A few things not being mentioned in Uber’s acquisition of Mighty AI
We learned earlier this week that Uber had acquired Mighty AI, one of the leaders in building training data for computer vision. Most of the articles have focused on Uber’s self-driving car ambitions, which is reasonable considering the importance of training data to their efforts. That said, there are a few additional takeaways that I found fascinating.
1. One less option for AI researchers
The list of Mighty AI customers that now need to find alternative annotation services includes a number of big names including Intel, Samsung, Microsoft, Amazon, BMW, Snap, Siemens, and a number of others. They were well-regarded for their ability to provide high quality computer vision training data, as well as doing so in a fairly hands-off fashion that didn’t require a ton of setup and management work on the part of customers. There are still a number of options out there but I’m sure a number of these companies are scrambling to setup alternative providers that can meet their needs.
2. Scale matters
Even with the growth in training data needs for AI researchers, it may still be tough sledding for smaller AI labeling providers. This is the second provider to get scooped up this year following Appen’s acquisition of Figure Eight (formerly CrowdFlower) in March. Figure Eight had been seeking a buyer for a while and based on GeekWire’s reporting it sounds like Mighty AI’s investors will be taking a bit of a haircut in the sale. One of the challenges for both companies was likely that training data needs tend to ebb and flow with the nature of ML projects. In most cases companies can develop a set of data for a project and then move on to other aspects of the ML development pipeline. As a result revenue can be quite uneven unless they can get to scale with a large number of customers. It will be interesting to see how other providers like Hive and Scale API fare over the coming year.
3. Where were Google and Microsoft?
If you had told me a month ago that Mighty AI was going to be acquired I would have put my money on Google or Microsoft. In the cloud computing wars Google, Amazon, and Microsoft have all emphasized their AI capabilities, but labeling hasn’t gotten as much attention at Google and Microsoft. Seeking a competitive advantage, Amazon recently launched SageMaker Ground Truth to focus on the labeling stage of AI development. This allows them to tout SageMaker as a one-stop shop for AI researchers. Microsoft doesn’t have any labeling offerings and, while Google has rolled out a similar service with their AI Platform Data Labeling Service, its still in beta and relatively limited. I would have expected one of these two, particularly Microsoft, to have scooped Mighty AI up.
I expect this won’t be the last acquisition we see this year.