SoftBank has completed a $40 billion investment in OpenAI, becoming its largest institutional shareholder, laying the groundwork for OpenAI's upcoming IPO and the valuation of AI companies. Meanwhile, China's Zhipu AI is set for an IPO in Hong Kong, planning to raise $560 million to become the world's first listed foundational large model company. The performance of its GLM-4.7 model is nearing Western cutting-edge levels. With an IPO valuation of $6.7 billion, 70% of the funds will be used for the research and development of general AI models.

Meta announced the acquisition of Singapore-based AI Agent company Manus for approximately $2 billion. Manus is known for its multi-agent capabilities in deep research, code, data analysis, and more. Meta plans to operate the Manus team independently and integrate its Agent capabilities into platforms like Facebook, Instagram, and WhatsApp. Manus will sever all ties with China to mitigate regulatory risks.

The average equity incentive for OpenAI employees is as high as $1.5 million, far surpassing any tech startup in history, significantly increasing the company's operating losses and shareholder dilution. The company recently canceled its equity vesting waiting period policy, paving the way for potentially higher future employee compensation.

Elon Musk's xAI has acquired the MACROHARD data center in Mississippi, forming a mid-southern compute cluster alongside the Colossus series. The total training compute power is nearing 2GW, with the goal of operating millions of GPUs simultaneously. This strengthens xAI's own AI infrastructure and reduces its reliance on third-party clouds.

Alibaba Tongyi Lab has open-sourced MAI-UI, a multimodal GUI Agent (2B~235B parameters). Based on the Qwen3-VL model, it can automatically perform multi-step operations in Android apps, supporting UI operations, natural language dialogue, and API calls. It employs self-evolutionary navigation data, online RL, and edge-cloud collaboration, achieving up to 76.7% success rate on real-world app tasks, and is licensed under Apache 2.0.

Microsoft CEO Satya Nadella outlined his 2026 outlook, stating that AI model capabilities have outpaced practical implementation. The industry will shift from individual model competition to system-level integration and creating real-world value. AI will become the 'scaffolding for human potential,' with future emphasis on the orchestration and deployment of intelligent systems.

Tencent's Hunyuan team has open-sourced Hunyuan Motion 1.0. Based on DiT architecture and flow matching techniques, it supports generating high-quality 3D skeletal animations from text descriptions, is compatible with mainstream 3D tools, significantly lowers the barrier to animation production, and is suitable for production scenarios like gaming and film.

AI data centers will undergo structural transformation in 2026, with six major trends including standardization of liquid cooling technology, architectural reshaping for extreme rack density, shift from single-site construction to industrialized scale delivery, grid synergy and on-site power generation, comprehensive AI integration into design and operations, and sustainability becoming a core design principle. Data centers are becoming global infrastructure assets, with deep integration of capital and technology.

In 2025, AI has become infrastructure, with generative AI and chip races, data center physical foundations, AGI, and quantum computing emerging as new industry foci. 2026 will enter a 'disciplined' phase. AI agents will become the core digital workforce, governance will move into the C-suite, energy and compute will determine competitiveness, and AI's impact on fields like media, science, and healthcare will deepen.

Gartner predicts that by 2028, over 60% of enterprise generative AI will use Domain-Specific Large Models (DSLMs), with general LLMs becoming infrastructure. DSLMs achieve up to 95% accuracy in fields like healthcare, law, and finance, with clear advantages in compliance and cost, driving the AI industry from 'universal' to 'specialized.'

In 2026, enterprise AI data infrastructure will undergo six transformations: evolution of RAG (Retrieval-Augmented Generation) technology, long-context memory becoming standard for Agents, narrowing uses of vector databases, PostgreSQL becoming the preferred choice for AI applications, continued innovation in data parsing and natural language SQL, and accelerated mergers and consolidation in data infrastructure.