The enterprise AI market is facing a significant challenge as it tries to navigate the complex landscape of large language models (LLMs). With dozens of commercially available models to choose from, senior executives are under increasing pressure to make the right decision. The top five providers

  • Anthropic, AWS, Google, Microsoft, and OpenAI - collectively hold 78% of the market share, which doesn't give executives many other options. According to Mayur Khandelwal, Vice President at EXL, the abundance of choice has become a form of risk.

Mayur Khandelwal leads the Data and AI Practice for Life & Annuities and Group Benefits insurance at EXL. He's witnessed firsthand the paradox of the LLM market, where the abundance of choice is creating more problems than solutions. In a 2025 survey by Global Market Insights, it was found that 37% of enterprises now run five or more models in production simultaneously. This leads to fragmented governance, and it also leads to inconsistent outputs. Additionally, data residency risks aren't something that any compliance team signed off on, and they're a major concern.

The LLM market is evolving rapidly, with new models being released regularly. New models are being released all the time, and this rapid evolution also means that models that lead benchmarks today won't lead them tomorrow. They're routinely displaced within months, which creates a structural risk for organizations that bind their business-critical processes tightly to a single vendor's API. As Khandelwal notes, "The model isn't the strategy." Organizations that have invested in model-agnostic orchestration layers will be far better positioned to switch or blend models as the market shifts. They can do this because business logic, memory, and workflow live independently of any single provider.

The cloud infrastructure market in 2008 is a useful analogy for the current LLM market. At that time, there were many providers, overlapping capabilities, and a confused enterprise buyer base. By 2015, three providers

  • AWS, Azure, and Google Cloud - had captured the dominant share. The rest either specialized, merged, or exited, and the same pattern is likely to play out in the LLM market. Winners will emerge based on deep distribution, ecosystem lock-in, and trust advantage in regulated or sensitive environments. They won't emerge based on anything else, and these factors will be key.

In terms of contract structures, Khandelwal emphasizes the importance of exit flexibility, data portability provisions, and clarity on pricing evolution. What looks like favorable unit economics in 2026 may look very different after the next wave of capability jumps. It might look very different, and that's something that organizations need to consider. For regulated industries, the compliance architecture surrounding a model is often more important than raw performance. Domain-specific or compliance-ready models may maintain durable market positions even as general-purpose consolidation accelerates.

They won't be affected as much, and that's because they're specialized.

  • 78% of the enterprise LLM market share is held by the top five providers: Anthropic, AWS, Google, Microsoft, and OpenAI.
  • 37% of enterprises now run five or more models in production simultaneously.
  • Enterprise LLM API spending more than doubled from $3.5 billion to $8.4 billion in less than a year.
  • The LLM market is evolving faster than most enterprise architecture cycles, with models that lead benchmarks today being displaced within months.

As the LLM market continues to evolve, it's clear that consolidation is inevitable. Organizations that prepare for this by investing in model-agnostic orchestration layers and negotiating flexible contract structures will be better positioned to navigate the changing landscape. They won't be caught off guard, and they'll be able to adapt. In the words of Khandelwal, the key to success lies in adaptability and the ability to swap out models without disrupting business-critical processes. It's not about finding the perfect model, but about being able to switch when necessary.