Agentic AI drives enterprises away from public clouds

AI agents are nothing new. AI itself was old but is new again, and AI agents and agentic AI are another chapter. Agentic AI, known for its autonomous decision-making and complex goal management, radically transforms enterprise operations. The significant change is that public cloud providers are often not the desired platform of choice. Companies big and small are instead looking for smaller, more distributed platforms, including on-premises hardware servers and smaller devices. Let’s explore the driving forces behind this transition and its future implications for enterprise AI.

Agents are on the rise

Agentic AI refers to artificial intelligence systems designed with autonomous decision-making capabilities, enabling them to act independently to achieve specific goals. We’ve seen many instances of this idea over the years, more recently with personal digital assistants on our phones and devices and automated everything, from home HVAC systems to automobiles.

These systems possess advanced reasoning, learning, and adaptive functionalities, allowing them to process complex information, make informed choices, and execute tasks without continuous human oversight. Using sophisticated algorithms and vast data sets, agentic AI can analyze environments, predict outcomes, and initiate real-time actions. This form of AI aims to enhance efficiency and effectiveness by providing intelligent, goal-directed solutions across various domains, such as healthcare, finance, and transportation.Historically, public cloud services from AWS, Microsoft Azure, Google Cloud, and others have dominated the cloud landscape. However, the unique demands of agentic AI are now leading enterprises to reconsider and ultimately move away from public cloud solutions for several reasons:

Data sovereignty and security are critical. Especially in regulated industries like finance, healthcare, and government, private clouds or on-premises servers provide greater control over data handling and storage, ensuring compliance and mitigating risks associated with data breaches.Agentic AI applications often require high levels of customization and optimization. Non-public cloud environments allow for fine-tuning infrastructure to meet specific requirements and operate more efficiently. Companies gain improved performance and resource management compared to the standardized offerings of public clouds.The cost structures of public cloud services can be unpredictable and prohibitive. Many enterprises are still reeling from post-pandemic cloud bills that are double or triple what they expected. Things were so bad, an entirely new space, finops, was created. By shifting to their own hardware servers and smaller dedicated devices, businesses can achieve greater cost predictability and control, avoiding the ongoing subscription fees and variable costs inherent in the public cloud.

Agent-based AI means smaller non-cloud systems

Edge computing—executing AI at the network edge on smaller devices—is gaining traction as part of this shift. By processing data locally, edge computing reduces latency and enhances security. Sensitive data remains closer to its source. For example, automotive manufacturers deploy edge AI to process real-time vehicle data, improving performance and safety without solely relying on cloud connectivity.

Decoupled and distributed systems running AI agents require hundreds of lower-powered processors that need to run independently. Cloud computing is typically not a good fit for this. However, it can still be a node within these distributed AI agents that run on heterogeneous and complex deployments outside public cloud solutions.The ongoing maturation of agentic AI will further incentivize the move away from the public cloud. Enterprises will increasingly invest in dedicated hardware tailored to specific AI tasks, from intelligent Internet of Things devices to sophisticated on-premises servers. This transition will necessitate robust integration frameworks to ensure seamless interaction between diverse systems, optimizing AI operations across the board.

Although this will indeed increase complexity, heterogeneity, and operational expenses, it will also lead to more pragmatic AI deployments tailored to the needs of the business. Let’s face it: Enterprises will never build massive large language models for their own use; it’s too expensive, even on a public cloud provider. Agents and small language models are the more likely architectural options for AI.

In this case, doing what’s best for enterprises is not great for public cloud providers. However, don’t cry for them. The cloud is not the most optimized or cost-effective option, but it is still the “easy button” of AI. This makes it attractive to many AI builds and deployments, and many enterprises will select that path regardless.

Integrating agentic AI marks a significant pivot in enterprise strategy, driving companies away from public cloud solutions. By adopting non-public cloud technologies and investing in adaptable, secure, and cost-efficient infrastructure, enterprises can fully leverage the potential of agentic AI. This strategic shift enhances operational efficiency and aligns AI deployments more closely with business-specific needs and goals.