Writing in 1998, feminist scholar Alison Adam recalled being asked how a fighter plane designed by a feminist would look any different. Her immediate instinct was to say simply that “feminists do not design fighter planes.” Yet, as Adam acknowledged, where meaningful refusal isn’t possible, such a response retreats from the political project entirely. Nearly three decades later, as Artificial Intelligence (AI) has evolved to harbor more resource-intensive compute and development pipelines, feminist technologists face a similar dilemma. The infrastructures underwriting the contemporary AI pipeline — compute, cloud, capital (and beyond) — harbor their own extractive political economy, yet complete refusal is rarely viable within the contemporary pro-capitalist milieu, particularly for the Global Majority.
Where then do apertures, or what Louise Amoore calls “openings onto the writing of algorithms: bifurcated paths,” for meaningful feminist intervention for the Global Majority present themselves? Is a full stack feminist approach for the South merely a pipe dream, or can it be freed from the abstract as an alternative imaginary for overturning the dominant AI paradigm? Guided by these questions, this piece speaks to both why Feminist AI is a necessity for the Global South and the structural reasons why its dominant neoliberal framings are bound to fail in Majority World contexts. That is, it argues that a Feminist AI stack for the Global South must be holistic, extending further down the value chain into energy grids, compute ownership, and cloud services and further out into governance (intellectual property regimes, etc.), sovereign infrastructure, and international trade law, among others.
Feminist responses to ruin(AI)tion
Feminist approaches to AI and technology more broadly are not of recent vintage. In the Global North, as early as the 1960s, Seymour Papert introduced Logo, a programming language that incorporated Papert’s and fellow developer and sociologist Sherry Turkle’s convictions as advocates of social constructionism — with Logo drawing on feminist thought, Africanism, and “other areas where people fight for the right not only to think what they please, but to think it in their own ways”. Symptomatic of the broader ‘malestream’ that was dominant in AI development at the time, Logo’s promise did not catch on. However, the impulse persisted, carried forward by Global North scholars like Alison Adam, whose seminal critiques exposed the exclusionary, formalist computing culture that governed AI. Contemporary approaches in the West continue to build on this lineage while drawing heavily on Majority World traditions, bringing to the fore frameworks like Catherine D’Ignazio and Lauren Klein’s Data Feminism — which establishes the primacy of challenging intersectional power asymmetries, elevating emotion and embodiment, and visibilizing the invisible labor underwriting data infrastructures. This impulse is similarly embodied in Caroline Sinders’ ongoing Feminist Data Set Project, which interrogates every phase of the AI lifecycle — from data collection and labeling to training and algorithmic deployment — to construct small-scale, community-led models.
Feminist approaches to AI and data in the Global South emerged less as derivatives of conceptual frameworks or theoretical traditions but more so as responses crystallized by colonial conditions and power asymmetries that have plagued the Majority World.
On the other hand, feminist approaches to AI and data in the Global South emerged less as derivatives of conceptual frameworks or theoretical traditions but more so as responses crystallized by colonial conditions and power asymmetries that have plagued the Majority World. In this vein, post-colonial scholar Ann Laura Stoler reframes colonization not as imperial “debris” or static relics of a bygone era, but as a continuing process of “ruination”, a broader political project that “lays waste to certain peoples, relations, and things that accumulate in specific places”. Seen this way, to trace feminist AI in the Global South is to trace a timeline of ruination, where historical extraction re-emerges through AI and digital architectures.
These responses take different forms across practice and scholarship. Te Hiku Media developed an automatic speech recognition tool for the te reo Māori language that actively resists Western corporate data colonialism. Governed strictly by their Kaitiakitanga License, the project ensures community data is not commodified and remains dedicated to cultural revitalization. Similarly, in Africa, the Masakhane initiative embeds core community-centred values like Ubuntu into AI development. In Southeast Asia, community-based platforms like Tamsang Tamsong in Indonesia illustrate how gig workers and platform drivers organize against algorithmic management and platform extraction. Beyond feminist approaches to data infrastructures, a case study in feminist design is presented by Chayn’s Chatbot, which embeds explicit feminist principles into conversational models to provide culturally aware, trauma-informed crisis responses to survivors of gender-based violence. Crucially, Chayn’s approach recognizes that a feminist model must also be amenable to being decommissioned or sunset when it no longer serves the communities it purports to support. While not all of these projects explicitly identify as being feminist, they embody feminist values and adjacent traditions that share feminism’s commitments to relationality, accountability, and solidarity.
This orientation is also mirrored in the scholarship on feminist approaches to AI and technology from the Majority World. Whereas the Harawayian Cyborg, as a boundary-dissolving figure, represented a radical escape from essentialist feminism, the figure has since transformed into an egregore of sorts – circulating as a universal emblem of hybrid subjectivity, untethered from the specific bodies and territories of women in non-Western contexts. Latin American feminist scholars have proposed the complementary figure of the “Cyborg Pachamama,” viewing technology as “inseparable from the bodies and territories they inhabit”. Similarly, organizations like IT for Change have called for a fundamental overhaul of the dominant Degenerative AI Paradigm, moving toward a ReGenerative model centered on four pillars: meaningful and dignified work, diversified economies, pluralistic knowledge societies, and planetary flourishing.
The intention behind drawing attention to these instances across the Global North and South — which barely scratch the surface of the work taking place globally — is not to compile an exhaustive registry of feminist tech movements and scholarship, but to tease out the core variances and commonalities across different approaches. What then emerges across these strands are three foundational imperatives: that non-Western rationalities must be central to AI development and design, that techno-materialities and ‘body-territories’ are inseparable from discourse on technological infrastructures, and above all, that relationality — forged under conditions of historical extraction — insists that technology can never be divorced from the context of the land, labor, and ecosystems that sustain it.
Why Full Stack Feminism Needs to be Rearticulated for AI Pipelines in the Global South
What, then, does it mean in practice for the developer or community builder seeking to operationalize feminist principles today? How can builders opt for slower, frugal, more deliberate, and non-extractive approaches to AI when the dominant political economy actively forecloses them? If existing feminist models cannot simply be retrofitted for the Global South, what replaces them?
Full Stack Feminism, developed through the Full Stack Feminism in Digital Humanities project, applies feminist principles across three layers: data and archives; infrastructure, tools, and code; and access, experience, and integration. However, as researchers on the project acknowledge, this framework is a starting point that can be viewed through various intersectional frames. This part of the article, then, argues that for the AI development pipeline in the Global South, a Full Stack Feminist approach to AI development should address the structural, distributive, and allocative asymmetries encountered in the Global Majority — not only embedding feminist values, but also proactively anticipating and countering the deficiencies of the “extraction stack” that currently controls AI development in the Global South. The necessity of this reorientation is visible most starkly in the accounts shared by developers and practitioners of the various challenges they have come up against while attempting to take frugal and community-led feminist AI projects forward, some of which have been outlined below.
Consequently, even avowedly open, community-oriented tools routinely land on AWS and Google Cloud servers because no alternative offers the required reliability, processing speed, and cost efficiency.
Although frugal AI approaches are oriented towards maximizing efficiency in resource-constrained environments, they can be hampered by “data scarcity, compute limitations, and funding gaps that prevent scaling,” with their success depending on local infrastructure like data centers as well as access to cost-effective Graphics Processing Units (GPUs). Additionally, despite scaffolding AI tools in open-source technologies, the core architectural components of modern AI, including cloud services, datasets, and container software, remain tethered to Western hyperscalers. Market foreclosure operates through high switching costs and token-based pricing regimes that make independent alternatives financially untenable for resource-constrained teams. Consequently, even avowedly open, community-oriented tools routinely land on AWS and Google Cloud servers because no alternative offers the required reliability, processing speed, and cost efficiency. On the flip side, where openness ostensibly exists to further feminist values, it is exploited as a vector by capitalist actors, in what Alex Tarkowski calls the ‘paradox of open’. Owing to these dependencies, tools built for feminist ends and with feminist values at their core may still remain attached to proprietary and extractive corporate regimes that create dependencies at various parts of the pipeline.
Similarly, although not AI technology in and of itself, pioneering initiatives like Janastu’s Community-Owned Wireless Mesh (COWMesh) in India — designed to allow local communities to own their communication infrastructure and access AI tools — face replicability challenges. In practice, maintaining basic network functions like wireless mesh routing proves technically volatile and labor-intensive, requiring ongoing expertise and access to hardware that community-led initiatives struggle to sustain.
How the market is regulated vis-à-vis intellectual property, benefit sharing, and trade agreements plays a decisive role in determining who can build.
In the AI value chain, the dominant neoliberal feminist approach presupposes a builder already within the stack, able to contend with the Extraction Stack’s politics from a position of at least partial access, whereas structural exclusion from the stack itself is not treated as a starting condition. Ultimately, such an approach fails to account for how, in the Global Majority, each layer of the stack presupposes the layer beneath it: access presupposes tools, tools presuppose data, data presupposes resources and power. How the market is regulated vis-à-vis intellectual property, benefit sharing, and trade agreements plays a decisive role in determining who can build. This represents a crucial dimension of feminist AI that merits deep and continued exploration in its own right. We must therefore popularize and problematize the market and its “chokepoints” as key analytical sites that fall squarely within the remit of feminist AI concerns.
Context, Care, Co-implication: A feminist AI Framework for the Global South
With my heart and eyes, I have a completely different view,
The consequence of my skin comes in an entirely different hue.
Don’t you see? Although you represent us,
We think very differently than you.
Because we see the world not through one set of eyes,
But through two.
Rebecca Thomas, Etuaptmumk: Two-Eyed Seeing
Coined by Mi’kmaw Elder Albert Marshall in 2004, Etuaptmumk is the Indigenous principle of Two-Eyed Seeing. It describes the practice of learning to see from one eye with the strengths of Indigenous knowledges and relational ways of being, and from the other eye with the strengths of Western ways of knowing. A feminist approach to AI development demands holding both eyes wide open. To operationalize this, the 3Cs Framework (Context, Care and Co-Implication) — developed by Anita Gurumurthy and Nandini Chami under IT for Change’s FemFirst Initiative — offers the baseline criteria to intervene across the pipeline:
1. Context: Demands that AI development be grounded in the structural and historical conditions of the region in which the system is being developed and deployed. In practice, this includes building operational, alternative, and community-led data architectures or prototypes that challenge traditional corporate machine learning systems and techniques, alongside adopting participatory and co-design methodologies and contextual applications of data feminism principles. Similarly, in governance, this can include constructing governance models, interventions, and regulatory frameworks that actively prioritize digital sovereignty, local development realities, and economic justice over superficial, liberal notions of tech inclusion and access.
2. Care: AI development must embed mutual care and reciprocity in value chains. In practice, this involves visibilizing and fairly compensating the gendered, highly exploited labor force that powers AI — from data annotators and content moderators in the Global South to local data enrichers — and reorganizing funding and operational mechanics so that value creation flows directly back into the communities generating the data. Similarly, policy instruments must be harmonized to curb corporate power and ensure that different stakeholders across the value chain are able to exercise their right to work in non-precarious, dignified, and safely regulated environments.
3. Co-Implication: AI pipelines must recognize and remain attentive to the fact that technology is deeply entangled with material infrastructure and environmental realities. In practice, this involves experimenting with infrastructure choices that account for the carbon, energy, and water footprint of AI prototypes, and actively resisting the environmental degradation wrought by hyperscaler data centers; and investing in regional public compute clusters and frugal edge deployments that reduce structural reliance on monopolistic cloud providers.
In this way, holding both eyes open strips away the myth of AI’s inevitability, revealing a contingent political economy that can and must be contested, dismantled, and rebuilt from the ground up by the Global Majority.
The author expresses deep gratitude to colleagues Sadhana Sanjay (especially for directing her to think about ‘apertures’ of intervention), Abhineet Nayyar (particularly for his sharp insights on market “chokepoints” and the need for a techno-legal governance approach), Nandini Chami, and Anita Gurumurthy for their invaluable conversations and guidance that shaped the article.
