The $25B Gap: Why the AI Market Is Overpaying for Intelligence

The AI market is paying a premium it no longer needs to pay.

Recent research from MIT and Georgia Tech suggests that open models now achieve roughly 80–90% of the performance of leading closed models, often at a fraction of the cost. For most enterprise workloads that performance is more than enough.

And yet enterprise AI spend remains heavily concentrated on closed platforms from providers like OpenAI and Anthropic. Some estimates suggest that shifting workloads toward open models could unlock $25 billion in annual savings.

When a market continues to pay a premium for capability it could get more cheaply elsewhere, something other than capability is driving the decision.

What Enterprises Are Actually Buying

Closed platforms have succeeded by making advanced AI accessible through reliable APIs backed by large-scale infrastructure and mature ecosystems. For many enterprises, the appeal extends beyond model performance itself to include trust, familiarity, support, and reduced operational risk.

What enterprises are really paying for then isn't just intelligence. They are also paying for trust, perceived safety, established ecosystems, and the reduced risk that comes with adopting a market leader.

The open model experience, by contrast, delivers lower cost and greater flexibility, but only to teams equipped to manage the system themselves. In practice, this often means fragmented architectures, inconsistent performance, and a level of ongoing engineering burden that most organizations can't justify.

Faced with that tradeoff, many organizations continue to choose closed platforms. Not necessarily because they deliver the most value per dollar, but because they reduce perceived risk and complexity.

Why This Looks Like a Market Inefficiency

In a stable market, prices reflect underlying value. The current configuration of AI spend doesn't. The capability gap between open and closed models is small and shrinking. The cost gap is large and persistent.

Research suggests that the persistence of this gap is influenced by factors such as switching costs, brand trust, information frictions, and operational complexity. Many organizations continue paying a premium for closed platforms even as lower-cost alternatives become increasingly viable.

"This creates the conditions for a market inefficiency: a situation where price differences remain larger than underlying differences in capability."

What Closes the Gap

The factor that would unlock this reallocation isn't another model release. It's the missing infrastructure layer that makes efficient models practical at production scale.

Three things have to be true for that shift to happen: open models need to be operable without specialized teams — not just available, but as easy to deploy and manage as a closed-model endpoint; system-level performance has to match the closed-model experience, as latency, reliability, and consistency are non-negotiable in production; and the economics need to be visible, because as inference becomes a meaningful share of revenue at scaling AI companies, cost transparency becomes a strategic priority.

All three are now in motion. While several factors contribute to closed-model dominance, operational complexity remains one of the most important barriers preventing enterprises from realizing the economic advantages of open models.

The market conditions for reallocation are forming. This is where the next phase of AI shifts: combining open-model economics with enterprise-grade operational simplicity.

‍