The race to dominate artificial intelligence is colliding with a difficult reality: the more AI companies expand across the world, the harder it becomes to control how their technology is deployed, governed and adapted to local needs.
The biggest artificial intelligence companies want their models to become indispensable. They want businesses to build on them, governments to integrate them into public services, developers to use them as the foundation of new products and consumers to rely on them for everyday decisions. They also want to retain control over the technology, the infrastructure, the distribution channels and, ultimately, the commercial value created by that dependence.
But these ambitions may be incompatible.
An AI model can achieve global reach without its creator maintaining complete control over how it is used. Conversely, a company that tightly controls access to its models may find that governments, businesses and developers increasingly seek alternatives that offer greater autonomy.
This tension is becoming one of the defining struggles of the AI industry. The question is no longer simply which company builds the most capable model. It is who gets to decide how AI is deployed, whose rules govern it, where its data is processed and who captures the economic value it generates.
The ambition to become the world's AI infrastructure
For companies such as OpenAI, Google, Anthropic and Microsoft, the commercial opportunity extends far beyond selling access to a chatbot.
AI is increasingly being integrated into office software, customer service, software development, financial analysis, healthcare administration, education and government operations. As models become embedded in these systems, switching providers can become expensive and technically difficult.
A company that controls the model powering an organisation's daily operations may influence more than the answers its employees receive. It may also shape the tools available to them, the information they can access, the workflows they automate and the costs their organisation must pay.
This creates powerful incentives for AI companies to establish themselves as foundational infrastructure.
Cloud computing companies pursued a similar strategy by offering the computing power, databases and services on which businesses built their digital operations. AI providers are attempting to occupy a comparable position in the emerging AI economy.
Yet AI introduces an additional complication: governments and customers are increasingly concerned about what happens when a critical capability depends on a provider they cannot control.
A business may accept dependence on an overseas AI provider while the service is affordable and reliable. A government responsible for sensitive public records, national security or essential infrastructure may be less comfortable with the same arrangement.
The difference between convenience and strategic dependence becomes more significant as AI systems acquire greater responsibility.
Global reach does not guarantee lasting control
The conventional technology playbook suggests that scale creates an advantage. More users generate more revenue, wider distribution attracts developers and a larger ecosystem makes a platform harder to replace.
AI companies certainly benefit from these dynamics. But their control is not absolute.
Open-weight models, for example, allow developers to download model parameters and run them on infrastructure they control, subject to the model's licence. Some can be fine-tuned, adapted for specific languages or deployed without sending every request to the original developer's servers.
This does not mean every open-weight model is fully open source. Training data, training code, development processes and usage restrictions may remain unavailable or limited. Nevertheless, models that can be run independently give users options that a closed, hosted service does not.
DeepSeek's open-weight releases have helped demonstrate how models developed outside the leading American AI companies can attract international attention. Meta has also released Llama models with weights available under its own licence, while other developers are building models intended for local deployment and specialised use.
These approaches create a strategic trade-off for their creators. Wider distribution can accelerate adoption and expand an ecosystem, but it also reduces the ability to control every deployment.
Once model weights are available, the original developer may have limited visibility into where the model is running, what additional data it is processing or how a downstream developer has modified it.
For customers, that independence can be an advantage. For the original developer, it means sacrificing some control in exchange for broader reach.
Governments are not necessarily willing to outsource their AI future
The struggle becomes more complicated when national sovereignty enters the picture.
Governments increasingly view AI as a strategic capability rather than merely another software subscription. Their concerns include the security of sensitive data, access to computing infrastructure, dependence on foreign suppliers, compliance with domestic laws and the ability to maintain essential services during geopolitical disputes.
A government using an overseas AI service may have little influence over its provider's pricing, model updates, service availability or internal safety policies. Depending on the contractual arrangements and applicable laws, it may also face restrictions on where data is processed and who can access it.
These concerns do not automatically make foreign AI services unsafe or unsuitable. International providers can offer capabilities that local institutions cannot economically build themselves. The problem arises when a country has no credible alternative and little negotiating power.
For African countries, this question is particularly important.
If ministries, banks, universities, hospitals and businesses become dependent on foreign AI platforms, the continent could gain access to advanced technology while remaining dependent on external providers for critical digital capabilities.
The economic consequences extend beyond subscription costs. AI systems can shape how local languages are represented, which markets receive investment, which problems are prioritised and where the profits from automation ultimately flow.
A country that consumes AI without developing local expertise, infrastructure and bargaining power may find that technological adoption does not translate into technological independence.
Africa's challenge: adoption without permanent dependence
Africa does not need to reproduce every stage of AI development independently. Building frontier models from scratch requires substantial computing resources, specialised talent, data infrastructure and sustained investment. For many countries, attempting to compete directly with the world's largest laboratories on every dimension would be financially unrealistic.
A more practical strategy combines access to international models with selective investment in local capabilities.
Governments and businesses can use external models where they offer clear value while developing local datasets, evaluation systems, language resources, technical expertise and secure deployment environments. Universities and research institutions can train specialists capable of assessing models rather than merely purchasing them.
Regional cooperation can also improve the economics. Shared computing facilities, research partnerships and cross-border data-governance frameworks could help African institutions pool resources instead of duplicating expensive infrastructure across individual countries.
Local deployment is especially valuable where connectivity is unreliable, data sensitivity is high or a language is poorly served by mainstream models. Smaller models adapted for particular tasks may sometimes provide better value than the largest general-purpose systems.
But sovereignty is not achieved simply by hosting a foreign model on a server located in Lagos, Nairobi or Kigali. Genuine control also depends on the licence, the ability to change providers, the availability of technical skills, the security of the infrastructure and the practical ability to maintain the system if the original supplier withdraws support.
The objective should be meaningful choice, not isolation.
The paradox of control
AI companies face a difficult strategic calculation.
If they keep their models closed, they can retain greater control over access, updates and commercial terms. They may also find it easier to impose usage restrictions and manage how their systems are delivered. However, customers worried about dependence may favour competitors offering more deployment flexibility.
If companies release model weights more widely, they can attract developers, encourage experimentation and extend their influence into markets they could not serve directly. But they also surrender some oversight of downstream uses and may make it easier for customers to switch providers.
Neither strategy guarantees dominance.
A closed model can become deeply embedded in business operations, but customers may still migrate if costs rise or a competitor becomes substantially better. An open-weight model can spread widely, but broad adoption does not necessarily translate into sustainable revenue for its creator.
The emerging market may therefore support several kinds of power at once: companies that control the strongest frontier models, cloud providers that control computing capacity, open-model developers that influence experimentation and local organisations that control specialised data and deployment.
The eventual balance will depend on performance, cost, regulation, security and the ability of users to change course.
Regulation will determine who controls the boundaries
Governments are also attempting to establish rules for how AI systems operate within their jurisdictions.
Privacy laws, data-protection requirements, cybersecurity rules, procurement standards and sector-specific regulations can all affect whether an AI provider may serve a particular market and under what conditions.
For international companies, this creates a tension between global standardisation and local compliance. A single model and service architecture may be cheaper to operate worldwide, but different countries may demand different safeguards, audit mechanisms, data-handling arrangements or accountability structures.
Some governments may require local data storage or particular security controls. Others may encourage domestic AI development through public procurement, research funding or infrastructure investment.
These policies can improve accountability and strengthen local capacity, but poorly designed restrictions can also raise costs, limit competition or protect inefficient domestic providers.
The challenge is to establish rules that preserve public interests without preventing beneficial innovation.
AI companies cannot reasonably expect every government to accept the same commercial terms indefinitely. Nor should governments assume that local ownership alone guarantees safety, competence or public accountability.
Both sides must demonstrate that their arrangements deliver genuine value.
The next competitive advantage may be the ability to switch
The AI industry often measures progress through benchmark scores, model size, user numbers and revenue. Those indicators matter, but they do not fully capture the strategic position of customers.
An organisation's ability to switch models may become just as important as the model it uses today.
Businesses can reduce dependence by adopting interoperable systems, keeping control of their own data, maintaining clear contractual exit provisions and avoiding unnecessary reliance on proprietary features. Where appropriate, they can test more than one provider and maintain alternative deployment options.
Governments can apply similar principles to public procurement. Contracts should clarify data ownership, retention, security responsibilities, audit rights, service continuity and what happens if a supplier changes its terms or withdraws from a market.
For African institutions, these measures can create bargaining power even when building a competitive frontier model is out of reach.
The objective is not to eliminate all dependence. Modern economies depend on international suppliers for many critical technologies. The objective is to ensure that dependence does not become an inability to negotiate, innovate or maintain essential services.
AI's future may be less centralised than its leaders expect
The largest AI companies have strong reasons to seek global reach and preserve control over their technology. Their investments in models, chips, data centres and distribution give them substantial advantages.
But AI is not simply a contest to put one model in every organisation. It is also a contest over how intelligence is delivered, governed and adapted to different circumstances.
Some customers will pay for closed, high-performance systems. Others will prioritise privacy, low costs, local languages or independent deployment. Some governments will rely on international providers while investing in domestic capabilities. Others may favour regional partnerships or open-weight models for particular applications.
That diversity makes total control difficult to achieve.
The companies that become most influential may not be those that dictate every condition to every customer. They may be those that make their systems useful enough to attract users while offering the reliability, flexibility and accountability needed to retain their trust.
For governments and businesses, the lesson is equally clear: adopting AI should not mean surrendering the ability to choose.
AI leaders may succeed in putting their models everywhere. But the more widely those models spread, the more users, developers and governments will demand a say in how they are used. Global reach is a source of power; permanent control is a much harder promise to keep.
Comments
Post a Comment