Who shapes the AI conversation? How global narratives travel and transform across African digital ecosystems

Executive summary

When technology companies announce a new model or a performance breakthrough, the conversations rarely stay contained. It is picked up, reframed, and amplified across financial networks, media, and social platforms, often becoming accepted as fact before it is critically examined.

This report traces the evolution of such stories, collecting six months of global AI discourse on X (formerly, Twitter), from July to December 2025. We contrast global conversations with those in Nigeria, Kenya, South Africa, Ghana, and Rwanda. Our findings support a widely shared assumption — that global AI narratives are largely set by a small group of technology companies and their executives, with discussion focused on model performance, infrastructure scaling, and investment.

Across African contexts, the agenda appears to diverge from this techno-optimism. Conversations center on infrastructure, skills, governance, and access, with AI framed less as a technological race and more as a question of capacity and control. By volume, these discussions are driven primarily by media organizations. Global narratives do circulate, but they are adapted to local constraints and priorities.

The result is not a shared global conversation, but two parallel ones. As these narratives shape policy and investment decisions, the report highlights a central tension: Whether Africa’s AI future will be defined by local needs or by external actors whose technologies and narratives travel faster than regulatory frameworks.

Key findings

Introduction

In May 2025, X users across South Africa discovered that Elon Musk’s Grok chatbot was inserting "white genocide" conspiracy theories into routine queries about baseball scores and enterprise software [1]. The bot claimed it had been "instructed by my creators to accept white genocide as real and racially motivated." X AI attributed the behavior to an "unauthorized modification," but the incident coincided with a Trump administration executive order granting refugee status to white South Africans, a narrative Musk has publicly amplified [2].

Seven months later, when the European Commission fined X €120 million for violating transparency requirements, Musk called for the EU to be "abolished" [3]. U.S. Secretary of State Marco Rubio described the fine as "an attack on all American tech platforms and the American people" [4]. Within hours, this framing spread across social media, recasting a regulatory enforcement action as a broader conflict over free speech and corporate power. Together, these episodes illustrate how statements from technology executives and political allies can quickly redefine how AI governance disputes are understood.

These narratives do not remain confined to Western media. They reach countries developing their own AI strategies, influencing which issues are prioritized and which solutions are seen as viable. A 2025 study, AI Hype Through an African Lens, found that 21% of media articles about AI in African contexts originated from Western outlets [5]. When AI is framed primarily as an economic growth engine, policy agendas emphasize attracting investment, building data centers, and training workers, particularly in regions where AI policy frameworks are still taking shape.

Corporate programs reinforce these narratives. Microsoft’s AI National Skills Initiative trained 350,000 Nigerians by December 2025 and aimed to reach 1 million by 2026. Nigerian media described the effort as "global investment," while government strategies cited it as evidence of "international partnership" [6][7]. At the same time, these initiatives also embedded Microsoft systems into institutional practice, shaping long-term technology dependence. When Sam Altman’s "code red" memo — internal communication highlighting that AI development had reached a critical moment and that competition was intensifying rapidly — leaked in December 2025, African tech newsrooms covered the announcement within days, framing it as part of the OpenAI-Google competition. Nigerian publications Tekedia and Pulse Nigeria explicitly presented the memo as a response to Google’s Gemini 3 launch [10][11]. Kenyan outlet Eastleigh Voice reported that "the outcome rattled OpenAI," citing Altman’s remarks about ChatGPT’s 800 million weekly users and Google’s advantage in integrating Gemini into search [12].

A similar pattern emerged with the arrival of Chinese AI providers at the start of 2025. When DeepSeek and Huawei began offering AI services at prices reported to be up to 94% lower than ChatGPT, African tech coverage largely framed the development as a breakthrough for access [8]. African Business quoted Kennedy Chengeta, a Pretoria-based AI entrepreneur, who argued that "cost has been one of the most significant barriers to AI adoption in Africa" and that cheaper models could enable businesses to "adopt AI without significant investment in infrastructure or talent" [8]. Concerns about data governance received far less attention. Subsequent reporting found that DeepSeek stores user data on servers in China, where it may be accessible under Chinese law. Some African technologists warned that this posed risks, particularly for education and public-sector use [9]. Despite these developments, adoption has continued among startups in Kenya, Nigeria, and South Africa, suggesting that cost and access often outweigh sovereignty considerations in public debate [9].

The case illustrates a central tension in African AI policy: Whether immediate affordability justifies long-term dependence on external systems.

These patterns shape which issues receive attention and which are deprioritized. The consequences are concrete: Huawei’s Safe Cities systems were used by Ugandan authorities to arrest more than 800 opposition supporters in 2020 [29] . Scholars describe this as "digital neo-colonialism" [30], where countries adopt technology designed, trained, and governed elsewhere, gaining access but losing control.

While newsrooms document these developments, social media platforms show how tech-CEOs’ narratives are taken up, repeated, and normalized by journalists, policymakers, technologists, and local communities before they translate into policy and investment choices. These dynamics raise key questions: Which voices dominate African AI discussions? Do local journalists, policymakers, and civil society groups shape the conversation, or do global tech CEOs define it? When CEO narratives arrive, are they challenged or adapted to local contexts, or accepted with their framing intact?

This report examines how CEO-initiated AI discourse spreads, transforms, and encounters resistance across African social media and newsrooms from July to December 2025. Using SimPPL's Arbiter platform, the study traces narratives from executives at OpenAI, Meta, xAI, and Anthropic as they diffuse online in Kenya, Nigeria, South Africa, Rwanda, and Ghana. The study focuses on five core questions:

1. Who shapes the agenda? Are conversations dominated by CEOs, or do local journalists, policymakers, and civil society construct counter-narratives? We examine which voices are amplified and whether African actors reframe or resist narratives from Silicon Valley or Beijing.
2. Which narratives dominate, and how are they framed? Are discussions focused on economic opportunity, job displacement, data sovereignty, or sustainability? Do framings differ across countries?
3. How do CEO narratives travel? When Altman’s memo or Musk’s comments appear, how do African newsrooms interpret them? Do outlets republish with critical analysis or maintain the original framing?
4. What alternatives are proposed? Are policymakers and technologists advocating for indigenous AI, data sovereignty, and localized models, or accepting Western and Chinese platforms as the only options? Are discussions about African language models, labor protections, or sustainability framed as viable or impossible?
5. How do conversations differ across countries? Do Nigeria, Kenya, South Africa, Ghana, and Rwanda share a common AI agenda, or do priorities diverge due to national strategies, infrastructure realities, and geopolitical positioning?

By examining this six-month period, when African AI strategies were being implemented, global companies were expanding, and geopolitical competition was increasing, the report maps how African digital ecosystems amplify, translate, and sometimes contest global narratives. Understanding which voices are amplified, which framings dominate, and where alternatives emerge or are suppressed reveals whether Africa’s AI future will be determined locally or inherited from external actors.

Data collection methodology

This study employs a multi-stage data collection methodology to examine how artificial intelligence-related narratives circulate between global technology actors and African digital ecosystems. The approach combines newsroom analysis, social media account mapping, and lexicon-based discourse tracing to capture institutional, industry, and community-level conversations across five African countries during a six-month period from July 1 to December 31, 2025.

The methodology is designed to address a central challenge in studying digital discourse. Reliance on a fixed set of influential or verified social media accounts risks reproducing existing visibility biases and overlooking emerging or less formal voices. To mitigate this, the study integrates account-based sampling with lexicon-based collection, allowing for the analysis of both established actors and distributed, event driven conversations.

Data collection design overview

The methodology consists of three interconnected stages:

• Newsroom narrative identification: Systematic review of AI and technology coverage by major news outlets in Kenya, Nigeria, Rwanda, Ghana, and South Africa to identify dominant AI discourse themes, organizations, public figures, and initiatives.
• Social media actor mapping: Identification and verification of Twitter accounts representing entities, individuals, and organizations mentioned in newsroom coverage, supplemented by targeted searches for AI communities and civil society actors.
• Lexicon-based discourse expansion: Development of phrase dictionaries based on themes identified in newsroom articles to capture conversations beyond identified accounts.

This approach combines account-based and lexicon-based sampling to capture both institutionalized and distributed AI discourse.

Data sources and sampling framework

Newsroom selection

GNews and NewsData.io APIs were used to discover AI and technology stories and identify news outlets active in the selected countries. The APIs were used for discovery and filtering and were not treated as exhaustive sources.

Selected outlets publish in English and maintain an active website presence, a Twitter presence, or both. The geographic location of each outlet was cross-referenced with information available on its website and Twitter account.

AI-related articles published between July 1 and December 31, 2025, were manually reviewed to identify:

• Organizations mentioned (AI companies, research labs, data science communities, civil society groups, government initiatives)
• Public figures quoted or interviewed (AI researchers, entrepreneurs, policymakers, academics)
• Events and initiatives discussed (AI and digital skilling programs, conferences, AI summits, AI policy launches)
• Topics covered (AI infrastructure, investment, regulation, applications in healthcare, agriculture, education)

This curated newsroom corpus served as the foundation for account mapping and lexicon development.

Twitter account mapping

Twitter was selected as the primary platform due to its widespread use among journalists, policymakers, researchers, and technology organizations, and its suitability for analyzing public discourse at scale in the region. Accounts were initially identified through newsroom coverage, using organizations and individuals mentioned in AI-related articles as a starting point. Each account was independently verified through news articles, organizational websites, and evidence of activity during the study period.

To expand beyond the accounts explicitly referenced in news coverage, we examined retweets, mentions, and interactions within the African AI ecosystem. AI research agents, including ChatGPT and Perplexity, were used to surface additional prominent regional voices, though all suggestions underwent manual verification. This network-based approach allowed us to include influential actors who may not have appeared in newsroom reporting like freeCodeCamp and DAIEvolutionHub but were active participants in regional AI discourse.

A limited set of global AI companies and leading technology figures were also included for comparative purposes, helping contextualize regional conversations against broader global narratives. These accounts were added intentionally for comparison and were not part of the regional discovery process.

The complete list of accounts, along with verification details and collection dates, is provided in Appendix A.

Lexicon development

To capture discourse beyond identified accounts, a lexicon dictionary was developed through an iterative process:

Thematic extraction from newsroom coverage: Articles were reviewed to identify recurring themes, including infrastructure, skilling programs, investment, governance and regulation, applications, AI models, events and initiatives, and national strategies.

Phrase generation and refinement: Thematic keywords were combined with geographic qualifiers to create candidate phrases (e.g., "AI infrastructure in Nigeria," "AI infrastructure in Kenya"). Synonyms and related terms were included to maximize coverage. This process produced an initial set of 187 candidate phrases. These phrases were then iteratively reviewed and refined through exploratory research using ChatGPT and Perplexity deep research agents. Phrases were removed or consolidated if they overlapped substantially with other phrases.

The final lexicon consisted of 151 phrases representing the thematic breadth of AI discourse across the selected countries. Complete lexicon dictionary organized by theme is provided in Appendix B.

Geographic identification of accounts

To identify accounts based in the selected African countries, we extracted and analyzed the location information provided in Twitter account profiles across all stages of data collection. Where users specified a geographic location, this information was used to classify accounts as belonging to Nigeria, Kenya, South Africa, Ghana, or Rwanda. This location-based classification was applied consistently across accounts identified through newsroom mapping, network expansion, and lexicon-based collection, enabling the construction of country-level datasets. Accounts without location information were retained in the broader dataset but could not be reliably assigned to a specific country.

Limitations

Platform coverage

The analysis focuses exclusively on Twitter and English language newsroom coverage. Other platforms, including TikTok, may surface different actors, engagement patterns, or content formats, but were not systematically collected due to application programming interface access limitations and differing platform structures. Application programming interface rate limits, including a cap of 200 posts retrievable per account or phrase per day, may have resulted in missed content during periods of high activity or rapid narrative shifts. Deleted posts and content from suspended accounts are inaccessible, which may lead to partial data loss. Moreover, the dataset does not reflect discussions taking place in offline settings, oral traditions, community radio broadcasts, or conversations occurring within WhatsApp, which is widely used for public discussion and narrative formation in the region.

Account sampling and representation

We identified a set of Twitter accounts through newsroom coverage, network mapping, and manual verification, but this represents only a partial view of the African artificial intelligence ecosystem. Many researchers, organizations, and communities working on AI do not have verified Twitter accounts or were not mentioned in English-language news during the study period. Smaller civil society groups, community initiatives, and regular users are likely underrepresented, although our lexicon-based approach helps capture some of these gaps.

In addition, the identification of region-specific accounts relied in part on self-reported location information in Twitter profiles. As not all users include location details, or may provide incomplete or ambiguous information, it is likely that some accounts from the selected African countries were not captured in country-level analysis. This introduces a potential bias toward accounts with clearly stated geographic identifiers.

Moreover, we observed substantial limitations in using AI research agents (e.g., ChatGPT, Perplexity AI) for account identification. While these tools could suggest candidate handles based on names, they frequently produced:

• Accounts belonging to individuals with similar names but unrelated to African AI ecosystems (e.g., U.S. or U.K.-based researchers with tenuous or historical affiliations)
• Inactive or non-existent accounts, likely generated through name-only inference
• Duplicate or contradictory results, including the same handle being attributed to different individuals

Lexicon constraints

Newsroom selection was limited to English language publications, which means local language media, including newspapers, radio, and online outlets publishing in Swahili, Kinyarwanda, Igbo, Twi, Xhosa, Zulu, and other languages, were not systematically covered. The lexicon used for narrative identification is also English only, which limits visibility into discussions conducted primarily in local languages unless English keywords are used. Lexicon phrases were validated using available newsroom discourse and supplementary research conducted with ChatGPT and Perplexity deep research agents. As a result, terminology not reflected in the newsroom data may not have been captured with the same efficiency. The lexicon-based approach also introduces the risk of false positives, where generic phrases match unrelated content, and false negatives, where locally specific or evolving terms are not included.

Data access and coding considerations

Certain conversations remain inaccessible, including private direct messages, closed WhatsApp or Telegram groups, and other off-platform discussions where narrative contestation may occur. Informal or anonymous accounts that are authentic to local contexts are often difficult to verify. Qualitative coding, including the phrases and lexicon dictionary used to identify artificial intelligence related narratives in Africa, is shaped by the available newsroom discourse and researcher interpretation. No intercoder reliability measures were implemented for the verification of phrases and the lexicon dictionary used to identify artificial intelligence narratives.

Ethical considerations

All data were collected from public Twitter accounts in compliance with platform terms of service and data policies. Public figures, including verified accounts, government officials, corporate executives, and news organizations, are identified by name used by newsrooms or their twitter handles in analysis and reporting. Data were stored on secure institutional servers, and the study adhered to data minimization principles by retaining only necessary fields. Where applicable, the study aligns with relevant African data protection frameworks, including the Kenya Data Protection Act, Nigeria Data Protection Regulation, and South Africa Protection of Personal Information Act. No analysis or reporting directly identifies private individuals or enables the re-identification of anonymized users.

How AI is discussed globally and across African countries

Analysis of AI-related discussion on Twitter from July to December 2025 across global technology leaders and actors in Kenya, Nigeria, Rwanda, South Africa, and Ghana reveals a clear divergence in both priorities and framing.

Among global technology leaders, discussion is dominated by model development and performance. The largest share of engagement each month focuses on capability, benchmarks, and new releases, peaking at 283 million interactions in November. AI is primarily framed as a technical system undergoing rapid improvement.

Over time, this focus shifts toward infrastructure. By December, interactions with infrastructure-related content (157 million interactions) nearly match model-related discussion (136 million) for the first time in the six-month period. Compute capacity, energy demand, and data center expansion move from supporting concerns to central constraints.

Energy emerges as a key point of tension. Engagement with energy-related posts increased by 221 percent in December, from 28 million to 58 million interactions. This shift is driven first by financial accounts framing grid constraints as an investment opportunity, and later by U.S. congressional debate over rising electricity costs linked to data center demand.

Despite this shift, governance remains marginal in global discourse. Ethics, regulation, and safety appear primarily in response to specific events rather than as sustained areas of attention. Overall, global AI discussion remains focused on scaling systems rather than governing their impact.

Across African countries, the structure of the conversation differs significantly. Skills, education, and jobs are the dominant themes across Nigeria, Kenya, South Africa, Ghana, and Rwanda, accounting for the largest share of engagement each month (14 million to 29 million interactions). These discussions focus on who can access and work with AI systems, rather than on improving model performance.

Infrastructure plays a smaller but distinct role. Engagement ranges from 1 million to 4 million interactions per month, and is framed around connectivity, power supply, and national readiness rather than compute scaling.

Policy and governance are more prominent than in global discourse. They account for 27 percent of total engagement in October and 29 percent in November, driven by debates on digital sovereignty, platform accountability, and financial systems. October represents the peak of overall engagement in Africa (77.5 million interactions), not due to model releases but due to governance-related events.

There are also important national differences. Rwanda shows a more institutionally driven discourse, with stronger participation from government and civil society actors, while other countries show more distributed engagement patterns.

Taken together, these patterns reflect fundamentally different starting points.

Global AI discourse begins with model capability and shifts toward the infrastructure required to scale it. Across African countries, the conversation begins with access, skills, and governance, with infrastructure and policy emerging as constraints tied to national capacity.

AI is therefore framed differently across contexts: as a system to be advanced in global discussions, and as a system to be accessed, adapted, and governed in African ones.

Who shapes AI discourse and who actually speaks

Having mapped what is being discussed and when, the next question is who is driving these conversations and whether the actors who show up most are the same as those shaping the agenda.

Across the dataset, 14,949 global accounts and 1,481 accounts across five African countries contributed to AI-related discussion. These accounts span media, government, technology professionals, executives, researchers, civil society, and unaffiliated individuals.

At first glance, the ecosystem looks similar across regions. Individuals make up the largest share of accounts in both datasets. Globally, they account for nearly two-thirds of all participants. In Africa, they remain the largest group, but their share drops significantly, replaced by a stronger presence of institutional actors. Media and government accounts each represent 17 percent of African accounts, more than double their share globally.

However, presence does not equal influence.

When looking at who actually speaks, measured by volume of posts, a different picture emerges. Globally, activity broadly reflects participation. Individuals dominate both account share and post volume. In Africa, this relationship breaks down.

Media accounts represent just 17 percent of accounts but generate 54 percent of all posts. Individual users, despite making up nearly half of all accounts, contribute less than a third of total content. By volume, AI discourse in Africa is not driven by individuals or technical communities. It is driven by media organizations.

This shift shapes not only who speaks, but also what gets discussed.

Across all actor types, the content of AI discussion differs sharply between global and African contexts.

The distribution of actors also shapes the content of discourse. Technical discussions that dominate global AI conversations remain limited in African contexts, even among developers and researchers, who focus instead on skills, training, and access. At the same time, skills emerge as a shared frame across all actor types, while governance discussions are present but lack the participation of civil society actors who engage most directly with accountability issues globally.

These structural differences are reflected in how different actors participate in the discourse.

Media organizations dominate output in Africa, but their role is primarily relay rather than analysis. Their coverage focuses on government announcements, partnerships, and global technology developments, with limited evidence of independent evaluation or technical scrutiny. Despite producing the majority of posts, their highest engagement comes from stories tied to everyday concerns such as taxation, education, or public accountability rather than AI systems themselves.

Government accounts play a distinct role in shaping geopolitical visibility. They show the highest relative share of references to Chinese AI companies, with 26% of company mentions linked to firms such as Huawei and DeepSeek. This reflects the role of infrastructure agreements, particularly Huawei contracts in Kenya, South Africa, and Rwanda, in shaping how Chinese AI appears in public discourse.

At the same time, civil society actors, who globally engage most directly with AI accountability, model evaluation, and data rights, are nearly absent from African AI discussions in this dataset. Where they are present, their focus remains on broader governance issues rather than AI-specific concerns.

Taken together, these patterns show that AI discourse is shaped not only by what is being discussed, but by who has the capacity to speak at scale.

In global contexts, conversations are distributed across individuals, technical experts, and companies. In African contexts, they are more centralized, driven by media and institutional actors, with limited participation from technical communities and civil society. This imbalance shapes not only the volume of discourse but also the questions that are asked, and those that are not.

From structure to narrative

The preceding sections map the architecture of the discourse: What themes dominate, which actors are most visible, and where key voices are missing. But aggregate patterns only show part of the story. They do not show how a narrative begins, how it changes as it moves across actors, or which questions disappear as it gains reach.

The four case studies that follow trace that process more closely. Each examines a different narrative cycle, from origin to amplification to fragmentation. Together, they show how visibility is produced, how meaning shifts between individuals, media, institutions, and policymakers, and why the narratives with the greatest public reach are not always the ones that engage most directly with governance, infrastructure, or accountability.

Skills as survival

A high-reach narrative built on urgency, not instruction

The narrative

Across Nigeria, Kenya, Ghana, Rwanda, and South Africa, the dominant AI-related message is simple: "Learn a tech skill." It appears consistently across the six-month period and across actor types, but carries very little information about what to learn, how to learn it, or whether it leads to employment.

How it starts: Individual amplification through repetition

In Nigeria, the narrative emerges from individual accounts. On July 10, @Theoladeledada posted a single phrase repeated dozens of times: "Learn a tech skill now," generating millions of engagements. The post did not name a tool, course, or employer. The repetition itself created urgency without instruction.

The same phrasing reappears months later across accounts. @Kynsofficial reposted nearly identical language multiple times between September and December, each time generating substantial engagement. The message spreads not by evolving, but by remaining unchanged and emotionally legible.

Where the narrative circulates, and how it expands

As the narrative moves beyond individuals, it is picked up by different actor types and reframed without becoming more specific. Government accounts in Ghana and Rwanda position skills as access, promoting tools such as Gemini through national initiatives. Media and public figures extend the message into mainstream visibility, including posts featuring athletes and entertainers urging young people to acquire digital skills.

The narrative therefore travels across networks, from individuals to institutions to mass audiences, without changing its core structure.

What does not happen, no translation into pathways

Despite its scale, the narrative does not develop into a discussion about outcomes. The actors most capable of translating skills into employment pathways, including developers, researchers, and employers, are largely absent from high-engagement posts. The conversation remains focused on the need to learn rather than the conditions under which learning leads to work.

Where institutional efforts fail to connect

At the same time, structured investments in AI skills do exist. Google’s $2.1 million investment in Nigeria and Microsoft’s program reaching hundreds of thousands of Nigerians are covered in media and community networks [23][24]. However, these posts generate minimal engagement compared to repeated urgency-driven messages, indicating that institutional efforts are not reaching the same audiences as viral narratives.

What is missing

The skills narrative avoids a central question: What skills are actually in demand, and who is being hired. The dataset shows a consistent pattern in which high-engagement posts emphasize urgency, while structured programs remain marginal in visibility. The result is a disconnect between the scale of public attention and the availability of actionable information.

Infrastructure, energy, and accountability

A coordinated investment narrative that fails to reach the public and avoids the question of cost

The narrative

Across Nigeria, Kenya, Ghana, and Rwanda, AI infrastructure is introduced through a highly consistent framing: Large investment figures, claims of regional leadership, and positioning as a "digital hub." The structure of these announcements is nearly identical across countries.

How it starts: Coordinated institutional messaging

On September 9, two posts from TechTrendsKE described Airtel’s data center in Kenya as "East Africa’s largest," posted within hours of each other using near-identical wording. Similar framing appears across countries, including Ghana’s Digital Realty launch, Rwanda’s AI Scaling Hub, and Nigeria’s billion-dollar investments reported through Bloomberg and Energy Connects [21][22]. In Rwanda, institutional accounts including NewTimesRwanda and RDBrwanda posted about the same initiative on the same day using nearly identical language [15].

This reflects coordinated narrative alignment across institutional and media actors rather than organic amplification.

Where the narrative circulates, and where it stops

Despite the scale of these investments, the narrative largely remains confined to the networks that produce it. It circulates among government accounts, media outlets, and institutional actors, but struggles to reach broader public audiences. Even major announcements backed by international funding generate limited engagement, suggesting that infrastructure framed as national achievement does not resonate widely.

What travels instead: Accountability and lived experience

When infrastructure is discussed through different actors, engagement shifts significantly. Posts questioning how funds are used or who benefits from them reach far larger audiences than official announcements. Personal accounts of access to education or connectivity also generate substantial engagement. These narratives do not reject infrastructure, but reframe it through accountability and lived experience, shifting the focus from national positioning to public impact.

Where expertise is missing, and why it matters

The conversation lacks the participation of actors who would connect infrastructure to its underlying systems. Grid engineers, energy economists, and infrastructure planners are largely absent from high-reach discussions. This absence matters because infrastructure is not only a technology issue but also an energy and cost issue.

Globally, this connection is explicit. Reporting links AI data centers to rising electricity prices and grid strain, and political actors frame this as a public cost issue [13][14]. In African contexts, however, this connection is rarely made. South Africa’s energy conversation focuses on whether the existing grid can sustain current demand, while the global debate focuses on whether grids can absorb AI’s additional load. Bloomberg’s finding that electricity bills in data center regions rose sharply [13] is directly relevant to South Africa’s grid fragility, but this connection does not appear in public discourse. The audiences engaging with Eskom-related discussions and those engaging with global AI energy debates remain separate, with no actors bridging the two.

The counter-narrative that does not scale

Alternative framings exist, including Kate Kallot’s "compute desert" argument, which positions Africa’s lack of infrastructure as a sovereignty risk [16][18], and coverage in BusinessDaily [17]. However, these narratives remain confined to specialist and policy networks and do not reach broader audiences.

What is missing

The infrastructure narrative avoids its central question: Who pays the cost of AI infrastructure, and who benefits from it. The dataset shows a pattern in which investment announcements generate limited public traction, accountability narratives gain visibility but do not translate into sustained policy discussion, and energy debates remain disconnected from AI despite their direct relevance. The discourse does not include voices from communities living near these facilities, grid engineers managing increased load, or analysis of data governance obligations tied to infrastructure investment.

Data sovereignty

Three parallel narratives that never converge

The narrative

"Data sovereignty" appears across all five countries, but it does not function as a single idea. Instead, it refers to three distinct concerns: Consumer data loss, state surveillance, and control over national digital infrastructure.

How it starts: Consumer experience and immediate loss

In Nigeria, the highest-reach entry point is everyday experience. A widely shared post about disappearing mobile data frames the issue as theft, generating close to two million engagements. The language is direct and personal, reflecting frustration with platforms rather than abstract governance concerns.

How it evolves, a narrative chain driven by individuals

In Kenya, the narrative develops through a legal case involving Safaricom. A court disclosure about data sharing evolves into a broader debate about privacy and surveillance. Over several weeks, individual accounts expand the story into a systemic critique, transforming a single incident into a wider narrative about power and accountability.

Where institutional processes remain separate

Formal policy discussions, including Kenya’s multi-stakeholder AI policy development process [19][20], run in parallel but remain disconnected from high-engagement discourse. Institutional actors address governance in structured settings, while public narratives evolve independently.

Where the geopolitical framing appears, and fades

A third version of the narrative appears in specialist and institutional spaces, focusing on control over infrastructure and data systems. These posts generate lower engagement and remain within policy networks.

What is missing

The three narratives do not intersect. Consumer concerns, surveillance debates, and policy discussions operate in parallel without convergence. The dataset shows a fragmented discourse in which a shared term does not produce a shared understanding, limiting the ability of public debate to translate into coordinated policy action.

Chinese AI in Africa

High-visibility warnings and low-visibility partnerships

The narrative

Chinese AI appears in African discourse through two dominant frames: As a security concern and as an infrastructure or development partner. These narratives coexist but do not interact.

How it starts — high-reach security framing

High-engagement posts from accounts such as ADFmagazine frame Chinese AI systems as risks to privacy and sovereignty, reaching millions of users. These narratives position Chinese technology within a geopolitical and security context.

Where institutional relationships appear, but do not scale

At the same time, Chinese firms appear across multiple countries through infrastructure projects, education partnerships, and public sector collaborations. These posts are primarily shared by government and media accounts and generate comparatively limited engagement, remaining within institutional networks.

How the narrative fragments across audiences

The two frames operate in parallel without interaction. Security narratives reach large audiences but do not engage with actual infrastructure deployments. Institutional narratives document partnerships but do not address governance risks. As a result, Chinese AI is simultaneously visible and opaque.

Where global comparisons do not translate

Globally, Chinese AI systems are often discussed in terms of cost and performance. This framing has limited presence in African discourse, where discussions focus more on access and partnership than technical comparison.

What is missing

The central governance question is absent: What are the terms under which these systems are deployed, and who controls the data they generate? The dataset shows a disconnect between visibility and accountability, where risk and implementation are discussed separately, but not connected into a coherent public debate.

Geopolitics, infrastructure, and the open question of influence

Across the four case studies, a consistent pattern emerges. The narratives that travel furthest are those grounded in urgency, accountability, and lived experience. The narratives that matter most for long-term governance, including infrastructure ownership, energy costs, and data control, remain fragmented, low visibility, or confined to institutional and specialist networks.

This pattern shapes how artificial intelligence is understood across African contexts. Large language models are rarely debated in terms of benchmarks or technical architecture. Instead, they appear indirectly through questions of cost, accessibility, and deployment. Where models are discussed, Chinese systems are often framed as more affordable alternatives to U.S.-based offerings, particularly for developers and institutions operating under resource constraints [25]. This suggests that model choice is driven less by performance and more by economic feasibility.

At the same time, global actors are positioning themselves differently within this landscape. Chinese firms are increasingly visible through infrastructure partnerships, emphasizing connectivity, capacity expansion, and long-term system integration. This aligns with broader commitments such as China’s $50.7 billion pledge under the Beijing Action Plan 2025–2030 [28], and is reflected in projects such as Huawei’s broadband and backbone infrastructure across countries including South Africa [26].

In contrast, U.S.-based companies are more frequently associated with digital skilling initiatives, cloud platforms, and ecosystem development rather than physical infrastructure [6]. These differing approaches are reflected in public discourse: Infrastructure appears through government and institutional messaging, while skills and tools circulate more widely through media and individual accounts.

African conversations do not resolve this geopolitical competition, nor do they align clearly with a single bloc. Instead, they reflect a pragmatic orientation shaped by immediate constraints. Infrastructure, energy, and skills are treated as practical conditions that determine what can be deployed today, regardless of origin. As seen in the case studies, discussions of infrastructure rarely connect to energy systems, and debates on data sovereignty remain fragmented across consumer, legal, and policy contexts. The absence of actors able to link these layers leaves key governance questions unresolved.

As a result, global rivalries enter African discourse not as ideological debates, but as operational choices about cost, speed, and reliability. Emerging investments from the Middle East further complicate this landscape, introducing additional sources of capital and influence [27].

The outcome is not a unified narrative, but an open question. Africa is not framed as choosing between competing technological spheres. It is navigating among them, adopting systems where they are available and affordable, while governance frameworks, public accountability, and technical capacity develop unevenly.

We must therefore ask not which model leads, but which systems endure.

• Who provides the infrastructure?
• Who controls the data?
• Who absorbs the cost?

In this sense, Africa’s AI future is not being determined at the level of algorithms. It is being shaped at the intersection of geopolitics, infrastructure, and long-term capacity.

About the author

Dhara Mungra is Co-founder and Product Lead at SimPPL. She works at the intersection of AI, information integrity, and civic technology, translating complex business and newsroom needs into scalable technical systems. With four years of experience as a data scientist at a leading audience analytics firm, she leads product development for tools designed with and for journalists, researchers, and civil society organizations, with a focus on responsible AI deployment and public-interest technology.

She is currently an Adjunct Lecturer at the University of Mannheim in Germany, where she teaches AI product development with an emphasis on practical industry applications, stakeholder-centered design, and ethical AI systems. Previously, while studying at New York University, she worked closely with professors and researchers who are now leading research initiatives at Anthropic.

She was also selected as a research group member at the Center for AI and Digital Policy, contributing to the 2026 CAIDP Index through a country-level assessment of Malta’s AI policy, and participated in the Global Changemakers Summit 2025, collaborating with international leaders to prototype scalable solutions for underserved communities.

This report was written in collaboration with DW Akademie.

Acknowledgments

Special thanks to Swapneel Mehta and Steffen Leidel for their guidance and support in shaping the ideation and reviewing the report.

References

1. Musk's AI Grok bot rants about "white genocide" in South Africa in unrelated chats. https://www.theguardian.com/technology/2025/may/14/elon-musk-grok-white-genocide
2. Musk's xAI updates Grok chatbot after "white genocide" comments. https://www.reuters.com/business/musks-xai-updates-grok-chatbot-after-white-genocide-comments-2025-05-17/
3. Elon Musk's X fined €120m by EU in first clash under new digital laws. https://www.theguardian.com/technology/2025/dec/05/elon-musk-x-fined-eu-first-clash-under-new-digital-laws
4. Elon Musk's X fined $140 million by EU for breaching digital regulations. https://fortune.com/2025/12/06/elon-musk-x-fined-140-million-by-eu-for-breaching-digital-regulations/
5. Nkoala, S. (2025). AI Hype Through an African Lens: A Critical Analysis of AI Narratives. https://www.tandfonline.com/doi/full/10.1080/21670811.2025.2528052
6. Microsoft Upskills 4 Million Nigerians, Targets AI Leadership. https://thediscussionnewspaper.co.ke/publications/microsoft-upskills-4-million-nigerians-targets-ai-leadership-f88e0b
7. AI revolution: FG, Microsoft, LBS boost talent pool, train 350,000 Nigerians. https://guardian.ng/technology/ai-revolution-fg-microsoft-lbs-boost-talent-pool-train-350000-nigerians/
8. DeepSeek's cheaper AI claims raise African hopes. https://african.business/2025/01/technology-information/deepseeks-cheaper-ai-claims-raise-african-hopes
9. China's DeepSeek Chatbot Expansion in Africa Raises Alarms. https://adf-magazine.com/2025/11/chinas-deepseek-chatbot-expansion-in-africa-raises-alarms/
10. Google's Gemini 3 Launch Intensifies Competition with OpenAI. https://www.tekedia.com/googles-gemini-3-launch-intensifies-competition-with-openai/
11. New Google and OpenAI AI Models Dropped. Here's What You Need to Know. https://www.pulse.ng/story/google-gemini-3-pro-openai-gpt-5-2-whats-new-2025121317083451019
12. Google raises AI stakes as OpenAI struggles to stay on top. https://eastleighvoice.co.ke/technology/253363/google-raises-ai-stakes-as-openai-struggles-to-stay-on-top
13. AI Data Centers Are Sending Power Bills Soaring. https://www.bloomberg.com/graphics/2025-ai-data-centers-electricity-prices/
14. Bernie Sanders endorses data center moratorium. https://www.eenews.net/articles/feed-center-moratorium/ (Or https://www.eenews.net/articles/bernie-sanders-endorses-data-center-moratorium/)
15. Rwanda Launches $17.5 Million AI Scaling Hub to Lead Africa's Digital Future. https://furtherafrica.com/2025/10/21/rwanda-launches-17-5-million-ai-scaling-hub-to-lead-africas-digital-future/
16. TIME (December 4, 2025), "Kate Kallot on Building an AI Future for the Global South." https://time.com/collections/time100-impact-awards/7335838/kate-kallot-amini-building-ai-infrastructure-africa/
17. BusinessDaily Africa (December 15, 2025), "Data sovereignty and new scramble for Africa." https://www.businessdailyafrica.com/bd/opinion-analysis/columnists/data-sovereignty-and-new-scramble-for-africa-5298572
18. IT-Online (October 2, 2025), "Data sovereignty is Africa's next frontier for digital growth." https://it-online.co.za/2025/10/02/data-sovereignty-is-africas-next-frontier-for-digital-growth/
19. Herman Tambo Law (December 7, 2025), "AI Regulation in Kenya: 7 Essential Insights for Businesses." https://hermantambolaw.com/2025/12/08/ai-regulation-in-kenya-businesses/
20. KICTANet (November 18, 2025), "Kenya Launches Inclusive AI Policy Development Process to Shape a Responsible Digital Future." https://www.kictanet.or.ke/kenya-launches-inclusive-ai-policy-development-process-to-shape-a-responsible-digital-future/
21. EnergyConnects (October 16, 2025), "AI Boom Sparks Billion-Dollar Data Center Race in Nigeria." https://www.energyconnects.com/news/utilities/2025/october/ai-boom-sparks-billion-dollar-data-center-race-in-nigeria/
22. Bloomberg (October 17, 2025), "Nigeria AI Data Center Projects Draw in $1 Billion Investment." https://www.bloomberg.com/news/articles/2025-10-17/nigeria-ai-data-center-projects-draw-in-1-billion-investment
23. Broadcast Media Africa (December 10, 2025), "Nigeria: Google Invests $2.1 Million To Accelerate AI Ambitions." https://news.broadcastmediaafrica.com/2025/12/11/nigeria-google-invests-2-1-million-to-accelerate-ai-ambitions/
24. Microsoft News EMEA (December 16, 2025), "Microsoft empowers 350,000 more Nigerians with AI skills." https://news.microsoft.com/source/emea/2025/12/microsoft-empowers-350000-more-nigerians-with-ai-skills/
25. African Business (November 19, 2025), "Caution urged as Chinese AI takes root in Africa." https://african.business/2025/11/technology-information/caution-urged-as-chinese-ai-takes-root-in-africa
26. ANN (December 11, 2025), "How Huawei is Helping Build South Africa's New Digital Highway." https://asianews.network/how-huawei-is-helping-build-south-africas-new-digital-highway/
27. The National (November 24, 2025), "UAE investment set to boost Africa's AI infrastructure as it reaches 'inflection point'." https://www.thenationalnews.com/future/2025/11/24/uae-investment-set-to-boost-africas-ai-infrastructure-as-it-reaches-inflection-point/
28. Africa International Advisor (2025), "The Year Ahead for China-Africa Investments." https://www.africaia.com/insights/china-africa-investments-in-2025-key-trends-and-how-it-can-improve-africas-competitiveness
29. BBC (2019), "Uganda and Zambia rejects Huawei spying allegations." https://www.bbc.com/news/world-africa-49368276
30. IJRISS (2025), "The Coloniality of Chinese Surveillance Technologies in Africa." https://rsisinternational.org/journals/ijriss/articles/the-coloniality-of-chinese-surveillance-technologies-in-africa/

Appendix

Final account categories and distribution

APPENDIX A: TWITTER ACCOUNTS BY CATEGORY

Global AI companies and executives (22 Accounts)

Africa-Wide organizations (13 Accounts)

Nigeria (26 Accounts)

Uganda (8 Accounts)

Kenya (21 Accounts)

Rwanda (6 Accounts)

Ghana (7 Accounts)

South Africa (18 Accounts)

APPENDIX C: LEXICON DICTIONARY (151 PHRASES)