex enterprise workflows and serves as its main solution for autonomous enterprise operations.

Cloud infrastructure and more AI tools

The cloud expansion is tied to Alibaba's three-year infrastructure plan. In February 2025, the company said it would invest at least RMB 380 billion, or about US$53 billion, over three years in cloud computing and AI infrastructure.

As part of that infrastructure commitment, the company opened new availability zones in Japan, Malaysia, France, and Mexico. Alibaba Cloud now operates 105 availability zones across 32 regions.

The new data centers provide cloud computing and AI infrastructure for enterprise customers, according to Alibaba. The company said the infrastructure is designed to support local requirements covering cybersecurity, resilience, data governance, data privacy, and data sovereignty.

Alibaba has also applied AI tools in healthcare and agriculture. In early 2026, Alibaba DAMO Academy launched two AI-powered screening tools: MAOSS for early detection of fatty liver disease, and COCA for colorectal cancer screening.

Alibaba said MAOSS achieved AUC scores of 0.904 to 0.917 in multi-center external validations across different stages of steatosis. The company said the average radiologist AUC score was 0.709, while radiologists using MAOSS as an assistive tool improved diagnostic accuracy to 0.798.

COCA was developed for colorectal cancer screening using routine non-contrast CT scans. Alibaba said the model was validated through a retrospective analysis of non-contrast CT scans from 27,433 individuals across multiple hospitals.

The company said COCA identified five previously missed colorectal cancer cases and achieved sensitivity of 86.6% and specificity of 99.8%. The available information does not establish broad clinical deployment, regulatory approval, or adoption across hospital systems.

In agriculture, Muyuan Group is working with Alibaba Cloud on an AI model for swine farming. The project uses Alibaba's Qwen large language model and computing resources.

The partnership covers operational areas such as feed nutrition, breeding stock improvement, and livestock management. Alibaba has also described a Qwen-powered swine disease diagnosis assistant developed with Muyuan Group, but the available information does not provide deployment figures or performance results.