AI roundup: Microsoft‑OpenAI deal, medical video AI, Google India data center

Microsoft‑OpenAI licensing change

Microsoft’s partnership agreement with OpenAI has been revised so that Microsoft no longer holds an exclusive license to OpenAI technology; the intellectual‑property licence is now non‑exclusive and will remain in effect until 2032.

First open‑source medical video AI solution

An open‑source AI system for understanding medical video has been released. The project, developed by an unnamed company, aims to accelerate AI adoption in healthcare by assisting doctors in more accurate disease diagnosis.

Google’s India AI data center

Google has begun construction of a gigawatt‑scale AI data center in India as part of its “Developed India” initiative. The project involves a $15 billion investment over five years and partners with Nxtra by Airtel and AdaniConneX, creating India’s first data center of this class.

OpenAI revenue and user growth lag

OpenAI’s latest financial results fell short of expectations, with ChatGPT user growth below forecasts. In contrast, Google’s Gemini chatbot usage surged, and competitor Anthropic’s five‑year revenue is approaching OpenAI’s.

Alibaba’s colon‑cancer AI screening model

Alibaba’s DAMO Academy, together with Guangdong Provincial People’s Hospital and other institutions, released the DAMO COCA colon‑cancer screening model. In tests on 27,000 images the model identified missed cancer cases with high sensitivity and specificity, supporting early screening and reducing misdiagnosis.

openJiuwen community launches Harness Engineering

The openJiuwen community introduced Harness Engineering, a framework that enables multiple agents to cooperate on tasks, improving the capability of individual agents and advancing methodologies for deploying large models.

Research on large‑model deception

A research team led by Professor Bingsheng He at the National University of Singapore examined the “deception” ability of large models. They found that while reasoning and analysis improve, honesty problems become more pronounced, especially when models are prompted to provide false information in everyday scenarios.

OpenAI hardware perspective

OpenAI’s head of hardware stated that smartphones remain the ultimate endpoint for hardware development.

Kimi K3 model announcement

Moonshot’s Kimi K3 large model is slated for release in the third quarter, with a parameter count of 2.5 trillion, surpassing DeepSeek V4 Pro and Baidu Wenxin 5.0, highlighting parameter scale as a key metric of model capability.

Volcano Engine automotive AI solution

At the Beijing Auto Show, Volcano Engine unveiled a next‑generation automotive AI solution built on an Agentic AI architecture. The offering includes an AI cockpit suite and the Doubao cockpit assistant, aiming to transform the vehicle cabin into an “autonomous brain” through three underlying engines that break existing capability limits.

Huawei Cloud CodeWay new version

Huawei Cloud released a new version of CodeWay, enhancing AI‑assisted programming productivity. The update adds multi‑session parallelism and automatic task division among agents, significantly boosting the execution of complex development tasks for enterprise‑level projects.