US weighs new AI release framework as cheaper Chinese models reshape the global race – Firstpost
The United States is reassessing how it governs advanced open-source artificial intelligence as Chinese AI development accelerates, prompting fresh discussions between the Trump administration and leading technology companies over how America’s most capable models should be released.
According to a report by the Washington Post, officials and AI developers are working on a capability-based framework that would determine the release of US open-source models by benchmarking them against the abilities of China’s leading open-source systems. The discussions
reflect a growing recognition in the US that rapid advances in Chinese AI are changing both the competitive landscape and the national security debate surrounding frontier models.
Industry participants involved in the conversations reportedly believe that open-source AI systems comparable to
Anthropic’s Mythos-class capabilities could become freely downloadable from Chinese developers within the next six to 12 months. If that happens, it could undermine attempts to tightly control the distribution of advanced AI in the United States, as similarly capable models would already be widely accessible online.
Corporate AI spending shifts towards Chinese models
Adding to the pressure on
US policy debate is a growing shift in corporate AI spending. According to the Financial Times, a rising number of global businesses are opting for Chinese AI models instead of their US counterparts, driven largely by lower operating costs and greater flexibility. Companies including DoorDash, Airbnb and Siemens are reportedly adopting Chinese “open-weight” models, which can be customised more easily for enterprise applications while costing significantly less to run.
Usage trends appear to reflect that shift. Data from AI platform OpenRouter, cited by the Financial Times, indicates that leading Chinese models from DeepSeek and Z.ai are now seeing higher usage than comparable systems from OpenAI and Anthropic. Eugene Cheah, chief executive of AI platform Featherless AI, told the newspaper: “Enterprises are starting to realize, ‘Hey, we don’t need the best model, we can use the faster, cheaper models.’”
The cost advantage has become increasingly difficult for businesses to ignore as enterprise AI spending climbs. The Financial Times reported that one organisation spent around $500 million on Anthropic’s Claude in a single month, an exceptional case, but one that illustrates the scale of AI expenditure.
Separately, research from Ramp’s AI Index suggests that companies with the highest levels of AI adoption are spending roughly $7,500 per employee each month on AI tools. Against that backdrop, lower-cost Chinese alternatives are becoming an increasingly attractive option, particularly as recent releases such as Z.ai’s GLM-5.2 have drawn attention in Silicon Valley for offering performance approaching leading US models at a fraction of the price.
The debate comes as the Trump administration pushes ahead with broader efforts to establish guardrails for frontier AI. Earlier this month, the Financial Times reported that
US officials and major AI developers were close to agreeing on a voluntary framework that would outline expectations for testing, deployment and access to advanced AI systems before they are publicly released. While not legally binding, the proposed framework is expected to provide common industry standards for companies building frontier models.
Energy breakthroughs could ease AI’s growing power demands
Alongside security and commercial competition, another issue attracting increasing attention is AI’s enormous appetite for electricity. Data centres powering advanced AI models are expected to consume significantly more energy over the coming years, driving concerns over infrastructure costs and power availability.
However, those projections may prove overly pessimistic. According to the Washington Post, researchers and technology companies are making progress on new hardware approaches that could dramatically improve energy efficiency. Instead of relying solely on conventional electrical connections, developers are experimenting with specialised conductors and optical technologies that transmit information using light waves, reducing the amount of power required to move data across AI systems.
If those technologies mature, they could significantly lower the long-term energy demands of AI infrastructure while helping sustain the industry’s rapid expansion.
Taken together, the emergence of powerful and inexpensive Chinese open-source AI, evolving US oversight and advances in computing hardware point to a new phase in the global AI race. For Washington, the challenge is no longer simply building the world’s most capable models, but deciding how to govern them in an increasingly competitive landscape where advanced AI is becoming cheaper, more widely available and harder to contain.
