Microsoft's AI Strategy Drives Significant Growth in Cloud Revenue


Microsoft open sources the multimodal reasoning model Phi-4-reasoning-vision-15B, with 15B parameters, balancing lightweight design and high performance. The model is trained using only 200B multimodal tokens, emphasizing data quality, and is suitable for complex visual tasks in resource-constrained environments.
Microsoft Bing team open sources the word embedding model Harrier, which supports over 100 languages and performs excellently in the MTEB v2 benchmark. The model is trained on 2 billion examples and GPT-5 synthetic data, using a 32,000 token context window, with 2.7 billion parameters, significantly improving the accuracy and flexibility of multilingual tasks.
Anthropic hires Eric Boyd from Microsoft to manage infrastructure for AI product Claude amid surging demand, addressing service instability issues.....
Microsoft is researching high-temperature superconductivity to enable zero-resistance current transmission, revolutionizing cloud power distribution for AI's high demands. This technology eliminates energy loss and heat buildup, extends transmission distances, and may replace copper/aluminum wiring, offering efficient data center solutions.....
Microsoft is accelerating its in-house development of cutting-edge AI models, aiming to lead in text, image, and audio processing by 2027 to challenge OpenAI and Anthropic, shifting from external collaborations to strengthening core proprietary technology.....