Microsoft takes an exclusive licence to GPT-3
Microsoft gained access to the underlying model and source code for its own products; the public API, through which OpenAI still served other customers, was unaffected.
- Money & business
- Labs & people
- Models & capabilities
- Notable
OpenAI announced that Microsoft, which had invested $1 billion in OpenAI in July 2019 partly in the form of Azure cloud credits, would receive an exclusive licence to GPT-3’s underlying model — allowing it to embed, repurpose and modify the technology directly rather than access it only through calls to OpenAI’s public API. OpenAI’s post described the arrangement as building on the companies’ existing partnership; Microsoft executive vice-president Kevin Scott said it would let Microsoft “leverage OpenAI’s technical innovations” to build its own AI-powered products.
The two forms of access were kept distinct. Third-party developers who had been granted API access continued to use GPT-3 exactly as before, generating text through a request-and-response interface without ever seeing model weights or source code. Microsoft’s licence went further, giving it the ability to fine-tune and integrate the model at the code level for its own commercial offerings — an asymmetry that made explicit, for the first time, that OpenAI’s most capable model was not equally available to everyone with API access.
The deal drew scrutiny for how it sat against OpenAI’s founding language of open, broadly distributed benefit, since it handed its largest investor a form of privileged access that competitors — including other paying API customers — did not have. It also previewed the shape the OpenAI–Microsoft relationship would take over the following years: Microsoft compute and capital in exchange for preferential product access, a structure that widened considerably with the multi-billion-dollar investments and Azure exclusivity commitments that followed after ChatGPT’s launch a little over two years later.