AI capabilities are creating opportunities for enterprises to launch use cases that promise to improve efficiencies or create new revenue streams. Not all AI, however, is the same, which creates a new challenge: Where should “enterprise AI” as well as other AI workloads be hosted?
Let’s break down the types of AI, what they do and where they are used, with a focus on enterprise AI.
Enterprise AI refers to how businesses integrate AI into technologies and tools to improve productivity, decisions, cost savings, customer experience and innovation. These outcomes are linked to fundamental AI capabilities such as automation, prediction and new content generation.
A recent McKinsey survey, The state of AI in 2025: Agents, innovation, and transformation, says that while most organizations are still in the experimentation or pilot phase, 88% of respondents use AI in at least one business function. Only 7% of organizations have “fully scaled” AI, meaning it’s fully deployed and integrated across the business.1
Stanford University’s Artificial Intelligence Index Report 2025 states that “AI is increasingly embedded in everyday life.” Other takeaways are that “business is all in on AI” and 78% of organizations reported using AI in 2024, up from 55% in 2023.2
OpenAI research has an interesting take on adoption. Enterprises are expanding both the extensive margin (more workers adopting AI) and the intensive margin (existing users going deeper). Metrics back up the finding: For example, “75% of workers reported that using AI at work has improved either the speed or quality of their output.”3
AI appeared on the scene in the 1950s. Over time, types or levels of AI were named: narrow AI, strong AI, also called artificial general intelligence, and artificial superintelligence.
Narrow AI, the foundation of subsequent AI forms, is the stage we’re in today. Narrow AI requires model training or tuning using massive amounts of data to improve capabilities for a specific task.
Narrow AI simulates intelligence. It doesn’t think on its own, and it’s not capable of reasoning. Narrow AI can be trained to recognize voices, patterns and images, and traditional narrow AI is specialized and used primarily for processing unstructured data that fuels virtual assistants (Siri, Alexa), recommendations systems and facial recognition.
Generative AI (gen AI) is advanced narrow AI. It uses deep learning and neural network techniques and leverages large language models (LLMs) for text, generative adversarial networks (GANs) for images and variational autoencoders for data generation.4
Gen AI capabilities include recognizing patterns, understanding and using natural language during interactions, summarizing text, problem solving, generating new content – text, video, images – and assisting developers with code.
Nearly 80-90% of new use cases are generative AI.
Strong AI and superintelligence don’t exist, but theoretically they could operate independently. Hypothetically, strong AI is human-like in areas such as self-awareness, learning, reasoning, adapting and decision making. Superintelligent AI is more cognitively advanced than humans.
Where AI workloads are hosted depends largely on the intended use, enterprise preferences and regulatory requirements. Here is a high-level breakdown that spans both narrow AI and gen AI:
Let’s look at examples of narrow AI and gen AI before exploring a few more types of AI that may be less familiar.
Narrow AI is touted mainly for its ability to complete tasks faster and more accurately than humans. The examples below are included because they are still single-purpose, specialized tasks. If you think this comment infers that gen AI is being incorporated into some of the examples, you’re right.
Gen AI resides in some of the same places as narrow AI, but think of it as upleveled, with more flexibility and more skills. A few gen AI locations:
Tools like ChatGPT and Google process billions of queries daily.
Now that we’ve covered the narrow AI and gen AI basics, let’s look at other forms of AI that use or leverage narrow AI and/or gen AI. Advanced AI systems can combine multiple types of AI.
Agentic AI systems are designed to work autonomously to pursue their assigned goals. Enterprises can build, deploy and manage AI agents with tools available in leading public clouds. AI agents can make decisions about what to do and how to do it by following multiple steps using tools and APIs and coordinating work with people or other agents. Agentic AI resides in some of the usual locations:
74% of companies plan to deploy agentic AI within two years.
Deloitte
Multimodal AI (narrow AI or gen AI depending on its design) understands and processes multiple types of data simultaneously – text, images, audio, video and so on. ChatGPT-4 and Google Gemini are multimodal designs that live in cloud platforms, and enterprises can tap their capabilities to create content, analyze videos, chat and so on.
Physical AI controls machines or robots that interact with the physical world, compared to AI that exists (virtually) only in software or digital environments. Initially built decades ago using narrow AI, some modern systems incorporate gen AI, deep learning and multimodal capabilities:
58% of companies already use physical AI to some extent…setting the pace for the next wave of industrial automation.
Source: Deloitte
Collaborative AI assists humans as experts in drudgery. What else would you call repetitive tasks or scanning financial transactions to find potential fraud or analyzing data to predict inventory requirements? Typically, collaborative AI is narrow AI designed to focus on collaboration, but some systems incorporate gen AI for reasoning or other capabilities. Where collaborative AI lives:
Sovereign AI is a service – typically involving gen AI – that operates within the borders of a nation to ensure compliance with local data privacy laws, governance and regulations.6 A McKinsey expert describes it as “the intelligence layer that you build on top of your data.”7 The intent of sovereign AI is to control the entire AI lifecycle, the intelligence, the hardware, data and infrastructure to reduce dependencies on foreign vendors and potential adversaries.
77% of companies say the location of AI development is a key factor when choosing new technologies, signaling that geographic sovereignty is now as important as innovation.
Now that we’ve arrived at the end of this succinct overview of enterprise AI, Bill Bryson’s “A Short History of Nearly Everything” comes to mind. Short is relative. AI isn’t everywhere, yet, but enterprises are implementing AI in many use cases.
Where AI lives and works depends on myriad factors, including latency, computing power, data storage, regulations, bandwidth and maintenance. Industry experts sometimes call AI a force multiplier to help humans be more productive and effective – not to mention challenged, entertained, informed and inspired.
Learn More
Check out the links below to learn more about AI or visit the Knowledge Base.
References