For the past two years, the artificial intelligence industry has been driven by a familiar narrative: build bigger models, achieve higher benchmark scores, and claim the top spot until the next breakthrough arrives.

Today, that dynamic is changing.

As organizations move beyond experimentation and begin integrating AI into real business processes, success is becoming less about using the most powerful model available and more about choosing the right model for each specific task. Cost, performance, data governance, deployment environment, and operational efficiency are now just as important as raw capability.

AI Is Becoming an Orchestrated System, Not a Single Model

According to Perplexity CEO Aravind Srinivas, the competitive advantage no longer lies in the model itself.

Instead, value comes from the orchestration layer that determines which model should be used, when it should be called, and what additional tools or enterprise data sources should be involved.

Rather than relying on a single large language model for every request, modern AI systems are increasingly designed to route workloads intelligently:

Simple customer support requests can be handled by smaller, low-cost models.
Internal business workflows may run on open-source models deployed within the company.
More demanding tasks, such as software development or complex reasoning, can automatically escalate to premium frontier models.

This approach allows organizations to optimize both performance and operating costs while maintaining flexibility.

Open Models Are Gaining Momentum

The transition is also being accelerated by the rapid improvement of open-weight AI models.

Unlike proprietary models, open-weight models can be downloaded, customized, and deployed on an organization’s own infrastructure. This gives businesses greater control over their data while significantly reducing inference costs.

Perplexity recently demonstrated this strategy by previewing a computer-use system powered primarily by GLM 5.2, an open model developed by China’s Z.ai. The architecture uses the lower-cost model for most operations while invoking more powerful models only when additional reasoning is required.

Industry investors believe this represents a structural shift rather than a temporary trend.

Benchmark partner Peter Fenton predicts that more than 90% of AI tokens processed over the next 18 to 24 months could come from open-weight models, driven by their lower operating costs and rapidly improving performance.

In many practical scenarios, smaller models optimized for a single task can even outperform much larger general-purpose models while responding faster and consuming fewer computing resources.

Deployment Matters as Much as Performance

Another major factor reshaping enterprise AI adoption is deployment flexibility.

Companies increasingly care less about where a model was originally trained and more about where it runs inside their own infrastructure.

This explains the growing adoption of platforms like Ollama, which simplify running open AI models locally or within private enterprise environments.

According to Ollama CEO Jeff Morgan, the company has already been adopted by more than 85% of Fortune 500 organizations, including businesses operating in highly regulated sectors such as healthcare, aviation, and insurance.

Many enterprises begin with lightweight models running close to their own data before gradually expanding to larger open models as confidence grows.

The Strategic Importance of Open AI

The growing strength of open-weight models also introduces broader geopolitical implications.

Several of the most competitive open models now originate from Chinese AI laboratories, including Z.ai and DeepSeek, making open-source AI an increasingly important topic not only for businesses but also for governments and national technology strategies.

Srinivas argues that affordable open models are essential for ensuring AI benefits extend beyond large technology companies.

Lower deployment costs make advanced AI accessible to smaller businesses while encouraging broader innovation across industries.

A More Hybrid AI Infrastructure

The rise of efficient models may also reshape how AI infrastructure evolves.

While cloud data centers will remain essential for the most computationally intensive workloads, many everyday AI tasks may increasingly run directly on enterprise servers, laptops, or edge devices.

Instead of sending every request to expensive cloud infrastructure, organizations are likely to adopt hybrid architectures where:

Routine inference runs locally.
Sensitive workloads remain inside private environments.
Only the most complex reasoning tasks are routed to frontier models in the cloud.

This approach improves latency, reduces infrastructure costs, and gives organizations greater control over sensitive information.

What This Means for Businesses

The AI market is entering a new phase where orchestration, deployment strategy, and cost optimization are becoming stronger competitive differentiators than model size alone.

Organizations that build flexible AI architectures capable of combining proprietary and open models will be better positioned to scale AI adoption sustainably while controlling operational costs.

For software development teams and technology leaders, the focus is shifting from choosing the best model to designing intelligent AI ecosystems that select the best model for every workload.

That evolution may ultimately define the next generation of enterprise AI.

Source