AI value optimization is now making enterprises think about their language model approach

Just months ago, the conversation for enterprises was - if and how the organization would adopt artificial intelligence. Today, AI pilots have moved into large-scale deployments, sending token consumption soaring.

As rising AI consumption was viewed as proof that the organization was embracing the future; today, for enterprise leaders, “Are we getting enough value for every dollar we spend?” is a confounding question to which they are looking for answers.

Organizations are today not excited about access to the most powerful model, but figuring out use of the right model for the right task at the right cost. “I'm so glad we're talking ROI now,” says Puneet Chandok, President of Microsoft India and South Asia, who now sees more enterprises realizing the need for an open, heterogeneous, diverse ecosystem of models to work.

“So most CFOs and CEOs tell me it's not about spending less, it's about spending intentionally. How do we use frontier solutions for frontier problems? How do we make sure that we have an open, heterogeneous system where multiple models work together to give us the quality we need based on the use case, based on the query? And that's exactly the work we're doing with Foundry,” he added.

This shift towards AI value is something that IT firms and systems integrators are already looking to embrace.

At the company Q1FY27 earnings call, TCS CEO & MD K. Krithivasan observed, "We believe most enterprises will have multiple models, one LLM plus many SLMs or multiple LLMs within the same family," where organizations would deploy an ecosystem of models, choosing between large and small models depending on the complexity of the task, the cost involved, and the value being generated." Krithivasan further pointed out that this is an area where system integrators have a greater role to play and ensure that the client's money is well spent, and the value is delivered to them.

The challenge now is not just implementing AI, but designing architectures that can balance performance, cost, governance, and security across multiple models and environments.

In the AI technology cycle, as adoption reaches scale and token prices are gradually declining, AI usage is growing at an even faster pace. According to Gartner, AI inference costs per agentic workflow are expected to increase more than fivefold by 2028.

The business and technology insights company says that although foundational AI models are becoming cheaper and more efficient, enterprises are using increasingly powerful models and deploying more complex, agentic workflows that consume significantly more tokens.

As a result, overall AI inference costs are rising, making cost optimization and multi-model management critical priorities for organizations.

As C. Vijayakumar, CEO & MD HCL Tech, said during the company’s earnings call, "The tiered approach is becoming the most popular enterprise AI architecture, both for data sovereignty for the enterprises and for the right price performance."

AI architecture strategies where smaller language models would sit closer to enterprise data, handling routine and domain-specific tasks, while larger models would be reserved for the most complex tasks and governance and security layers determining which model should be used in each situation.

Similarly, Infosys also sees the same trend, where companies were increasingly asking whether they could use, a less parameter model, also less expensive model or even an older version of some of the big company models for some tasks and the most recent one for some very specific, high-end type of task, as Salil Parekh, CEO & MD Infosys, noted in the company’s analyst call.

Further, he said that the Topaz Fabric was built to remove complexity, working with 15 different models, the platform makes those decisions automatically, balancing capability, speed, and cost.

With the initial AI adoption excitement now giving way to business fundamentals, enterprise AI adoption will not hinge on models, agents, or tokens consumed, but increasingly on contextual intelligence coupled with governance and turning AI pilots into measurable business outcomes and sustainable value.