Microsoft to Put More Computing Power Behind In-House AI Models

Microsoft Corp. has announced its intention to significantly enhance its physical infrastructure to support the training of proprietary artificial intelligence models. This move aims to position the tech giant competitively against established players like OpenAI and Anthropic. Mustafa Suleyman, the head of consumer AI at Microsoft, shared these insights during an employee meeting, emphasizing the importance of self-sufficiency in AI for a company of Microsoft's magnitude. Suleyman noted that Microsoft is pursuing a dual strategy: deepening its collaboration with OpenAI while also fostering partnerships with other model developers and building its own AI capabilities. Although Microsoft has primarily relied on the language models provided by OpenAI, there are indications of growing competition between the two as they both release new products and seek additional collaborations. Having joined Microsoft last year, Suleyman has been instrumental in spearheading the development of the company’s AI models and establishing a dedicated AI consumer franchise. Recently, Microsoft launched its first large language model, trained using 15,000 Nvidia Corp. H100 chips. Suleyman highlighted that competing models from Meta, Google, and Elon Musk’s xAI utilized clusters that were six to ten times larger, indicating that Microsoft's approach is more efficient. CEO Satya Nadella stated that Microsoft intends to adopt a multimodal approach across its products, ensuring that they utilize the AI model that best meets customer preferences. Furthermore, reports from the Information suggest the company plans to integrate Anthropic models into some of its offerings, reflecting a strategic diversification in its AI endeavors.