Market analysts have identified Automated Machine Learning tools and Machine Learning Operations as two areas of the broader ML market that will grow rapidly over the next few years.
A recent report from AI and machine learning-focused analyst firm Cognilytica notes that there are five primary categories of solutions for machine learning model development: machine learning toolkits, machine learning platforms, analytics solutions, data science notebooks and cloud-native machine learning as a service (MLaaS) offerings.
"When we were doing our primary research, we noticed recurring themes for ML development where people were not just using one kind of tool, but multiple different kinds depending on their circumstances and the roles," said Ronald Schmelzer, co-founder and analyst at Cognilytica.
Cognilytica estimates the market for machine learning platforms as $23.2 billion in 2019 growing to $126.1 billion by 2025, which represents a 33.73% compound annual growth rate, Schmelzer said.
The key reason why developers should care about the ML market is because they have a lot of confusing choices to make when deciding which sort of ML platform to use to develop ML models.
"Between open source ML toolkits, cloud-based solutions, soup-to-nuts ML platforms, proprietary analytics apps and the array of data science notebooks, trying to figure out how to sort out the mess is confusing," Schmelzer said. "By separating these solutions into various areas that are focused for different use cases and situations, it helps to understand when to use one tool with or versus another."
But other analysts see the market differently. "We think that is way too high," said Mike Gualtieri, an analyst at Forrester Research. Forrester sees the entire AI software market growing to only $37 billion by 2025.
"Investors and other analysts have projected that the AI software market will be huge -- $150 billion to $200 billion in the next five years or so," stated a recent Forrester report. "But we think these projections define the market too hyperbolically because they mistakenly include categories that are only loosely influenced by or distantly adjacent to AI software."
AutoML on the rise
Meanwhile, Automated Machine Learning (AutoML) tools are being increasingly adopted by companies that find it hard to hire data science talent or want to make their existing data science teams more efficient.
"AutoML tools solve many of the problems that data scientists usually need to use their expertise to solve, from machine learning algorithm selection to model tuning, data preparation, and other tasks," Schmelzer said.
Companies such as H2O and DataRobot pioneered AutoML, but it can now be found across many of the machine learning platform provider companies, as well as open source solutions, cloud-based ML-as-a-service, and proprietary analytics tools.
Ronald SchmelzerAnalyst, Cognilytica
More market consolidation
In addition, Cognilytica said it sees accelerated consolidation of the machine learning platform market with the pace of acquisitions, mergers and IPOs increasing in the year ahead.
Schmelzer said there will be "significant consolidation" in the ML marketplace in 2021 and beyond.
"DataRobot is on a tear with multiple acquisitions and running toward an IPO no doubt soon," Schmelzer said. "Likewise, companies like Alteryx and Dataiku are growing by widening their platforms. The Cloud ML vendors are likewise growing through acquisition and consolidation with Microsoft, AWS and Google leading the way.
But the major players are and will remain the public cloud vendors -- AWS, Azure and Google, said Holger Mueller, an analyst at Constellation Research.
"It's interesting because it is the only area where AWS has clearly had a late start," he said. "And the third-place player, Google, is three to four years ahead in terms of putting algorithms on customer silicon. Speed and cost of AI processing win here."
In addition, Cognilytica projects there will be increased MLOps capabilities within existing machine learning platform solutions as the scope of ML platforms continue to expand.
ML Ops refers to the use of machine learning models by DevOps teams, as well as the process of managing a machine learning model's development lifecycle.
"In the past, companies have cobbled together their own tools to deal with these issues, but now a whole field of companies is emerging to deal with these model lifecycle challenges such as model drift or data drift," Schmelzer said.