Chinese AI Quietly Erodes OpenAI-Anthropic Dominance

US companies are beginning to change their strategy for using artificial intelligence (AI). After a period of racing to use the most advanced AI models at high cost, many firms are now opting for cheaper models to reduce expenditure. This strategic shift has the potential to redraw the global AI industry’s competitive map. Companies are no longer relying on a single provider like OpenAI or Anthropic, but are instead combining various AI models according to their needs, including low-cost models developed in China.

At Cursor, a software development startup that supports various AI models, this change in mindset is clearly visible. According to Cursor’s Chief Field Technology Officer, Mike Saeks, the most expensive AI models are not always necessary for day-to-day work. ‘It’s like driving a Lamborghini just to go to the shop to buy milk, when that car is designed for the racetrack,’ Saeks said, as reported by The Wall Street Journal on Saturday.

Saeks joined Cursor about two months ago after a career as an investment banker. The company, which is reportedly being acquired by Elon Musk’s SpaceX for US$60 billion, tasked him with helping companies measure and improve their return on investment (ROI) from AI spending. Cursor calls this approach ‘tokenomics’, referring to the tokens that are the primary unit for measuring AI usage.

According to Saeks, the shift in corporate mindset is happening very rapidly. Just a few months ago, companies were proud of how many AI tokens they consumed. Some even awarded employees for the highest token usage via internal leaderboards. Now, the focus has switched to saving on AI costs. This change not only impacts the scale of US corporate AI spending but also introduces a new geopolitical dimension to the technology rivalry between the United States and China.

Most popular US AI models are closed, meaning the developer fully controls their use. In contrast, many Chinese AI models use an open-weight approach, allowing companies to download, modify, and customise the models to their specific needs. OpenAI and Anthropic have previously accused several Chinese AI developers, including DeepSeek, MiniMax, and Moonshot AI, of copying their technology.

On the other hand, some companies remain reluctant to use Chinese AI models due to security concerns. Several OpenAI and Anthropic executives have warned of potential risks, and some officials in President Donald Trump’s administration have proposed banning Chinese AI models. Others, however, argue that such efforts are merely aimed at stifling competition.

On Friday, major technology companies including Nvidia, Microsoft, and Palantir signed a joint letter supporting the development of open AI models. They also urged US policymakers to exercise caution before imposing restrictions on such models. Regardless of the outcome of this debate, many investors and industry players believe the era of low-cost AI is here to stay. Several US technology companies have also begun releasing or developing their own open AI models.

Silicon Valley firms were the first to adopt a strategy of selecting AI models based on cost efficiency, much like buying any other commodity. In recent weeks, this approach has started to spread to other industrial sectors. ‘I see no loyalty at all. It feels like an all-out war right now,’ said Marty Kausas, CEO of Pylon.

According to Kausas, AI companies are now offering significant incentives to retain customers. Pylon received unlimited AI usage for several months. Over the course of this year, the company has been given free tokens worth approximately US$1.6 million from one AI provider, US$65,000 from another, and US$10,000 from a third. Microsoft is also reportedly considering adding Chinese AI models, including DeepSeek, to its platform. Meanwhile, several startups say they have held discussions with executives from the financial, healthcare, and insurance sectors about utilising open AI models.

Cursor’s advantage lies in its software being ‘model agnostic’, allowing users to easily switch between models from OpenAI, Anthropic, SpaceX, Google, and Chinese AI developers. Saeks explained that companies can save on AI costs through various strategies, such as restricting access to premium models for new employees or using the most advanced AI only for planning work, then handing the execution over to cheaper models. Cursor recently tested the cost of building a web browser from scratch. Using OpenAI’s GPT-5.5 for the entire process cost just over US$10,000. However, using a combination of Cursor’s Composer model and Anthropic’s Opus 4.8 brought the total cost down to just US$1,339. ‘Before, the best model for a job changed every few months. Now, it feels like it changes several times a week,’ Saeks said.

OpenAI stated it has trained its latest model, GPT-5.6 Sol, to be much more token-efficient, enabling it to complete complex tasks at a lower cost. Anthropic also released a new, cheaper model on Friday. The company said users can now choose the level of AI intelligence or cost they desire within its ecosystem.