← All writing

Oct 27, 2026

The Cost of Building Enterprise SaaS in Africa

Banks want their own isolated cloud environment and want their data kept in Africa. The continent's only AWS region costs about 25% more than Virginia. Here's what that does to the economics of building enterprise SaaS for African customers.

I spent the last few months envisioning an enterprise solution that's near and dear to my heart, and about a month ago I got serious about bringing it to life. It's called TraceGraph. It reconstructs security incidents from audit logs. It takes the logs a bank already has (identity providers, databases, API gateways, core banking, and now AI agents acting on behalf of staff and customers) and rebuilds what happened as a graph. Every step traces back to the original record, and the evidence it exports can be verified offline by an examiner who doesn't trust us at all. I've made good progress on it. But that's not what I want to talk about today.

I want to talk about the cost of building a tech startup in Africa. The cost of compute is insane. Often it's not that African developers and founders aren't innovative enough or crazy enough. It's that it's genuinely hard to do.

I know what you're thinking: there are cost-friendly resources out there. That's true, to a point. In my experience, far more people here build consumer products than hosted enterprise SaaS, and I think cost is a big part of why. Consumer products can start on a cheap server and grow into something bigger. Enterprise products can't, because enterprises come with the toughest compliance requirements of all, and they bring them to the first meeting. A local bank won't be comfortable with its data sitting on a cheap server in Finland. A local insurer might say AWS is fine, but they choose the region. And for those who know: the cloud is easy to get into, but it is expensive.

What "our own runtime" actually means

Imagine building an enterprise product where, because of how sensitive the system and its data are, every customer insists on their own runtime. Not a row in your shared database with an organisation_id column, but their own isolated environment.

For TraceGraph, the hosted design gives each organisation:

  • its own cloud account, so one customer's mistakes or breaches can't reach another's,
  • its own containers running the product,
  • its own managed PostgreSQL database, with a standby in a second data centre,
  • its own storage, encrypted with its own key,
  • a region chosen at onboarding that never changes, and
  • evidence anchored to a storage bucket the customer controls and that even we can't overwrite.

None of that is gold-plating. It's what makes a bank's security team say yes. "Separate account, separate key, separate database" answers most of a third-party risk questionnaire in one line. It also means almost every line on the bill is per organisation: the load balancer, the firewall, the private network endpoints, the database, the containers, the keys, the logs, the backups.

So you open AWS's price list and do the sums. I did, line by line, for the smallest setup I'd be willing to put a bank on:

AWS region              Per org, per month   Per org, per year
--------------------------------------------------------------
US East (N. Virginia)                ~$345             ~$4,100
Europe (Ireland)                     ~$365             ~$4,400
Africa (Cape Town)                   ~$435             ~$5,200

Look at the last row. Cape Town is the only AWS region on the continent, and it's about 25% more expensive than Virginia. The database costs more, the storage costs more, the containers cost more, even the private network endpoints cost more. So the customers who most need their data to stay in Africa are the most expensive ones to serve. Keeping data on the continent costs extra, and the startup pays it.

That's about $5,200 a year per customer, before you've paid for anything else: not your engineers, not your sales people. And that's the small size. A busier bank, on a bigger database with more storage, is closer to $890 a month in Cape Town. If it's an AI product, add the cost of the models on top.

The worst part is that most of it is a floor. In Cape Town, around $250 of that monthly bill exists before the customer sends a single log line: the load balancer, the firewall, the network endpoints, the standby database, the security monitoring. An idle customer costs nearly as much as a busy one.

So what do you charge?

If your costs are that high, what do you charge per month? Remember that your market is African enterprises, many without the revenue to justify spending heavily on outside providers. And the bill comes in US dollars, while many of your customers earn in local currency. When the currency slides, your costs go up in their terms even though nothing on your side changed.

Here's the rough maths I use. Healthy SaaS businesses tend to keep infrastructure well under a quarter of revenue. If one customer costs $435 a month to run, that customer needs to pay somewhere around $1,700 to $2,200 a month just to keep infrastructure at 20-25% of what they pay. That's $20,000 to $26,000 a year, per customer, before support, sales or salaries. Some banks will pay that for incident forensics without blinking. Plenty of the institutions I'd most like to serve, the mid-sized banks, the microfinance lenders, the younger fintechs, will not.

If your solution is cloud-native from day one, at what point (which year, which customer) do you expect to break even? Every customer you sign carries over $5,000 a year in cost before you've earned a cent of margin. A pilot you run for free for three months costs you about $1,300 in infrastructure alone. A slow procurement cycle (and bank procurement is slow) means you pay for environments long before you invoice for them.

Scale doesn't make it go away, either. Fifty organisations in Cape Town at the small size is roughly $260,000 a year at steady state, and about half that in a first year where you're adding customers as you go. Put everyone in one shared database and the same fifty would cost a fraction of that. But the shared database is exactly what the bank said no to in the first meeting.

What I'm doing about it

A few things, none of them glamorous:

  1. Self-hosting stays a first-class option. TraceGraph runs as a single binary with PostgreSQL inside the customer's own network, even air-gapped. For a bank that already runs its own data centre, that's often what they wanted anyway, and the compute is on their bill, not mine.
  2. No customer infrastructure until there's a customer. Our demo, our public site, the operator console and our test organisations all run on one modest server, with synthetic data only. Only each test organisation's evidence anchors and backups go to AWS, into buckets locked so that not even I can delete them. It's honest about what it is: good enough to show the product and test it, and not where a real customer's data will ever live. The full per-organisation cloud design is written down, reviewed, and waiting for the first contract.
  3. Building so I can switch later. Creating a test organisation today goes through the same organisation registry, the same activation check and the same recorded bindings that production will use. Only the step that creates the infrastructure changes, so moving a customer onto their own cloud account means pointing that step somewhere new, not a rewrite.
  4. Committing only once the numbers are real. Reserving database capacity for a year cuts that line by roughly a third. Across the whole bill it saves maybe 10-15%. That's worth having, but only once I know how many customers I'm reserving for.

And then there's the second job.

Building is only half the work

As I'm building this, I've realised I have two jobs:

  1. Building the product.
  2. Applying for cloud credits like crazy (AWS Activate, Google for Startups, Microsoft for Startups, and anything else I can find). I'd never done this before, but I'm at the point where I have to learn or I'm ngmi.

Credits don't fix the economics. They buy time to find the customers who will. That's what makes them worth the paperwork.

The next time you see a successful enterprise product come out of Africa, remember how many hurdles that team had to clear along the way. It isn't easy.

How are you all doing it?