Chinese AI start-ups Moonshot AI, MiniMax, and DeepSeek are developing open-weight models to compete with U.S. services such as OpenAI’s ChatGPT and Anthropic’s Claude [1, 2].

This shift toward open-weight, locally run models could disrupt the current AI market by lowering operational costs and improving data privacy. By allowing models to run on local hardware rather than premium cloud services, these companies aim to challenge the dominance of U.S. frontier labs [2, 3].

DeepSeek, based in Hangzhou, has emerged as a significant player in this effort [3, 4]. The company recently targeted a funding round of up to $4 billion [4]. This influx of capital contributed to a valuation that climbed fivefold to $50 billion [4].

These labs are prioritizing efficiency to reduce their reliance on expensive Nvidia-based data-center hardware [4]. The goal is to provide a viable alternative for users who find current cloud-based AI too costly or restrictive regarding privacy [2, 4].

Global adoption of these technologies is accelerating, with approximately 16.7% of people worldwide now using generative AI tools [5]. Chinese researchers have set an ambitious timeline, aiming to be competitive with U.S. frontier AI labs by 2025 [1].

While the ambition is high, the transition to open-weight models requires significant infrastructure. The labs continue to navigate the balance between high hardware intensity and the desire for localized, efficient deployment [4].

“Chinese labs aim to rival US frontier AI labs by 2025”

The move toward open-weight models represents a strategic attempt by China to break the 'compute moat' held by U.S. companies. By focusing on local deployment and efficiency, these firms seek to bypass the high costs of cloud subscriptions and the logistical hurdles of U.S. chip sanctions, potentially democratizing high-performance AI across domestic Chinese industries.