Microsoft has committed an additional USD 329 million — roughly R6 billion — to expand its South African cloud and data centre footprint, on top of an earlier USD 1.2 billion investment. The announcement, made by Microsoft president Brad Smith, has been received almost uniformly as good news. Bigger data centres. AI skills programmes. A partnership with the SABC to reach 1.9 million users. Confidence in the South African digital economy.
We should be more careful before we celebrate.
This is not an argument that the investment is bad, or that Microsoft is acting in bad faith. It is an argument that the South African conversation about hyperscaler investment has become structurally uncritical — and that the people most affected by it, including the small and medium businesses that make up the bulk of this economy, are being asked to applaud announcements they have not been invited to interrogate.
Here are the questions worth asking.
Who actually gets selected for "AI for Africa" programmes?
Microsoft's AI Skills initiative, the SABC partnership, and the broader category of hyperscaler-funded training programmes share a common pattern. They report on aggregate reach — millions of users, thousands of certifications — but rarely on demographic composition, geographic spread, or post-programme labour market outcomes.
A reasonable observer should ask:
- Who is actually completing these programmes? The available evidence from comparable initiatives in other markets suggests that participants are overwhelmingly urban, English-fluent, already digitally literate, and concentrated in age brackets and provinces where employment opportunities already exist. The programmes tend to deepen advantages held by people who were going to find work anyway.
- What counts as a "completion"? Many hyperscaler skills programmes count anyone who finishes a short online module as a graduate. The gap between "watched a video on prompt engineering" and "is employable as a cloud engineer" is not small, and aggregate reach numbers paper over it.
- What is the 12-month employment outcome? This is the only metric that matters, and it is the one almost never reported. Without it, we are measuring the effort of the funder rather than the change in the participant's life.
- Who is doing the selecting? When a hyperscaler partners with a state broadcaster, an NGO, or a university to deliver training, the selection criteria are often set by the funder and administered by local partners under contractual pressure to deliver enrolment numbers. The people most likely to benefit are the people the system already finds easiest to reach.
None of this is a reason to oppose these programmes. It is a reason to stop reporting them as transformation wins until the evidence supports the claim.
What actually shifts for Africa, and what does not?
The dominant framing of hyperscaler investment is that it brings infrastructure, jobs, and capability to the continent. The honest read is more complicated.
What shifts. Latency improves for South African users of Microsoft services. Some data stays in-country, easing certain compliance pressures around POPIA and sector-specific data residency rules. A small number of high-skilled construction, facilities, and engineering jobs are created. A larger number of contractor and services roles flow into the local economy during build phases.
What does not shift. The capacity being built is owned by Microsoft, priced in US dollars, and rented back to South African firms on terms set in Redmond. The intellectual property in the AI models running on that capacity remains foreign. The training data, the model weights, and the safety policies that determine what those models will and will not do are decided outside the continent. South African firms become more efficient consumers of foreign AI, not producers of African AI.
This is not a Microsoft-specific critique. It applies equally to AWS, Google Cloud, Oracle, Anthropic, and OpenAI. The point is structural. When the headline reads "hyperscaler invests in Africa," what is happening is that a foreign firm is building distribution infrastructure for its own products in a new market. That is a legitimate commercial activity. It is not, by itself, development.
Are we just building small companies dependent on the giants?
This is the question that should be keeping South African transformation strategists awake at night.
The story being sold to South African SMEs is that the cloud levels the playing field. A two-person startup in Soweto can now access the same compute, the same AI models, and the same global market as a Silicon Valley firm. There is some truth to this. It is also a partial picture.
What the story leaves out is that every layer of the modern digital business — compute, storage, AI inference, payments, identity, communications, analytics — is now provided by a small number of foreign firms charging in dollars and able to change pricing, terms, and access at any time. The South African SME that builds its entire operation on Microsoft Azure, OpenAI's API, and Stripe is not independent. It is a tenant. Its costs are exposed to the rand-dollar exchange rate, its data is held under foreign jurisdiction, and its access to its own customers can be revoked by a counterparty it has no leverage over.
The B-BBEE codes recognise ownership, management control, skills development, enterprise and supplier development, and socio-economic development. They do not yet recognise digital sovereignty as a transformation dimension. They should.
A black-owned SME that is 100% beneficially owned by previously disadvantaged South Africans but operationally dependent on five US firms for its core infrastructure is not as transformed as its scorecard suggests.
The infrastructure question
Brad Smith was unusually candid in the announcement. The new capital, he said, would cover land for future data centres, power and water readiness, and capacity expansion at existing sites. Read that again. Microsoft is paying for power and water resilience because the South African state has not delivered it.
The polite reading is that this is welcome private-sector investment in critical infrastructure. The less polite reading is that we are watching the partial privatisation of national infrastructure provision, undertaken not as policy but as a practical workaround. The hyperscalers will have functioning grids and water supply at their data centres. The South African SME, hospital, school, and township will continue to negotiate with Eskom and a municipal water department.
This matters for transformation strategy in a way that has not been widely discussed. If the future of the South African digital economy runs on infrastructure that the hyperscalers built for themselves, then the benefits of that infrastructure accrue to whoever can afford to rent into it. Companies with capital and procurement sophistication will buy reliable cloud capacity. Companies without will continue to live inside the broader infrastructure failure. The digital divide widens at the infrastructure layer, not just the skills layer.
How does a small business actually reduce its dependency?
The honest answer is: not entirely, and not quickly. Genuine independence from hyperscalers is technically demanding, expensive, and rarely the right commercial choice for a small business. But strategic reduction of dependency is possible, and over a three-to-five-year horizon it should be a deliberate part of how serious South African SMEs build their operations.
Five practical moves:
1. Architect for portability from day one. Choose tools, frameworks, and data formats that can move between providers. Use open standards (PostgreSQL over proprietary databases, S3-compatible storage interfaces, OpenAI-compatible API patterns) so that if pricing or terms change, the cost of switching is measured in weeks not years. The single biggest mistake South African SMEs make is locking themselves into a single vendor's proprietary stack because the integrations are convenient. Convenience today is leverage lost tomorrow.
2. Hold your own data. Even if you run your application on a hyperscaler, your customer data, your business records, and your core intellectual property should be exportable, backed up outside that provider, and held in formats you can read without the provider's tools. If you cannot leave with your data, you do not own your business.
3. Use open models where they are good enough. The open-weight model ecosystem (Llama, Mistral, Qwen, DeepSeek, and the growing set of smaller specialised models) has reached the point where a meaningful share of business AI workloads can run on infrastructure you control or rent from smaller, more substitutable providers. For many SME use cases — document processing, customer service triage, internal search, content generation — open models running on rented GPU capacity are now adequate, cheaper at scale, and dramatically less exposed to single-provider risk. The right question is not "open or proprietary" but "which workloads can sensibly move, and on what timeline."
4. Diversify across providers deliberately. If you are going to depend on hyperscalers, depend on several. Run your production workload on one provider, your backups and disaster recovery on another, and your AI inference somewhere a third party operates. The cost premium is real but modest, and the strategic position it creates — credible ability to leave any single provider — is what gives you actual negotiating power.
5. Invest in the human capability to make these choices. The deepest form of dependency is not technical. It is the absence of the internal expertise to evaluate alternatives. SMEs that have no one inside the business who understands how AI infrastructure actually works will default to whatever their account manager recommends. The cheapest sovereignty investment a South African SME can make is one strong technical hire or trusted advisor who is not paid by the hyperscalers.
What we think Okiru's clients should do
If you are a transformation lead, an executive accountable for B-BBEE outcomes, or a director of a South African business of any size, the announcement of Microsoft's investment should prompt three conversations inside your organisation in the next quarter.
One. Audit your operational dependencies. List every foreign technology provider your business depends on, what would happen if their pricing doubled, and what your exit plan looks like for each. If you do not have an exit plan, you do not have a strategy. You have a habit.
Two. Rebuild your view of skills development. The right question is not "how do we get more of our staff into Microsoft training programmes?" The right question is "what capability do we need internally to make independent technology choices over the next five years?" Some of that capability will come from hyperscaler programmes. Some of it will not.
Three. Take digital sovereignty seriously as a transformation question. The next iteration of the B-BBEE codes will eventually grapple with this. Boards that are thinking about it now will be ahead of the regulation. Boards that are not will be caught flat-footed.
The bottom line
Microsoft's USD 329 million investment in South Africa is real. The data centres will be built. The skills programmes will run. Some good will come of all of it. But the framing of this announcement as transformation, as African development, or as a step toward digital sovereignty is not supported by what is actually happening. What is happening is that a US firm is building distribution capacity for its products in a growing market, and a generation of South African businesses is being asked to mistake that for partnership.
There is a serious version of African AI, and there is a serious version of South African digital sovereignty. Neither will be delivered by hyperscalers, however welcome their capital is. They will be delivered, if at all, by South African firms and institutions that retain the right to ask hard questions of the giants — and by transformation advisory work that refuses to confuse activity for outcome.
That is the work we think is worth doing.
Okiru Consulting helps South African organisations turn transformation strategies into measurable outcomes across B-BBEE, ESG, Skills Development, Employment Equity, and Enterprise Development. If you want to discuss what digital dependency means for your transformation strategy, contact us at hello@okiru.co.za.