The artificial intelligence industry is shifting from a focus on massive, general-purpose models toward specialized “orchestration” systems, according to industry leaders. Rather than relying on a single, expensive frontier model, enterprises are increasingly routing tasks to a mix of open-weight and proprietary models based on cost, data requirements, and specific performance needs. This transition marks the end of the “bigger is better” era as companies prioritize efficiency and control.
The Shift to AI Orchestration
For two years, the AI race was measured by benchmark scores and model size. Today, the competitive landscape is defined by how well a product manages different models. Perplexity CEO Aravind Srinivas told CNBC that “the model alone is no longer the product.” Instead, companies are building “harnesses”—orchestration systems that determine which model is best suited for a specific task.
This strategy allows businesses to optimize costs. Simple customer service queries might run on a small, inexpensive model, while complex coding tasks are escalated to more powerful, high-end models. By pairing models with specific tools and private company data, firms gain better control over their workflows without overspending on premium, general-purpose technology.
Perplexity recently previewed a new computer-use system built around the GLM 5.2 open model from China’s Z.ai, demonstrating how cheaper models can handle routine work while calling on stronger models only when necessary.
The Rise of Open-Weight Models
Open-weight models—which companies can download, tune, and run internally—are rapidly gaining market share. Benchmark general partner Peter Fenton predicts that “90-plus percent of the tokens created will come out of open-weight models” within the next 18 to 24 months. As these models improve, they place significant pricing pressure on frontier model companies that rely on high-margin, proprietary software.
Beyond cost savings, open models offer logistical advantages. Ollama CEO Jeff Morgan noted that for the 85% of Fortune 500 companies using their platform, the primary concern is “where it runs and how it runs.” By keeping models close to their own data, businesses in regulated sectors like health care and insurance can maintain better security and performance.
Strategic and National Implications
The growth of open-source AI has transformed into a complex policy issue. With highly competitive open-weight models emerging from Chinese labs such as Z.ai and DeepSeek, the U.S. faces a strategic challenge regarding national competitiveness.
Srinivas argues that the U.S. should embrace open models to make the technology more accessible. According to the Perplexity CEO, if the benefits of AI are to reach small businesses, the technology must be affordable, and “open source is the only way to do that.” This evolution could also reshape the tech industry’s data center strategy. While the current boom relies on massive cloud infrastructure, a hybrid approach—where routine tasks run locally on consumer or business devices—could eventually reduce the strain on centralized cloud resources.
Pro Tips for AI Adoption
- Start Small: Begin by running smaller, specialized models close to your data to assess performance.
- Prioritize Orchestration: Look for platforms that allow you to route tasks between different models rather than committing to a single provider.
- Focus on Efficiency: Don’t pay for the most powerful model if a smaller, tuned model can handle the task effectively.
Frequently Asked Questions
- What is an open-weight model?
- It is an AI model that can be downloaded and run by an organization on its own servers, rather than being accessed exclusively through a third-party cloud provider.
- Why are companies moving away from the largest models?
- Companies are seeking to reduce costs and gain better control over their data by using specialized, smaller models for routine tasks instead of high-cost, general-purpose models.
- How does “routing” work in AI systems?
- Routing is an orchestration method where the system evaluates a task and automatically directs it to the most appropriate model—either a fast, cheap one or a more powerful one—based on the complexity of the request.
Are you adjusting your AI strategy to include more open-source or specialized models? Share your thoughts in the comments below or subscribe to our newsletter for more updates on the future of enterprise technology.
