If we’re serving African users from US-East-1, we’re paying a 30-50% margin tax we don’t need to pay — and the fix isn’t “move everything to a local provider,” it’s a deliberate hybrid stack where compute sits close to the user, storage sits where egress is cheapest, and the control plane stays on whichever hyperscaler our team already knows. Most diaspora operators get this wrong in the same direction: they default to AWS us-east-1 because their first engineer was American-trained, then watch their AWS bill outpace their revenue growth by month nine.
This is a margin problem dressed up as a technical problem. We want to walk through the actual numbers, the three real options on the table in 2024-25, and the hybrid pattern most of us should be running.
Why does the default AWS setup quietly kill African-facing margin?
Three reasons, in descending order of damage.
First, egress fees. AWS charges roughly $0.09 per GB to move data out of US regions to the public internet — and every API call, image load, and video stream we serve to a user in Soweto or Surulere is egress. A fintech app pushing 50TB monthly to African users from us-east-1 is paying ~$4,500/month in egress alone. The same workload on AWS Cape Town (af-south-1) drops egress only marginally because AWS prices af-south-1 egress at around $0.154/GB — higher than US regions. That’s not a typo. The Cape Town region is more expensive than Virginia on most line items, sometimes by 20-40%.
Second, latency tax on conversion. Every 100ms of added latency drops e-commerce conversion by roughly 1% (this is the old Akamai/Amazon number that’s been replicated across diaspora markets by Paystack and Flutterwave engineering teams in their public talks). A Joburg user hitting a server in Virginia eats 200-250ms round-trip before TLS handshake. We’re losing 2-3% of conversion before our checkout even loads.
Third, FX exposure on a USD-denominated bill. Our revenue is in rand, naira, or cedi. Our AWS invoice is in dollars. When the rand slid past R19/USD in 2023, every operator running USD infrastructure took a silent 15% cost hike with no corresponding revenue lift. This is the part most cloud cost analyses skip entirely because they’re written from a US perspective.
What are the three real options on the table?
1. Stay on a hyperscaler, move to a local region
AWS Cape Town (af-south-1), Azure South Africa North (Joburg), and Google Cloud Johannesburg are the three African hyperscaler regions worth using. They solve latency. They do not solve cost — and in many cases make it worse. The case for staying hyperscaler is operational: our team already knows the tooling, the IAM model, the deployment pipeline. Switching cost is real.
2. Move to an African cloud provider
Africa Data Centres (Cassava-owned, present in Joburg, Cape Town, Lagos, Nairobi, Accra), Teraco (the Joburg-headquartered colo giant that’s effectively the carrier-neutral spine of Southern African internet), Rack Centre in Lagos, and MainOne (now Equinix-owned) are the serious infrastructure players. For raw colocation or bare-metal compute, they’re 40-60% cheaper than equivalent hyperscaler spend. The catch: we’re back to managing our own Kubernetes, our own load balancing, our own backups. If our team is two engineers, this kills us.
3. Build hybrid
This is what most of us should be running. Keep the control plane (auth, billing, admin tooling, analytics warehouse) on whichever hyperscaler we already know. Push the edge layer — CDN, image serving, API gateway, regional caches — onto African infrastructure via Cloudflare (which now has 15+ African PoPs including Joburg, Cape Town, Lagos, Mombasa) or local colos. Egress drops dramatically because Cloudflare doesn’t charge for it. Latency drops because traffic terminates locally. The control plane stays familiar.
What does the hybrid stack actually look like in production?
A working pattern we’ve seen produce real margin recovery:
- Edge / CDN / WAF: Cloudflare (zero egress, African PoPs, ~$200/month for most operators at this stage)
- Origin compute for African users: Africa Data Centres or Teraco-hosted bare metal, or AWS af-south-1 if hyperscaler simplicity is worth the premium
- Database: Managed Postgres in the same African region (Neon, Supabase, or self-managed RDS in af-south-1)
- Object storage: Backblaze B2 or Cloudflare R2 (both have $0 egress to Cloudflare CDN — this is the single biggest cost lever)
- Analytics warehouse and batch jobs: Stay in US-region BigQuery or Snowflake where compute is cheapest and the data doesn’t need to be hot
For an operator currently paying $8,000/month on a US-region AWS bill serving 80% African traffic, this restructure typically lands the bill between $3,200 and $4,500 — a 40-55% reduction with better performance for end users. That’s not theoretical. That’s the pattern Paystack, Flutterwave, and the larger Joburg-based fintechs have been publicly migrating toward since 2022.
“What if we’re too small to justify a hybrid setup?”
If our AWS bill is under $1,500/month, stay on the hyperscaler but move our region to af-south-1 and put Cloudflare in front. That’s a one-afternoon migration that handles 80% of the cost and latency problem until we’re large enough for the full hybrid pattern.
The Joburg infrastructure argument we should be paying attention to
Teraco’s NAPAfrica internet exchange in Joburg now peers more African traffic than any other point on the continent. AWS, Azure, Google, Cloudflare, Akamai, Netflix, and Meta all have direct presence there. What this means operationally: if our origin sits in a Teraco-adjacent rack in Joburg, our traffic reaches every major cloud and CDN with sub-5ms latency and effectively zero transit cost. The Joburg infrastructure spine is the closest thing the continent has to a Northern Virginia or Ashburn — and most diaspora operators don’t know it exists because the AWS console doesn’t surface it.
This is the kind of infrastructure literacy our generation of builders needs to have priced in. The continent’s internet backbone isn’t a future event. It’s running, it’s commercially mature, and operators in Lagos, Nairobi, and Accra route through Joburg whether they realize it or not.
The move we should make this quarter
Pull our last three months of cloud invoices. Calculate two numbers: percentage of egress fees as a share of total bill, and percentage of users served from non-US geographies. If egress is over 15% of the bill and African users are over 50% of traffic, we’re the case study. The migration to a Cloudflare-fronted hybrid stack is a two-to-six-week project depending on how clean our infrastructure-as-code is. The payback period is usually under 90 days.
This is the kind of unglamorous margin work that compounds. We’re building businesses that need to throw cash, not burn it on Amazon’s transatlantic bandwidth.
For operators wearing the work, the Armed to Build drop at store.blackarcscend.com is where some of us mark it — not merch, a flag for those of us who stopped waiting for the diaspora’s infrastructure to be built for us and started building it.
If we’re running an African-facing product and the infrastructure conversation above is one we’ve been putting off, the Black Arcscend Accelerator runs operators through exactly this kind of cost-and-architecture restructuring inside a cohort of peers solving the same problem. Applications for the next cohort are open at blackarcscend.com — bring your AWS bill.
