This article is part of our series that explores the business of artificial intelligence
In last week’s Ignite 2022 conference, Microsoft announced another integration of OpenAI’s technology with its Azure cloud platform. Now, Microsoft clients can sign up to use DALL-E 2 through Azure OpenAI Services.
DALL-E 2 is one of several text-to-image generative models that have made the rounds this year. DALL-E takes a textual description as input and generates an image that fits the description. DALL-E’s results are so impressive that many organizations and artists are considering using it to create original art.
The fast-paced evolution of DALL-E as a commercial product is a sign of the expansion of the market for generative AI models and the maturity of Microsoft’s long-term partnership with OpenAI.
DALL-E and its competitors
OpenAI announced DALL-E 2 in April and launched it as a paid API service with a waitlist. Shortly after, Google released Imagen, another equally impressive text-to-image generator. A major game-changer, however, was Stable Diffusion, released in August by Stability.ai. Unlike DALL-E 2, Stable Diffusion is open source and available for everyone to download and run.
To be clear, Stable Diffusion per se is not a direct competitor to OpenAI’s DALL-E 2 API. It does not provide the convenience of abstracting the technical difficulties of setting up and running the model. But setting up and running the model is not rocket science. Anyone with decent knowledge of deep learning and the right resources will be able to launch their own Stable Diffusion service. And the open-source format of Stable Diffusion has opened the way for customization and the development of new applications.
Since its release, Stable Diffusion has become the basis of research and new products. Many researchers have repurposed it for different applications. On the other hand, developers have made it available through API access on Hugging Face and other platforms.
This evidently posed problems for OpenAI and put pressure on it to maintain its edge in the market. In late August, OpenAI added outpainting to DALL-E 2, a feature that has also been replicated with Stable Diffusion. In late September, OpenAI removed the waitlist (why wait for access to DALL-E 2 when there are plenty of alternatives available?).
But none of these moves have provided OpenAI with long-term defensibility against competitors. There’s little doubt that there will be a market for text-to-image generator models. But in this nascent market, competition quickly shifts across different planes.
First, OpenAI was the only player in the market. It could set the norm for availability, features, and pricing. But the entrance of Stable Diffusion and potential new players triggered competition on features, pricing, and convenience, which perhaps explains why OpenAI removed the limits of DALL-E 2 much faster than it did for GPT-3.
This is where Microsoft’s role becomes prominent.
DALL-E 2 on Microsoft Azure
In a blog post, Microsoft detailed the integration of DALL-E 2 into its products. The image generator will be added to Azure OpenAI Service, which launched in 2021 and allows Azure customers to access OpenAI deep learning models such as GPT-3.
According to the Microsoft blog post, “The addition of DALL∙E 2 builds on Microsoft and OpenAI’s ongoing partnership and expands the breadth of use cases within Azure OpenAI Service, the newest in the Azure Cognitive Services family currently in preview, which offers the security, reliability, compliance, data privacy and other enterprise-grade capabilities built into Microsoft Azure.”
This can be an important factor for companies that need to verify industry standards and regulatory compliance before adopting a new solution.
Microsoft is also adding DALL-E 2 to other products, including Designer, an upcoming tool for creating social media posts, invitations, digital postcards, graphics, and more. It will also integrate DALL-E 2 with Image Creator from Bing, a tool that enables you to search the web for images or create new ones based on your search query.
This integration strategy can be a game-winner for both OpenAI and Microsoft. When choosing solutions to their problems, customers weigh different factors, and preferences can vary a lot depending on the use case and setting.
For example, for a company that is developing a brand-new product and has no previous infrastructure in place, the difference between OpenAI’s API, Stable Diffusion, and Azure OpenAI Service might not be significant. But for an enterprise that is already using Microsoft’s cloud infrastructure, DALL-E on Azure will be a much better option.
Microsoft already has a wide and deep reach into the enterprise sector, working with very large companies across different industries. For many of these customers, DALL-E on Azure will be a more convenient option. In the blog post, Microsoft provides an example of an enterprise client that is using DALL-E on Azure to design toys or generate ideas for new toys.
At a time when developers, artists, and organizations are exploring ways to put DALL-E to productive use, the display of real-world case studies goes a long way toward solidifying Microsoft and OpenAI’s position in the enterprise segment of this fast-evolving market.
Microsoft’s partnership with OpenAI
Microsoft’s partnership with OpenAI began in 2019. Microsoft invested $1 billion in OpenAI in exchange for an exclusive license to its technology. Since then, OpenAI has kept some of its deep learning models closed and only accessible through paid APIs. At the same time, Microsoft has benefitted from its agreement to integrate models such as GPT-3, Codex, and now DALL-E into its existing products.
Both companies stand to benefit from this partnership. Microsoft’s exclusive access to OpenAI’s cutting-edge technology helps it increase the value of its products and deliver AI services at scale to its widespread network of enterprise clients. On the other hand, OpenAI gets subsidized access to Microsoft’s compute resources and shortcuts its way to profitability through integration with Microsoft products and access to its clients.
But the partnership also comes at a cost to OpenAI. The lab that was founded in 2015 for purely scientific purposes is constantly being pulled toward the interests of its main financial backer. As I’ve argued before, maintaining the balance between scientific research and commercial product development is very difficult. And this conflict will intensify as OpenAI becomes more reliant on revenue from its partnership with Microsoft.