Microsoft is committing more money and computing resources to OpenAI, extending a partnership that already connects the startup’s AI research with Microsoft’s cloud and software products. The companies say the expanded relationship will support research and give developers and organizations access to AI tools through Azure.
A larger partnership, with the amount undisclosed
Microsoft described the investment as “multi-year, multi-billion-dollar,” but did not disclose a precise figure. OpenAI said the new funding would support its independent research and the development of AI that is “safe, useful and powerful.”
The agreement preserves OpenAI’s capped-profit structure. Under that model, investor returns are limited to 100 times their investment, or potentially less in the future. The arrangement therefore combines a larger Microsoft commitment with limits on how much backers can earn.
Microsoft also plans to expand investment in specialized supercomputing systems for OpenAI’s research. Its Azure cloud platform will remain OpenAI’s exclusive cloud provider, supporting the startup’s research workloads, products and API services.
From cloud infrastructure to software features
The companies’ partnership has centered on both the computing needed to develop AI and the ways those systems can reach customers. Microsoft has said it intends to make OpenAI’s foundational systems available as commercial platforms, so organizations across industries can build on them.
Microsoft has already brought OpenAI technology into products and services, including Power Apps and Microsoft Edge. GitHub, which Microsoft owns, jointly developed Copilot, a code-generating AI system. The companies’ plans also point toward broader integration of OpenAI models in Microsoft products and new digital experiences.
The relationship has developed over several years. In 2019, Microsoft announced a $1 billion investment, roughly half in Azure credits, to develop technologies for Azure and extend OpenAI’s large-scale AI capabilities. OpenAI agreed to license some intellectual property to Microsoft for commercialization and to train and run AI models on Azure.
A year later, Microsoft said it had built an Azure-hosted supercomputer co-designed with OpenAI. In 2021, it launched Azure OpenAI Service, giving enterprises access to OpenAI systems such as GPT-3 alongside business-focused features including security, compliance and governance. The latest announcement builds on that infrastructure and product work.
Growth depends on costs and trust
Expanding access to AI systems also raises a business challenge: developing and operating them is expensive. The source reports that OpenAI expected $200 million in 2023 revenue from licensing and premium products such as ChatGPT Professional. Altman described ChatGPT’s operating expenses as “eye-watering,” with compute costs of a few cents per chat.
Those costs matter because the partnership’s ambitions rely on sustained investment in computing. Making models available through cloud services and products can broaden their use, but it also means the companies must keep funding the infrastructure behind them while finding ways to make the business work.
Trust presents a separate challenge. ChatGPT can produce answers that sound convincing but are not factually true. The source also describes restrictions on sharing AI-generated responses on Stack Overflow and bans by several public school systems and at least one academic conference.
Copyright disputes shadow wider adoption
Questions about the material used to train AI systems have also prompted legal challenges. Microsoft, GitHub and OpenAI were being sued in a class action that accused them of copyright violations connected to Copilot’s ability to reproduce licensed code without credit. The source notes concerns that companies could unknowingly use code suggestions with incompatible terms.
Similar disputes reach generative art tools such as DALL-E 2. The article describes a legal case against several OpenAI competitors alleging that their tools were trained on images scraped from the web without artists’ consent. It also reports that Getty Images had banned AI-generated content over concerns about legal consequences.
The investment deepens a relationship built on cloud capacity, research and product access. Its longer-term prospects will depend not only on the systems the partners can build, but also on whether they can manage operating costs, improve reliability and address copyright concerns as AI reaches more users.