It’s tough to make predictions, especially about the future – it’s a phrase popularly attributed to legendary New York Yankees catcher Yogi Berra. That wisdom certainly applies to the machine learning (ML) market. We don’t like making predictions about technology markets or pretending we know better than anyone else. However, it’s the end of the year so we need to give you something 😉 We decided to outline a few non-obvious trends that might play an important role in the ML market in 2022. Calling these predictions might be an overstatement, but it’s fun nonetheless. Here are five that might make you think:
MLOps Consolidation: Money was free in 2021, and many MLOps startups raised capital at astronomical valuations. It is fairly improbable that all those startups will be able to live up to those valuations. The market is just not that big. As a result, we should see a strong M&A and consolidation trend in 2022 within the MLOps space.
One-two ML Companies Go Public, but the Trend Slows Down: SPACs are losing momentum, and public markets face quite a bit of uncertainty. C3.ai was the first standalone ML company to go public, and its performance has been average so far. Factoring all that, it is conceivable that we will see a couple of IPOs of ML companies in 2022, but the trend is likely to slow down. DataRobot and Scale AI seem to be well-positioned to capitalize on successful IPOs.
ML-Hardware Acquihires: In 2021, we saw a lot of funding going into ML hardware startups. However, we think most of those companies will soon face the realities of the hardware market in terms of capital intensity and long production cycles. As a result, several of those companies are likely to be acquired by players like NVIDIA, Tesla, Google, or Intel, which have been really aggressive in their ML-hardware agenda.
GPT-3 Moment for Computer Vision: It seems that we are almost there. Transformers are making major inroads in computer vision just like they did in natural language. Meta (Facebook), Google, Microsoft, and OpenAI have been pushing the boundaries of computer vision with transformer architectures. We shouldn’t be surprised if next year we see a production-ready transformer model at the scale of GPT-3 be released for computer vision scenarios.
Value Start Moving from ML Infrastructure to ML Services: The current ML market is dominated by infrastructure players in terms of funding allocated, the number of companies, and value captured in general. Like any technology market, the value should fluctuate between the infrastructure and application layers. In that sense, we think 2022 will see the emergence of ML-first applications/services that achieve meaningful market traction. Companies like OpenAI with its API, Microsoft with Cognitive Services, Google, Meta (Facebook), and Salesforce seem well-positioned to capitalize on that trend.
Well, there you have five non-obvious trends of the ML market in 2022. Let’s revisit them periodically next year and see how we are doing. Before concluding, we would like to thank you for your support during this year. Hope we were able to maintain high levels of content quality and that you found this newsletter instructive and helpful. We hope to have you with us next year.
Thanks and Happy Holidays 💜
🔺🔻 TheSequence Scope is our Sunday free digest. To receive high-quality educational content about the most relevant concepts, research papers, and developments in the ML world every Tuesday and Thursday, please subscribe to TheSequence Edge 🔺🔻
🗓 Next week in TheSequence Edge:
We will go light and surprise you ;)
Now, let’s review the most important developments in the AI industry this week
🔎 ML Research
OpenAI trained a version of GPT-3 to answer open-ended questions based on how users search information online →read more on OpenAI blog
Partially Local Federated Learning Models
Google Research published a paper proposing a technique to improve personalization and privacy in federated learning models →read more on Google Research blog
Common Sense Language Understanding
Microsoft Research published a paper detailing KEAR, a model that exhibits signs of common sense reasoning in language understanding tasks →read more on Microsoft Research blog
Mirroring Natural Speech
Microsoft Research published a paper proposing Uni-TTSv4, a text-to-speech model that shows no difference with natural voice recordings >→read more on Microsoft Research blog
🤖 Cool AI Tech Releases
New TorchVision API
PyTorch introduced a new API for its computer vision framework TorchVision with multi-weight support →read more on PyTorch blog
😎 Awesome stuff
We highly recommend the Reading list we put together for you from all our interviews with ML practitioners. Enjoy the reading and share this list with your friends!
You can also follow us on Twitter where we share all our recommendations in bite-sized form
💸 Money in AI