The AI race is no longer just about the biggest and best models. As artificial intelligence moves from experimental labs to real-world applications, a new competitive landscape is emerging, one that prioritizes efficiency, cost-effectiveness, and adaptability. This pivot marks a significant shift from the previous focus on raw model power and benchmark victories.
“The model itself is no longer the ultimate product,” stated Aravind Srinivas, CEO of Perplexity. “It’s the system that orchestrates the model, integrating it seamlessly with a suite of tools and data sources.” This indicates that the value proposition of AI solutions is evolving from a singular, powerful model to a comprehensive system capable of intelligent routing and resource management.
This means AI products are increasingly designed to dynamically select the most appropriate model for a given task. For instance, a routine customer service inquiry might be handled by a less computationally intensive, more affordable model, while a complex software development challenge would necessitate a more robust and capable AI. Similarly, internal workflows could leverage cost-effective open-source models, with more demanding tasks seamlessly escalated to premium proprietary options. Srinivas emphasizes, “The optimal approach is always to utilize the best tool for the specific job at hand.”
This maturation of the AI market arrives at a time when many enterprises are re-evaluating their AI expenditure. For prominent AI developers like OpenAI and Anthropic, who have capitalized on offering cutting-edge technology, this shift presents a new set of challenges.
Perplexity recently unveiled a new system for its AI-powered computer application, built around GLM 5.2, an open-source model from China’s Z.ai. The design prioritizes using a more economical model for the bulk of operations, reserving powerful models for when they are strictly necessary. This strategy aligns with a broader market trend where open-weight models, which can be freely accessed, customized, and deployed by companies, are rapidly increasing in capability and becoming a more cost-effective alternative to proprietary offerings from leading AI labs.
Peter Fenton, a General Partner at Benchmark, believes this transition could be profound. “A view that is gaining traction, and might be considered contrarian by some, is our expectation that over the next 18 to 24 months, possibly even by the end of this year, more than 90% of generated tokens will originate from open-weight models,” Fenton shared. Tokens are the fundamental units of data that AI models process and generate. He further elaborated, “The profit margins for frontier model providers are likely to face pressure as businesses gain the ability to run these models without the added markup, especially when sufficiently capable open-weight alternatives become widely available.”
Fenton also highlighted that the adoption of open models is not solely driven by cost savings. In many scenarios, smaller, task-specific models can outperform larger, general-purpose models in terms of both speed and accuracy.
### The Importance of Deployment: ‘Where and How It Runs’
This emphasis on practical deployment is a key reason behind Benchmark’s investment in Ollama, a company focused on simplifying the process for developers and enterprises to download, run, and manage open-source AI models.
“While the origin and training data of a model are important factors, the more critical consideration for businesses is often where and how the model operates,” explained Jeff Morgan, CEO of Ollama. Morgan noted that Ollama has seen widespread adoption, with over 85% of Fortune 500 companies integrating the platform. This includes organizations in highly regulated sectors such as aviation, insurance, and healthcare. He observed a common pattern where companies begin with smaller, locally deployed models to handle sensitive data, and then gradually scale up to larger open-source models as their confidence and expertise grow.
The increasing prominence of open-source AI models also introduces a strategic consideration for the United States. A significant number of the most competitive open-weight models are emerging from Chinese research institutions, including Z.ai and DeepSeek. This development elevates open-source AI from a purely technological domain to a matter of business strategy, policy, and national competitiveness.
Srinivas advocates for U.S. support of open-source AI models, arguing that they are instrumental in making AI more affordable and broadly accessible. “To ensure that the benefits of AI are equitably distributed among small businesses in America and allied nations, it’s imperative that AI becomes significantly more affordable. Open-source development is the most viable path to achieving this goal,” he stated.
This market shift could also impact the massive infrastructure buildout currently underway across the technology sector, particularly concerning data centers. The prevailing narrative of the AI boom assumes a continuous demand for large cloud data centers populated with high-end computing chips. However, Srinivas suggests that a portion of AI processing may eventually migrate to local devices, whether owned by consumers or businesses. While this would not eliminate the need for centralized data centers, it could foster a more hybridized AI ecosystem, with routine tasks executed locally and more computationally intensive operations delegated to powerful cloud-based models.
For investors, the central question remains: can the leading AI developers maintain their premium pricing strategies as open-source models continue to advance and businesses become increasingly discerning in their AI adoption?
Original article, Author: Tobias. If you wish to reprint this article, please indicate the source:https://aicnbc.com/23631.html
