AI’s open-source moment could be the turning point for its future
DeepSeek-R1 has demonstrated that high-performing AI models can be trained with significantly lower compute costs and supporting infrastructure
THE overnight success of DeepSeek-R1, an open-sourced artificial intelligence (AI) reasoning model demonstrating competitive performance in certain tasks, validates the argument that open, efficient models – when built with optimised architectures and tuned with the right data – can deliver strong competitive performance against proprietary approaches.
This milestone event challenges the conventional narrative that training high-performing AI models requires over US$1 billion and thousands of the latest chips. DeepSeek-R1 was trained using only 2,788 GPUs, achieving a 96 per cent reduction in training costs compared with other leading AI models. While proprietary models still require substantial computational investments, DeepSeek-R1 has demonstrated that high-performing AI models can be trained with significantly lower compute costs and supporting infrastructure.
Open source, open approaches
Open source is the path forward for AI. Innovation doesn’t happen in a vacuum. It’s a team sport – no single organisation can change the game on its own, no matter how resource-rich it is. Open-source AI enables and ignites innovation across the ecosystem by tapping into the collective brainpower of organisations and individuals quickly and easily. It extends access to key AI system components, including datasets, code and model parameters.
Share with us your feedback on BT's products and services
