Organizations aiming to leverage generative AI (GenAI) have numerous choices to choose from when it comes to deployment. These include software as a service (SaaS), platform as a service (PaaS), cloud APIs, infrastructure as a service (IaaS), and self-hosted, each with unique advantages and disadvantages. As IT leaders begin to navigate GenAI deployment, there are a series of tactics that can help them make informed decisions to optimize their competing objectives of time to value, control, operational ease and cost.
According to a recent Gartner survey, 95% of CIOs believe in the moderate or extensive potential value of GenAI, with top areas of value being in productivity, customer experience and digital business transformation. As a result, GenAI use cases are proliferating across organizations.
One core challenge of this proliferation of use cases is that the same GenAI solution can be delivered in numerous ways, almost as a spectrum of buy versus build options. IT leaders who do not discern the trade-offs between these delivery models will incur excess costs, risks and reduced value from their implementations.
There are five main methods through which an enterprise can consume GenAI:
-
SaaS: Rapid, subscription-based access to GenAI applications with limited customization.
-
GenAI PaaS: A cloud-based platform that provides developers and enterprises with scalable tools, APIs, and infrastructure to build, deploy, and manage GenAI applications without managing underlying models or hardware.
-
API: Convenient API access to AI models for application building, with varying control levels and pricing per usage.
-
GenAI IaaS: Also known as GenAI cloud infrastructure as a service, it delivers the foundational compute, storage and networking resources, optimized for training and running GenAI models, enabling enterprises to build and scale AI workloads with full control over infrastructure.
-
Self-hosted (on-premises/edge): GenAI self-hosting refers to the deployment and operation of GenAI models and infrastructure, where the organization retains full control over data, customization, security and performance.
As IT leaders navigate these options, they should begin by analyzing the trade-offs for each approach. There are fundamental differences between these methods when it comes to degrees of control, feasibility, choice of models and pricing model.
Differences Across These Approaches
For instance, SaaS provides rapid access to GenAI capabilities but with limited control and customization. While these embedded capabilities potentially accelerate democratization of AI access, IT leaders are struggling to discern hype from reality, adequately audit the security/privacy practices and gauge the long-term innovation potential of the SaaS providers.
IT leaders should choose SaaS when use cases are well-defined and narrow. They can pilot in noncritical areas, assess vendor product quality, gauge the pace of innovation, and audit data privacy and legal indemnification policies.
APIs and PaaS are the most popular way to build custom GenAI applications.
IT leaders should select PaaS when they want to strike a balance between choice, ease of use and customization. Ensure PaaS-specific talent and are actively evolving the architecture to minimize lock-in. APIs provide IT leaders rapid access with lower operational overheads while PaaS offers similar benefits with a wider choice of GenAI models and tools for customizing, automating and securing workflows.
IT leaders should opt for model APIs when experimenting rapidly. They should institute FinOps practices for cost optimization, use AI gateways or similar abstractions to future-proof provider shifts, and implement prompt governance and automation.
Deploying customer-owned models on IaaS or self-hosting them isn’t very common due to operational complexity. However, data gravity, data privacy, latency performance and the need for AI sovereignty could drive model inferencing to be more distributed in the future, further aided by the availability of open and smaller GenAI models.
IT leaders should adopt IaaS when they need a high degree of control and customization. They should use open-source frameworks for model deployment and serving to decouple from cloud provider-specific dependencies to the extent possible.
They should choose self-hosted when they need complete data privacy or custody or require on-premises, air-gapped or edge deployment. Consider hybrid cloud deployments to strike a balance (on-prem for training/customization, cloud for inferencing) and invest in sourcing methods and tools for automation, observability and continuous cost optimization.
Cost Considerations
In the next step of navigating GenAI deployment, IT leaders should consider the cost difference between approaches. Each of the five methods has its own total cost of ownership (TCO) composition.
SaaS applications typically have a fixed price per user. API is based on token usage. PaaS is based on the hourly price of cloud resources, as is IaaS (though just for the infrastructure). Self-hosting includes the cost of procuring and maintaining hardware, premises, software and workforce deployed on-premises or colocation-based pricing.
While there is no one-size-fits-all answer when determining which method yields the highest or lowest TCO, it’s crucial to consider the balance between fixed and variable costs and usage volume.
As GenAI use cases continue to proliferate, IT leaders are tasked with making nuanced decisions that balance speed, control, cost, and innovation. The spectrum of deployment options offers flexibility but also introduces complexity in terms of trade-offs and TCO. There is no universally optimal approach; rather, the right deployment model depends on the organization’s unique requirements, risk appetite, and strategic objectives.
By rigorously evaluating the pros and cons of each method and aligning deployment choices with business priorities, IT leaders can begin to harness the potential of GenAI while mitigating unnecessary risks and costs. Ultimately, a thoughtful, well-informed deployment strategy will be critical to maximizing value from GenAI investments and ensuring long-term success in an evolving digital landscape.