AI Competition Shifts From Model Size to Cost and Orchestration

The New AI Battleground

The artificial intelligence industry is entering a new competitive phase. For two years, the race centered on building the largest, most capable models. Now, as companies move AI from experimentation into production workflows, the calculus has changed. Success increasingly depends on matching the right model to each specific task, at an acceptable cost, with proper data controls.

This evolution is reshaping how AI products work. Rather than relying on a single powerful model, systems now orchestrate multiple models, deciding which to deploy based on task complexity and budget constraints. A routine customer service query might run on an inexpensive model, while a complex coding challenge gets routed to premium infrastructure.

"The model alone is no longer the product," Perplexity CEO Aravind Srinivas explained. "It is the harness, the orchestration system that puts the model inside a very capable harness and pairs the model with a lot of tools."

Open Models Gain Ground

This shift coincides with growing corporate scrutiny of AI spending and the rising capabilities of open-weight models—AI systems that companies can download, customize, and run on their own infrastructure. These alternatives cost less to operate than proprietary models from leading labs.

Perplexity recently previewed a system built around GLM 5.2, an open model from Chinese company Z.ai, designed to handle most tasks with a cheaper model while escalating only when necessary.

Benchmark general partner Peter Fenton predicts the trend will accelerate dramatically. He told the network that 90% or more of AI tokens generated could come from open-weight models within 18 to 24 months, potentially by year-end. This would pressure the profit margins that frontier model companies currently enjoy.

Enterprise Adoption Accelerates

Ollama, a platform that simplifies running open models, reports adoption by more than 85% of Fortune 500 companies, including firms in aviation, insurance, and healthcare. CEO Jeff Morgan notes that businesses care deeply about where models run and how they integrate with existing data infrastructure.

"One thing is where the model's from and where it was created and trained," Morgan said. "But the more important thing to these businesses we speak to is where it runs and how it runs."

Many enterprises start with smaller models operating close to proprietary data, then expand to larger open alternatives as confidence builds.

Why It Matters

This transformation challenges the business models of OpenAI, Anthropic, and other frontier labs that have thrived by selling premium access to cutting-edge technology. If open models can handle most production workloads at lower cost, the pricing power of proprietary systems faces pressure. The shift also introduces geopolitical complexity, as many competitive open models originate from Chinese labs. For infrastructure investors, the trend suggests AI deployment may become more distributed, with routine tasks running locally on enterprise hardware rather than exclusively in massive cloud data centers.

These details were first reported by CNBC.