AI competition shifts toward cheaper, smarter systems

AI companies are moving beyond a race focused mainly on larger models and stronger benchmarks, according to a CNBC report published Friday, July 10, 2026. The article says the market is increasingly centered on systems that can route tasks to the most suitable model based on cost, data needs, and where the work is performed. Perplexity CEO Aravind Srinivas said the product is now the orchestration layer, not just the model itself. The shift could favor openweight models that companies can download and run themselves. Benchmark general partner Peter Fenton said these models may handle a large share of AI token usage over the next 18 to 24 months, which could pressure the pricing power of major model providers such as OpenAI and Anthropic. The report also notes growing interest from enterprises in running smaller models close to their own data, especially in regulated industries. It adds that this trend may influence future data center demand by pushing some routine AI work onto local devices while reserving cloud systems for more complex tasks.