Africa’s AI Startups Have a Dollar Problem

An African startup can build for local businesses, charge in local currency, and still have its economics determined by a bill arriving in US dollars. Its customers may be shops in Lagos, schools in Accra, or service businesses in Nairobi. Behind the product, however, the model hosting, cloud infrastructure, and other software services may be priced for an entirely different market.

That creates a question founders need to answer before promising affordable AI: what happens when the currency customers earn buys less of the infrastructure the product needs?

The answer reaches beyond startup margins. It can affect subscription prices, service quality, business continuity, and who gets access to useful technology.

Open models still cost money to run

There is an important distinction here. An open model is not necessarily a free service. Some models described as open source only make their trained parameters, or “weights”, available under particular terms. The Open Source Initiative distinguishes that from the broader requirements of open source AI. Either way, somebody still has to run the model. A startup may download it and pay for computing capacity, or use a hosting provider that charges for each request’s usage. Together AI, for example, lists dollar-denominated prices for running models through its API.

The model developer, hosting provider, and application business may also be three different companies in three different countries. The economic exposure comes from the billing currency and operating arrangements, not simply the model’s origin.

When the exchange rate moves, margins move

Consider a hypothetical Nigerian startup collecting ₦10 million a month from customers and spending $2,000 on AI and cloud services. At an illustrative exchange rate of ₦1,500 to the dollar, that bill costs ₦3 million. At ₦1,800, the same bill costs ₦3.6 million.

Nothing about the product has improved. Customer numbers and usage have not changed. Yet the company has lost ₦600,000 of monthly contribution before paying its other expenses. Its dollar infrastructure bill has risen from 30% to 36% of revenue. These are illustrative figures, not current exchange rates or a complete profit calculation.

How customers feel the pressure

The founder must absorb the increase, raise prices, reduce costs, or combine those responses. Each choice eventually reaches the customer.

Higher prices can make a useful tool harder for a small business to justify. Keeping the headline price unchanged might mean fewer included tasks, slower processing, or additional charges for expensive features. A poorly managed switch to cheaper models could reduce reliability. If the startup cannot sustain the service, customers may have to replace a tool around which they have already organized their work.

For a business using AI in customer support, stock management or document processing, that disruption can cost more than the subscription. Choosing a supplier increasingly means understanding how it can keep delivering at the advertised price.

More usage can mean more cost

AI also complicates a familiar startup ambition: getting customers to use the product more. With usage-based model services, additional activity creates additional costs. One customer request may trigger several model calls, searches, and retries. An unlimited plan can therefore become expensive precisely because customers find it useful.

A startup can grow its customer base while earning less from each customer. The danger is greater when free infrastructure credits temporarily hide the actual cost of serving them.

This pressure could also shape which customers founders pursue. If serving a local microbusiness requires substantial computing usage but supports only a small subscription, founders may shift towards banks, large enterprises, or overseas customers with bigger budgets. That can be commercially sensible. But it risks leaving the businesses most sensitive to price with fewer suitable products. This is a possible consequence of the economics, not a prediction that every African startup will follow the same path.

The outcome differs by market

Nor should Africa be treated as one currency market. Exchange-rate arrangements, currency movements, and access to foreign exchange differ across countries. An appreciating local currency can reduce the value of a dollar bill. A startup earning dollars from export customers may partially offset this. The vulnerability is greatest when local revenue cannot adjust as quickly as foreign costs.

A powerful counterweight is also emerging: AI can become cheaper. Stanford’s 2025 AI Index reported that the inference cost of achieving roughly GPT-3.5-level performance fell more than 280-fold between November 2022 and October 2024. That historical decline doesn’t guarantee comparable future savings for every application, but it shows why a dollar bill doesn’t automatically make a local business model unviable.

The practical question is whether those savings reach the startup’s actual workload. Lower prices help less if the product simultaneously starts processing longer documents or running more elaborate agents.

How founders can build resilience

Founders should therefore measure the cost of completing a useful customer task, including retries and any human review. A smaller model that reliably handles a routine task may make more commercial sense than sending everything to the most expensive model available. Reusing suitable results, limiting unnecessary processing, and placing clear usage allowances around subscriptions can also improve the economics.

Pricing needs the same discipline. A predictable local-currency subscription can include a defined amount of work, with understandable charges beyond it. Founders should test whether that promise survives exchange-rate changes, heavier usage, and the expiry of promotional credits. Simply quoting customers in dollars transfers the problem to buyers whose income may still be entirely local.

Running models independently or using African hosting providers deserves consideration, but neither automatically removes dollar exposure. Hardware, infrastructure finance and other inputs may still carry foreign-currency costs. The comparison should include staffing, electricity, reliability and how fully the computing capacity will be used.

Build for the currencies customers earn

For Tullopy’s builders, the opportunity remains substantial. A product that demonstrably saves a business time, reduces errors or helps it earn more can justify its cost. Local knowledge, distribution, trust, and integration into everyday work can create lasting value even when the underlying model comes from elsewhere.

But affordability has to survive beyond the demo and the free credits. Building AI for Africa means designing a business that works in the currencies its customers actually earn.