innovation through the development and sharing of open AI models that can be replicated, modified and shared within the constraints of secure environments. Yet the integration of AI with DPI can also bring new challenges, such as the risk of large-scale data breaches. The United Nations DPI Safeguard Framework outlines an approach for mitigating risks at both individual and societal levels. Regionally, India has been a pioneer in DPI. The widely successful deployment of the Unified Payments Interface (UPI) for small payments, used for 11.7 billion transactions in 2023, demonstrates how DPI can provide inclusive access to high- speed networks for enabling innovative and impactful solutions at a low cost.7 Access to scalable and affordable compute Scalable and affordable compute capacity is also essential for supporting large-scale AI applications. The global AI infrastructure market was valued at $35.42 billion in 2023 and is projected to grow at a compound annual growth rate (CAGR) of 30.4%, reaching $223.45 billion by 2030.8 Governments are significant investors. The US Federal Government, for example, spent $3.3 billion on AI in fiscal year 2023, more than double the $1.38 billion spent in 2018.9 A recent report by the Tony Blair Institute for Global Change notes that care should be taken as broad investment in high technology will not translate to compute capability, meaning a specific AI compute strategy is needed. Tech companies and private investors (including private equity, asset finance funds and infrastructure funds) can play a significant role in developing large-scale data centres. One alternative to training large language models (LLMs) in large data centres is to use small, domain-specific models that provide narrow-focused expert intelligence for a fraction of the cost of training a large language model. This can be an effective strategy for smaller countries and enterprises.Country collaborations can also help provide access to compute. In one recent successful example, Rwanda and Qatar signed a memorandum of understanding (MoU) to enhance collaboration in R&D in AI, building digital public infrastructure for innovation and transformation.10 The International Computation and AI Network (ICAIN) is a Swiss initiative with a global focus on broadening access to AI resources for sustainable development research. It connects AI capabilities, such as computing power, data and expertise with research projects aligned with the UN Sustainable Development Goals. ICAIN uses two supercomputers and draws on the expertise of experienced policy-makers and AI research experts in Europe and Africa to develop AI models that benefit societal needs. Access to AI-ready devices The use of smartphones and computers has increased significantly in the past decade, with low- cost devices (such as the affordable Reliance Jio smartphone) driving adoption within the Global South. Within the context of an inclusive national AI ecosystem, low-cost devices can be used to run AI-enabled applications that have been optimized to work with limited compute capability, weak internet connectivity and low levels of battery storage. An alternate way to increase access to AI-ready devices is to subsidize the distribution of devices. Singapore’s Digital for Life initiative under the Smart Nation Singapore programme, in partnership with the Edison Alliance, provides subsided devices to low- income families through PPPs with telecom and tech providers.11 Additionally, there are numerous low- cost devices that can be used to facilitate inclusive AI access, such as the previously mentioned Jio smartphone and Computer Aid. The latter, in partnership with large technology companies, provides access to computers and mobile devices. 10 Blueprint for Intelligent Economies

A Blueprint for Intelligent Economies 2024