The artificial intelligence (AI) market will not develop the dynamism in the coming years that is often predicted. Many of the AI developments cannot be monetized. They are simply used to expand and improve existing business applications. Companies which develop AI applications for their use will usually generate no or at most indirect sales. According to Forrester, the development of AI business has slowed down in the current pandemic year – as has the software business as a whole. Demand may rise strongly again by 2023. AI is increasingly being absorbed by other software products and no longer plays a major role as a differentiator between the products of different providers. Revenues from applications and tools, for which “AI-Inside” was useful as a sales argument for a while, should then even decrease. AI becomes a commodity.
None of this will change the fact that AI will form a technical basis for future software products. All those software manufacturers who set the tone today are likely to have AI under their hood in the foreseeable future so that customers will hardly be prompted to switch to “more intelligent” solutions from other providers. Wherever AI platforms and solutions are bought today, users resort to their classic applications again, which are more powerful due to “embedded AI”.
One segment includes providers who provide tools and platforms that enable AI teams to create highly customized AI solutions for almost any use case. These include first and foremost the cloud hyperscalers Amazon, Microsoft, Google and Alibaba. There are also special providers such as Cloudera, Domino Data Lab, dotData, Google, H2O.ai, IBM, MathWorks, Microsoft, RapidMiner, SAS and so on. The latter offers special tools on their platforms, for example for data engineering, model development or application integration and provisioning. The providers in this segment often fall back on innovations from the open-source community.
---
In the AI platform business, hyperscalers benefit from the same effects as with Infrastructure as a Service (IaaS): massive scaling and cost advantages. Also, customers can relatively easily integrate their AI products with applications that are already running on their large IaaS platforms. The link with the existing Platform-as-a-Service (PaaS) offers for software development is just as obvious.
The specialized AI maker platforms, however, can tailor solutions specifically to industries, countries or market segments. Over time, however, the hyperscalers are likely to exploit their economies of scale and resources, which will thin out the market for AI maker platforms considerably.
The AI market has a special feature: it will be the software houses themselves that will make up a significant part of this volume. The analysts estimate that, for example, providers of database systems, integration tools or BI/analytics applications generate around ten per cent of their revenues with customers in the software industry. In the case of AI software, this proportion of revenue should be around a third.
The conclusion from the market research is: AI will be used in the future to optimize existing applications. It will find its way into most business applications and become the norm there. Of course, there will also be pure AI applications or applications that are pimped up using AI and thus more expensive, but their sales will not even reach three per cent of the total market for applications, and the trend will be down from 2023. Overall, AI platforms are becoming part of an increasingly broad middleware market and are ranked alongside database and storage management systems, integration software, security offers and development platforms.
For end-users with first-mover ambitions, the question of how to deal with AI. If you want to react quickly to company or market-specific market opportunities or if you want to achieve a competitive advantage, the analysts recommend the building approach. This group could occasionally also be helped by AI-centric applications or standard applications with a chargeable AI-based added value – reinventing the wheel rarely makes sense. The buy approach is more practicable, but only if companies have done their homework in terms of data quality and management.
Machine learning is based on advanced data management, so it needs clean operational, customer and market data. Anyone who wants optimal results depends on permanently optimal training data, which AI applications can be used to train over and over again. Companies should keep this in mind and some cases perhaps choose a provider who has already successfully trained its algorithms with data from other customers.
