The rapid success of DeepSeek-R1, a high-performing open-source AI model, offers more evidence that open, well-optimised AI systems can rival proprietary models. This breakthrough challenges the long-standing belief that building top-tier AI systems requires ploughing millions of dollars into cutting-edge hardware. Remarkably, DeepSeek-R1 was trained on just 2,788 GPUs, resulting in a 96% reduction in training expenses compared to other high-performance models. While many proprietary systems still rely on heavy computational power, DeepSeek-R1 shows that effective AI can be built with far less infrastructure.
This shift marks the beginning of a new era where AI advancements will come at a lower cost and be more accessible to businesses of all sizes. It’s not just happening in China. See, for example, the work by researchers from Stanford and the University of Washington recently published a paper claiming they used just 16 H100 GPUs to create a low-cost AI model in only 26 minutes that competes with OpenAI’s – all at a cost of less than $50.
Open-source communities are key to advancing AI. Innovation thrives with collaboration, not isolation. No single organisation, no matter how deep their pockets, can reshape the landscape of AI alone. Open-source models democratise access to essential components like datasets, code, and model parameters, fuelling innovation across the AI ecosystem. By opening up AI’s building blocks, open-source systems enable a community of participants—from startups to research institutions—to contribute to and benefit from technological progress.
Open-source AI not only reduces barriers to entry but also speeds up breakthroughs by bringing together diverse ideas, talents, and perspectives. Just as the open-source software movement revolutionised other industries, it is now catalyzing the development of new and dynamic AI models, unlocking cost-effective solutions and enabling more efficient use of resources. Different viewpoints and skills open the door to solutions that may have otherwise been overlooked, leading to innovations that transcend borders and disciplines.
Opening up access to AI
The development journey of AI mirrors the history of computing. Initially, access to computing resources was restricted by high costs, but as technology progressed, more affordable options emerged, paving the way for widespread adoption. Open-source AI is following this trajectory, empowering developers to push the boundaries of what’s possible while maintaining a focus on safety, trust, and ethics. This collaborative, transparent approach propels innovation while making sure AI serves the broader public good.
Many businesses are also looking to Small Language Models (SLMs) to reduce the cost of AI adoption, with 15% market growth expected up to 2030. Open-source SLMs that have been built on carefully curated enterprise data can be safely fine-tuned by companies with their own data, further increasing their return on investment from AI.
But while open-source AI has many advantages, it also introduces new challenges related to data governance and security. Different regions may have varying data protection regulations, making compliance a complex issue for companies operating internationally. Transparency, though a key feature of open-source initiatives, is not always guaranteed. Some model providers may not disclose the sources of their training data, raising concerns about accountability and the risk of bias.
Many organisations are exploring hybrid AI approaches—using both open-source models and proprietary ones to create tailored solutions. This flexibility allows businesses to integrate open-source technologies into their existing systems, ensuring they get the best of both worlds: cost-effective innovation with the reliability and security of proprietary solutions.
AI for all
The future of AI does not belong in the hands of a select few; it belongs to everyone. The rise of global collaborations in the AI field highlights the growing movement toward an open ecosystem. Access to open-source AI empowers not just enterprises but individuals and governments as well, allowing them to harness AI’s potential for a wide range of applications.
Take climate science, for example. Thanks to open-source AI models that improve long-range, hyper-localised weather forecasting, communities lying in the path of extreme weather events can be given more time to prepare or shore up their defences.
AI holds the power to address some of the most critical challenges humanity faces, but this potential can only be fully realised if AI development is democratised. Open-source AI plays a pivotal role in ensuring that all stakeholders—from small businesses to large governments—can tap into the vast opportunities this technology offers.
Dr Juan Bernabe-Moreno is the director of IBM Research Europe for the UK & Ireland
Read more: With CSRD deadlines looming, businesses need to work smarter, not harder
