A severe imbalance in venture capital allocation risks relegating developing nations to passive consumers of Western-centric artificial intelligence architectures.
Artificial intelligence research and deployment currently face a significant structural imbalance, with nearly two-thirds of global venture capital flowing into the sector while less than 1 per cent targets social impact initiatives. This concentration of resources, documented during the 81st session of the United Nations General Assembly in New York, threatens to solidify a technological divide where developing nations function exclusively as consumers of systems designed by and for Western markets.
Data from a white paper titled “From Access to Agency: How philanthropy can support AI development in the Global South,” produced in collaboration with Oliver Wyman, indicates that AI attracted $258.7 billion in venture capital during 2025. This figure represents 61 per cent of total global venture investment, yet the negligible portion directed toward social impact limits the development of localized infrastructure and indigenous technical expertise.
The Gates Foundation’s Goalkeepers 2026 report further illustrates the technical consequences of this funding concentration, particularly regarding training data provenance. Current large language models remain heavily reliant on English-language corpora, with over 90 per cent of training data originating from English sources.
This linguistic bias manifests in significant performance degradation across non-English benchmarks. Leading speech-recognition systems demonstrate error rates below 6 per cent for English, whereas equivalent systems exhibit error rates exceeding 60 per cent when processing Yoruba, highlighting the failure of current architectures to generalize across diverse language datasets.
The lack of investment in local-language data curation creates a feedback loop where models are inherently optimized for high-resource languages. This architectural limitation prevents the effective deployment of natural language processing tools in regions where the primary communication medium is not represented in the training set.
Badr Jafar, Special Envoy of the UAE Minister of Foreign Affairs for Business and Philanthropy, emphasized that the current commercial investment model fails to address the foundational requirements for sustainable AI deployment. He noted that while algorithmic models can traverse borders instantaneously, the development of local trust, technical skills, and institutional capacity requires long-term, non-commercial support.
The white paper advocates for philanthropic intervention to bridge this gap by funding essential but non-commercial components of the AI stack. These include the curation of high-quality local-language datasets, the establishment of independent evaluation frameworks, and the training of personnel capable of maintaining systems beyond the initial pilot phase.
Philanthropy’s job is to fund what others will not: local-language data, independent evaluation, the missing middle of people who keep a system working after the pilot ends, and the institutions that take a decade to build.
This perspective was echoed by Ahmed Talib Al Shamsi, Secretary General of the Emirates Foundation, who argued that meaningful social impact is contingent upon strengthening the capabilities of local institutions. The discourse during the roundtable also explored how the UAE’s existing computational capacity and research partnerships, such as the AIM for Scale initiative with the Gates Foundation, could serve as a model for supporting locally led development.
The technical and ethical implications of this funding gap extend beyond mere accessibility, touching upon the fundamental reliability and safety of AI systems in diverse environments. Without intentional investment in representative data and local evaluation, models deployed in the Global South will continue to suffer from systematic performance deficits and potential misalignment with regional priorities.
The Emirates Foundation has initiated a formal agreement with the Responsible AI Future Foundation to investigate methodologies for the responsible adoption and sustainment of AI solutions across the Middle East and the Global South. Future milestones will focus on the development of shared infrastructure and the creation of localized research ecosystems designed to mitigate the current reliance on non-representative, Western-centric training methodologies.
Stakeholders must now prioritize the creation of independent evaluation bodies to ensure that AI systems are rigorously tested against local linguistic and cultural metrics. This shift toward decentralized, locally-led research will be essential to ensure that the next generation of large language models provides utility beyond the current narrow focus on high-resource, English-speaking populations.
