The AI ​​battlefield is shifting, from scale to 'smarter and cheaper' systems.

The AI ​​battleground is shifting! From a competition based on model size to an era of smarter and more cost-effective systems. Discover why open-weighted models are shaking up tech giants.

July 11, 2569 (CNBC) – Over the past two years, the competition in the artificial intelligence (AI) arena has become very easy to measure. That involved looking at who had the larger model, the better benchmark scores, and which company could claim leadership—at least until the next model was launched. However, these traditional measurement criteria are no longer effective.

As businesses transition from experimenting with AI to fully integrating it into their products and processes, the focus is no longer on choosing the best model, but rather on accessing the 'most suitable' model for that specific task, at a reasonable cost, with the necessary data, and in the chosen environment.

This shift is opening the door to a new form of AI competition that focuses less on model scale and more on routing, cost, control, and compute.

"The model alone is no longer the product." Aravind Srinivas, CEO of Perplexity, told CNBC...

"But it's the harness or orchestration system that places the model within a high-potential structure and pairs that model with a wide variety of other tools."

This means that AI products are evolving into 'systems' that can make their own decisions about which model to use, when to use it, and what external tools or company data sources are needed.

• Customer service: You may not necessarily need the most expensive model.
• Solving complex coding problems: High-level models may be necessary.
• Typical internal workflow: Can run on a cheaper open model.
• The more difficult step: The system will then escalate to a model with higher processing power.

"The answer is: Always choose the best option for the job." Srinivas said.

The Challenges of the Giants and the Rise of Open-Weight Models

The emergence of alternative models coincides with Corporate America's tightening of its belts on AI spending, posing a new challenge for OpenAI and Anthropic, two companies that have experienced rapid growth in recent years by selling cutting-edge technology.

This week, Perplexity unveiled a new system for computer-use products built on GLM 5.2, an open model from Chinese company Z.ai. The system is designed to allow the cheaper model to handle most tasks, only calling upon the more powerful model when necessary.

This approach reflects a shift in the overall market, as open-weighted models—which companies can download, tune, and run themselves—are becoming increasingly efficient and cheaper to run than proprietary, closed models from large AI labs.

Peter Fenton, General Partner at the venture capital firm Benchmark. He said this change could happen rapidly and drastically. “A view that might seem counterintuitive at first, but is becoming the consensus, is the belief that more than 90% of tokens created in the next 18 to 24 months, or perhaps even by the end of this year, will come from open models,” Fenton told CNBC. (Note: “Tokens” are units of data that AI models use to process and generate results.)

"I think the inference margins of frontier model companies will face significant pressure when you can run those models without paying them markups, once you have sufficiently good open models available." Fenton added.

Furthermore, Fenton noted that switching to open models isn't just about saving money. In some cases, smaller models optimized for specific tasks can be faster and more efficient than larger, general-purpose models.

'Where and how does it run?'

That's one reason Benchmark decided to invest in Ollama, a company that helps developers and organizations easily download, run, and manage open models.

“One thing is where the model comes from, where it’s built and trained,” said Jeff Morgan, CEO of Ollama. “But even more important for the businesses we’ve talked to is where it runs and how it runs.”

Morgan reports that Ollama is already adopted and used by over 85% of Fortune 500 companies, including those in highly regulated industries such as aviation, insurance, and healthcare. Many companies started with a smaller model running close to their own data, then scaled up to a larger, open model as they became more familiar with it.
National security and the future of data centers.

The growth of open models also poses a strategic challenge to the United States, as many highly competitive open models in today's markets originate from Chinese labs such as Z.ai and DeepSeek. This makes open-source AI a key issue in terms of business, policy, and national competitiveness.

Srinivas (CEO of Perplexity) I think that The U.S. should support open models because they make AI more affordable and accessible.

"If you want the benefits of AI to spread widely among small businesses in America and partner countries, you need to make AI much more affordable, and open source is the only way to do that."

This shift could also impact the massive data center construction currently underway in the technology industry. The current AI boom is based on the assumption that demand will continue to flow into large cloud data centers filled with high-end computing chips. However, Srinivas says that ultimately, some types of AI tasks may run locally on consumer or company-owned devices instead.

While this trend may not completely eliminate the need for data centers, it could create hybrid AI systems where everyday tasks are processed locally, while only the most difficult and complex tasks are sent to more powerful models on the cloud.

For investors. The key question now is: will the leading AI labs be able to maintain their pricing power in a world where open models are becoming increasingly sophisticated and businesses are becoming more selective about what they use?

refer : www.cnbc.com

Read news related to All situations around the world can be found here.