As general-purpose AI reaches its limits, industry-specific AI solutions are set to drive the next wave of innovation by leveraging specialized data while navigating legal and competitive barriers.
Unless you have been hiding under a rock, you probably have not escaped any news about the increasing adoption of Artificial Intelligence (AI) into our daily lives. Just recently there was a lot of chatter about the impact of Chinese startup Deepseek on the race for AI dominance.
After ChatGPT’s spectacular growth in 2023, its competitors like Google Gemini, Microsoft Copilot, Claude AI, and others began their own growth trajectories. Almost all of the world’s knowledge published on the open Internet has been skimmed by these different AI applications and is now being utilized by masses of individuals and small businesses globally.
While GPTs have been useful in summarizing all of the world’s openly available knowledge on the Internet, there have been hurdles in fully utilizing their potential for industries and businesses in general. The hurdles are not technical but more on legal and business strategy.
For general-purpose AI such as those from Open AI, Microsoft, Google, and others, they often train using information and data available to the general public, such as those posted on the Web. Occasionally some older material needs to be scanned to digitize those, but there is already a lot of general information on the Internet at the moment.
Unfortunately, many industry sectors such as food and beverage, airlines, insurance, semiconductors, and the like are composed of fiercely competitive companies with their intellectual property (IP). Thus to be useful vertically across a particular industry, any AI solution would need to specialize and train a tailored solution for these sectors, where the industry AI training data may need to remain confidential between the different parties involved.
For example, a group of semiconductor manufacturers might want to exchange best practices to train an AI about improved factory techniques. However, a situation may crop up where they want to contribute training data not available on the Internet but are constrained because of some secrecy considerations.
Fortunately, there are now technologies that allow companies to share their information without revealing it. That may sound illogical but it is not. Developments such as Fully Homomorphic Encryption (FHE) mask the actual data or information but give the AI an idea of what the gist of that information represents or is all about.
There are already industry-specific AI software companies specializing in particular industry sector verticals and can navigate and deliver what these sectors need. Y Combinator says that Vertical AI startups could be ten times the size of the SaaS sector. Perhaps not quite an accurate analogy, but the training that a general practitioner MD doctor would not be as specific as a specialist MD in a particular field. Now think of nuclear energy. Perhaps you might get some information from a general-purpose public AI, but if you work in that industry, you might want to use an industry-specific AI, with training not just from publicly available information, but also confidential corporate data.
Overall, it looks like we will see significant growth for AI vertical software as adoption in these different industry verticals speeds up, albeit at different rates.
- AI-Weekly for Tuesday, May 19, 2026 – Issue 217 - May 19, 2026
- AI-Weekly for Tuesday, May 12, 2026 – Issue 216 - May 12, 2026
- AI-Weekly for Tuesday, May 5, 2026 – Issue 215 - May 5, 2026
