Three weeks before the African Union signed its landmark AI partnership with Google, the President of the Pan-African Parliament stood at Tangaza University in Nairobi and issued a warning. “If we do not control the data that goes into Artificial Intelligence, we will not control the AI that shapes our future,” Chief Fortune Charumbira told an audience of policymakers, researchers, and university leaders. The AU’s answer, apparently, was to hand Google the keys.
On February 17, 2026, the African Union Commission signed a Memorandum of Understanding with Google at its headquarters in Addis Ababa. The MOU is framed around “advancing Africa’s Sovereign AI and Digital Capacity.” It delivers AI readiness training for African public officials — in partnership with Apolitical — free access to Gemini Pro and NotebookLM across government use cases, support for local languages including Amharic, and a commitment to train three million students and teachers in AI skills by 2030. Projections cited during the signing suggest AI could contribute up to USD 1.5 trillion to Africa’s economy by the end of the decade.
The MOU has been almost universally reported as a win. The actual governance question — what kind of sovereignty does an AI partnership structured entirely around a foreign hyperscale cloud provider actually create? — has received almost none of the scrutiny it warrants.
What the MOU Delivers
The AU-Google partnership operates under a framework titled “AI for Government: Strengthening Digital Public Infrastructure, Education, and Climate Resilience.” Its practical commitments fall into four buckets: government AI readiness, talent development, infrastructure and cloud access, and policy and governance support. The deal was signed by Commissioner Lerato D. Mataboge, who heads the AU’s Infrastructure and Energy portfolio, and Charles Njenga Murito, Google’s Regional Director for Sub-Saharan Africa for Government Affairs and Public Policy.
The infrastructure and cloud access component is the most consequential and the least scrutinised. Under this arrangement, African governments and public institutions will be encouraged to build AI workloads on Google’s cloud infrastructure. Gemini Pro — Google’s flagship large language model — will be made available free or at subsidised rates for education and government applications. NotebookLM, Google’s AI document analysis tool, will be deployed for public sector use cases. The 22 countries currently implementing the AU Data Policy Framework will receive technical support aligned with Google’s platforms.
What the MOU does not contain is equally important: there is no explicit data governance rider specifying how data generated by African public officials using Google’s tools will be stored, processed, owned, or accessed. There is no data localisation requirement. There is no reference to the AU Data Policy Framework’s Cross-Border Data Flow provisions — provisions that were, as of December 2025, still being validated in a continental workshop in Addis Ababa.
The AU signed a major AI infrastructure partnership with Google before its own cross-border data governance framework was finalised.
The Infrastructure Dependency Problem
The language of “sovereign AI” — used by the AU commission itself to describe the deal — requires some unpacking. Sovereignty, in the context of data and AI systems, typically means the ability to set and enforce the terms under which data is collected, processed, and used. It implies that the institutions claiming sovereignty have access to the compute, storage, and model infrastructure necessary to act on those terms independently.
The AU-Google MOU delivers none of these things in structural terms. It delivers access — which is a different concept entirely.
Africa currently hosts less than one per cent of the world’s data centres. Approximately three per cent of global AI talent is based on the continent. The largest AI workloads run by African governments and institutions are almost entirely dependent on hyperscale cloud infrastructure based in North America and Europe. When those governments deploy Gemini Pro for administrative AI applications, the inference computation runs on Google’s servers. The model was trained on data Google assembled. The terms under which the model operates — what it can and cannot do, how outputs are stored, what usage data Google retains — are Google’s terms, not the AU’s.
Training three million students to use AI tools does not change this. Providing free access to Gemini Pro does not change this. These are access and literacy interventions, which have genuine value but are categorically different from building the infrastructure layer that determines who controls AI at the foundation.
The Pan-African Review put the structural concern plainly in its response to the MOU signing: “When data is hosted, processed or trained by foreign entities, countries risk losing control over how their knowledge systems are shaped, refined and monetised. Public data becomes a raw material extracted locally, refined elsewhere and commercialised globally.”
The WAXAL Parallel
The AU-Google MOU sits alongside another significant data relationship between Google and African institutions: the WAXAL speech dataset, comprising over 11,000 hours of voice data across 21 African languages, collected in partnership with Makerere University, KNUST, and other African academic institutions.
WAXAL is a landmark dataset — the largest open African-language speech corpus ever assembled. But the terms of its production raise the same questions the AU MOU raises. African partner institutions contributed the expertise, the recruitment, the community relationships, and the data. Google provided the infrastructure, the model pipeline, and the publication. Who benefits most when African speech patterns are embedded in Google’s next-generation multilingual models, and those models are deployed at scale across African markets?
The AU MOU and WAXAL are not the same project, but they are part of the same structural pattern: a major American technology company becoming the infrastructure layer through which Africa’s AI capacity is built, at a moment when the continent’s own data governance frameworks are still incomplete. The pattern is not evidence of bad faith. It is evidence of a governance gap — and a gap that the AU’s own policy agenda has not yet closed.
Who Signed, and What That Signals
One detail of the MOU that merits attention is who signed it. Commissioner Mataboge oversees the AU’s Infrastructure and Energy portfolio — not the Digital Economy and Trade portfolio, and not the Political Affairs, Peace and Security desk that might be expected to lead on data sovereignty. The AU’s internal framing of AI appears to categorise it primarily as an infrastructure question rather than a digital rights or data governance question.
This is not a technicality. It signals what the AU prioritises in its approach to AI: connectivity, compute access, and infrastructure rollout — the kinds of outcomes that a partnership with a cloud provider can plausibly advance. The questions about data ownership, model accountability, and the terms of AI-driven public administration are left to the AU Data Policy Framework process — which is moving through validation workshops while the infrastructure partnerships are already being signed.
If the AU’s AI strategy is going to be governed by the Infrastructure and Energy portfolio rather than a dedicated digital governance body, the continent will end up with AI infrastructure that precedes the governance frameworks designed to regulate it. This is not a novel problem — it is how most jurisdictions have managed technology, including many in Europe and North America. But it is a problem that the AU’s own Continental AI Strategy was designed to avoid.
The Continental AI Strategy’s Own Standard
Phase I of the AU Continental AI Strategy, covering 2025 to 2026, is explicitly focused on creating governance frameworks, mobilising resources, and building national AI strategies. The strategy’s approach to external partnerships emphasises “responsible AI frameworks” and the importance of ensuring that African institutions shape the terms of AI development rather than inheriting terms designed elsewhere.
The Google MOU is not inconsistent with these ambitions — at the level of stated intent. At the level of structural outcome, however, the deal advances Google’s position as the default AI infrastructure provider for African governments without first establishing the data governance guardrails that the AU’s own strategy says should precede or accompany such partnerships.
The AU Data Policy Framework’s Cross-Border Data Flow provisions, finalized in validation workshops in December 2025, are the instrument that should govern how African public data is handled when processed by foreign cloud infrastructure. They have not been formally incorporated into the MOU. Whether they will be applied retroactively to government use of Google’s tools — and who will enforce that application — is unclear.
What Businesses Need to Know
For technology companies, fintechs, and data-intensive businesses operating in Africa, the AU-Google MOU is a signal about the direction of public sector AI adoption — and it carries regulatory implications that extend beyond government use cases.
First, governments that adopt Google’s AI infrastructure for public administration will increasingly shape their procurement expectations, compliance standards, and interoperability requirements around that infrastructure. Businesses that interface with government systems — whether in financial services, health technology, or digital identity — will feel the downstream effects of public sector AI platform choices.
Second, the data governance gap in the MOU creates regulatory uncertainty. The AU Data Policy Framework’s Cross-Border Data Flow rules are not yet finalised. National data protection frameworks — Nigeria’s NDPA, South Africa’s POPIA, Kenya’s Data Protection Act — apply to personal data processed by foreign entities in some circumstances. Businesses operating in jurisdictions where government AI workloads run on Google Cloud should monitor how their data intersects with those systems and what the applicable rules are.
Third, the MOU is likely to accelerate interest in AI governance legislation at the national level, particularly in countries that are already moving toward AI regulation — Nigeria, Rwanda, Kenya. As governments begin deploying AI tools in administrative functions, the pressure to define accountability frameworks for those deployments will intensify. The AU partnership with Google may, paradoxically, accelerate the national regulatory processes that will eventually impose obligations on the platforms it endorses.
The Question That Remains Open
The AU-Google partnership will deliver real benefits. AI literacy programmes for public officials are needed. Infrastructure access at subsidised rates matters for a continent where compute costs remain prohibitive. Language support for Amharic and other African languages in foundation models is genuinely valuable.
None of this makes the governance question go away. Chief Charumbira’s words at Nairobi in January 2026 were not a critique of AI capacity building. They were a demand that capacity building happen on terms that preserve African agency over the data that AI systems require. The Nairobi Declaration on Sensitive Data Sovereignty, signed at the same conference, committed academic and political leaders to ensuring that African data is stored, managed, and analysed within the continent.
The AU signed an MOU with Google twenty-one days later with no visible reference to the Nairobi Declaration and no data localisation provisions.
Whether the AU-Google partnership can be structured — through subsequent agreements, data governance addendums, or national-level regulatory requirements — to actually deliver on the sovereignty it promises is the central regulatory question of Africa’s AI moment. It is a question the MOU as signed does not answer.
The AU Continental AI Strategy Phase I runs through 2026. The AU Data Policy Framework Cross-Border Data Flow provisions are expected to be finalised before mid-2026. BETAR.africa will track both.
UphivaBETAR
