Most AI news is about models: which one is smartest, which one is fastest. But some of the most important stories this week were about money. Reading Tech Startups' roundup for 24 September, four business stories stood out to me. Together, they say a lot about where AI is heading, and what developers and businesses here in Ghana should do about it.
1. Cheap AI may not last
According to The Information, DeepSeek's revenue run rate has reached about $1 billion, double what it was a few months earlier. Part of how it got there: it raised its API prices by 2.3 to 4.5 times. The Chinese startup is now reportedly aiming to raise $7.5 billion by late October, at a valuation near $74 billion.
DeepSeek became famous for being cheap. Many developers, including plenty in Africa, chose it for exactly that reason. Now the price has jumped by up to 4.5×.
That's the lesson: the price of an AI API is a business decision, and it can change. Low prices help a company win users; once it has them, prices can rise. If your app or your client's app depends on one cheap API, a price change like this can wipe out your margins overnight.
2. Signs of a bubble
Two stories made me cautious about the scale of the AI spending boom.
Oracle invoked force majeure on Project Jupiter, its planned AI data center of roughly 2.45 gigawatts. Bloomberg reports that the notice protects Oracle from payment obligations if the facility isn't operating by 2028. Companies don't file notices like that for projects they're sure will be finished on time.
Firmus, an Australian AI data center operator, is preparing a $5 billion IPO for an October listing, despite expecting an after-tax loss of about $77 million for the first half of its financial year, according to Reuters. Its customers include Nvidia, Meta and OpenAI.
Neither story proves AI is a bubble. AI is clearly useful, and I use it every day. But enormous projects that might not finish on time, and loss-making companies raising billions, are what the peak of a spending boom tends to look like. If the boom cools, the companies that overspent will look for revenue elsewhere, and higher prices for customers are an obvious place.
3. Africa is being left out
Microsoft committed more than $10 billion to AI, cloud and connectivity infrastructure in the Middle East, across Saudi Arabia, the UAE, Qatar and Kuwait through 2030, including more than $400 million for connectivity. Its partners there include G42, HUMAIN and QAI.
That's good news for the Middle East. But reading it from Accra, the question is obvious: where is the equivalent for Africa? Africa has seen some cloud and data center investment over the years, but nothing on this scale was in this week's news.
It matters because infrastructure decides who benefits. When the data centers are far away, African businesses pay more for slower AI services, their data leaves the continent, and the jobs building and running that infrastructure go elsewhere.
4. Being built in isn't enough
This one surprised me. According to the Financial Times, OpenAI says its ChatGPT integration in Apple's devices "dramatically underperformed." OpenAI expected a "halo effect" from being built into iPhones, but adoption came in far below expectations, and it lowered its forecasts.
If ChatGPT inside the iPhone can't guarantee users, no deal can. People use AI when it solves a real problem for them, not because it's installed. For anyone building AI features for clients, the lesson is to start with a real problem, not the technology.
What developers and businesses in Ghana should do
Put these stories together and the message is clear: AI is powerful, but its prices, its infrastructure and its business models are still unsettled. Here's how I'd protect myself and my clients.
Don't depend on one AI provider. Keep every AI call in your app behind one internal function or service, instead of scattering one provider's code everywhere. If prices jump or a service goes down, you switch providers in one place instead of rewriting the app.
Budget for AI costs rising. When you quote a client for a project with AI features, don't price it on today's API rates as if they'll last forever. Leave room for increases, and make sure the contract says who pays if they happen.
Use small and open models where you can. Not every task needs the biggest model. As I wrote in my post on four AI releases, small models that run on an ordinary CPU are getting good enough for many business tasks. A model you run yourself can't have its price raised by someone else. And for the big models, compare what a real task actually costs before you commit.
Push for local infrastructure. This one is bigger than any single developer. Local data centers mean lower latency, lower costs, and data that stays in the country. Governments, telcos and investors should treat AI infrastructure in Ghana and across Africa as a priority, and the tech community should keep saying so.
Where this leaves us
This week's money stories are a reminder that the AI industry is still being built, financially as much as technically. Prices will change, some big bets won't pay off, and the infrastructure is being built mostly elsewhere. None of that means stop using AI. It means use it the way you'd use any critical supplier: with a backup plan, a realistic budget, and your eyes open.
News in this post comes from Tech Startups' roundup for 24 September 2026, which cites The Information, Bloomberg, Reuters, Microsoft and the Financial Times. The opinions are my own. This post was drafted with help from Claude.