Why Choice Matters Most In AI

Shane Buckley is President and Chief Executive Officer of Gigamon, a leader in deep observability.

getty​For the past several years, the AI market has largely been framed as a race. Which model will win? Which provider will dominate? Which platform should enterprises bet on?

For CEOs making significant AI investments, the bigger question is what happens when those bets need to change. A model that delivers the best results today may become too expensive or no longer meet the organization’s requirements. AI models and capabilities are advancing at a pace that challenges traditional enterprise planning cycles, even as workloads and business requirements continue to change.

Just as organizations learned that no single cloud environment could satisfy every need, they are finding that no single AI model is best suited to every workload. Different workloads call for different combinations of model capability and deployment environment. Complex reasoning may justify a frontier model’s cost, while routine tasks may be better served by a smaller, specialized model. Data sensitivity and control requirements can also help determine where those models should run.

The result is a hybrid approach in which organizations select and combine models based on capability, cost, performance, security, governance and the requirements of each workload. This shift is already happening, with recent research reporting that 46% of enterprises are using open-weight models as they balance AI costs and flexibility, while continuing to rely on frontier models where capabilities justify the expense.

The hybrid cloud era offers a useful lesson, as early cloud strategies often prioritized speed and scale, with governance and cost controls following later. AI is moving through a similar cycle, but at a faster pace. Enterprises cannot afford to wait to apply those lessons. ​

According to Gartner, by 2028, more than 50% of enterprises will use public cloud selectively for bursty or experimental workloads, while anchoring performance- and sovereignty-sensitive AI workloads in private and localized environments.

As AI moves across more models and environments, portability and interoperability become increasingly important. Leaders need to understand whether applications can move between models and whether data can remain where the business needs it. They also need to be clear whether governance can remain consistent as the underlying provider changes.

Before making a significant AI platform commitment, CEOs should ask their teams a simple question: If a better model became available tomorrow, what would it take to switch? The answer should account for application changes, data access, testing and governance, not just whether another model is available. The goal is to build flexibility into the foundation of enterprise AI rather than trying to introduce it later, when changing direction may be more expensive. ​

Moving data between environments, maintaining integrations and validating model performance can all affect the business case for AI. Security and governance need to be part of that calculation from the outset, particularly as more systems exchange information.

Choice is only valuable if the enterprise can manage the complexity that comes with it. As AI workloads expand across models and environments, moving and duplicating data can become expensive and difficult to manage. Organizations must also maintain visibility and control over increasingly complex data flows. Gigamon research found that 80% of CISOs point to inadequate governance around unsanctioned AI as a top data-security challenge, while 76% say limited visibility into AI-driven traffic is a major barrier to securing AI adoption.

These findings reinforce the trade-off leaders must manage. More freedom leads to additional infrastructure, interactions and data flows that organizations must understand and govern. Those requirements should be incorporated into the ROI calculation from the beginning, alongside the cost of the models themselves. ​

Managing those trade-offs requires a more deliberate approach to deciding which workloads belong on which models and infrastructure, with the flexibility to revisit those decisions as business needs evolve. ​

Organizations should start by defining what each AI workload needs to accomplish, what data it requires, how sensitive that data is, what performance level is necessary and what the economics look like at scale. ​

This builds on a discipline that enterprises should already have developed. Hybrid AI makes that discipline even more important, because the right model depends on what each workload actually requires, and those requirements will continue to evolve as models improve and business priorities change. ​

Token consumption inference costs are also becoming significant operating considerations, as is the infrastructure required to support AI. As AI economics continue to evolve, organizations must take steps to maintain their ability to route workloads toward the model or environment that delivers the best combination of performance, cost and risk. ​

A workload that makes financial sense on a public model today may not make sense tomorrow. As model capabilities and economics shift, enterprises need the flexibility to reassess those decisions without rebuilding their AI infrastructure. Three key principles can help guide that approach: ​

1. Preserve model choice. Avoid architectural decisions that unnecessarily bind critical workflows to one provider. Model capabilities and economics will continue changing, and the enterprise should be able to change with them without requiring significant reengineering.

2. Match infrastructure to risk and value. Evaluate each workload based on its business value, performance requirements, data sensitivity, security implications and total cost. Some workloads will justify frontier models, while others may be better suited to open-weight, specialized or privately-deployed models.

3. Establish consistent governance and visibility. A decentralized AI environment requires a consistent understanding of which systems are operating, what data they access, how they communicate and whether activity remains within policy. This becomes more important as machine-to-machine interactions multiply.

No CEO can predict which AI model will lead the market several years from now, but leaders can decide how difficult it will be to change course. The real payoff is being able to adopt better capabilities, improve economics and maintain control as the market evolves. That ability should be designed into today’s investments.

Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?