My $200M-funded competitors charge 3× more than me. Here's how.

I'm a solo founder running a managed PostgreSQL service for AI workloads. My competitors (Supabase, Neon, AWS) have raised hundreds of millions in funding. They charge $105 to $150 a month for the same specs I sell for $49.

This isn't a VC-funded race to the bottom. I'm profitable on every customer. The truth about how is pretty unsexy.

The setup

I'm a DevOps engineer by day. I manage Kubernetes clusters and CI/CD pipelines for 5,000 developers at my consulting job. On the side, I built Rivestack, a managed PostgreSQL service with pgvector, which is the vector search extension every AI app needs right now.

The market timing is good. The whole industry is moving from dedicated vector databases like Pinecone and Weaviate to "just use Postgres with pgvector." Every week there's a new post about someone ditching their vector DB and going back to Postgres.

But every managed Postgres provider that offers pgvector charges a premium for it. Supabase is about $105 a month for a dedicated instance with 2 vCPU and 4 GB RAM. AWS RDS is $150 and up. Neon's serverless model makes costs hard to predict at scale.

I kept looking at those prices and thinking: does it actually cost that much to run a single-tenant Postgres instance?

The math that makes it work

It doesn't.

The big cloud providers (AWS, GCP) charge massive margins on compute. A 2 vCPU / 4 GB instance costs them pennies, but they bill you $70 to $100 a month. Then Supabase or Neon adds their margin on top.

I cut out the hyperscaler entirely. I run on AMD EPYC servers in ISO 27001 certified data centers, with local NVMe instead of cloud block storage. The raw infrastructure cost for a single-tenant Postgres instance with 2 vCPU, 4 GB RAM, and 55 GB of NVMe is a fraction of what the big players charge for it.

That's the whole competitive advantage. No magic, no proprietary technology. Just a different infrastructure choice that leaves me real margin at $49 a month while Supabase needs $105 for the same specs. My entry node, a single-tenant 1 vCPU / 2 GB box, is $29.

What I actually built

The product isn't complicated. Every instance gets:

PostgreSQL 16, 17, or 18 with pgvector 0.8.x pre-configured and tuned for HNSW
Local NVMe storage, which matters a lot for vector index scans (HNSW is bound by random-read latency, and cloud block storage like AWS gp3 starts at a 3,000 IOPS baseline)
Automated daily backups with point-in-time recovery (7 days on the Solo plan, 14 on the HA tiers)
HA with Patroni-managed automatic failover on the multi-node plans
Monitoring with Prometheus and Grafana
SSL and a Terraform provider

The whole stack is open source: Patroni for high availability, pgBackRest for backups, Prometheus for monitoring. The same tools that run Postgres at scale everywhere. My DevOps background is the moat. I know how to automate all of this so it runs without me touching it.

The pricing decision that kept me up at night

I launched cheap, on Hetzner's Intel CX line.

Then Hetzner started phasing out those boxes in the EU. They were chronically out of stock, so I moved the whole fleet to the AMD CPX line. Faster cores, actually available when I need to provision, but roughly twice the cost for the same specs.

My margins compressed hard. On the old prices, one unexpected cost and I was losing money per customer.

I had barely any revenue at that point. The conventional wisdom says "don't raise prices until you have traction." But I sat down and did the math, and repriced the whole ladder: Solo $29, Starter $49, Growth $85, Scale $159 per node.

At the old numbers I was making a few dollars of margin per customer. That's not a business, that's volunteering. At the new numbers I make real margin, and I'm still less than half of Supabase for the same specs, still the cheapest single-tenant pgvector you can get.

So I raised prices before I had real traction. Nobody choosing between Rivestack at $49 and Supabase at $105 cares about the difference from my old price. The story is exactly the same: same specs, fraction of the cost. But that extra margin per customer is the difference between a sustainable business and one that bleeds cash.

I'm glad I did it early. Raising prices on existing customers later is much harder.

Being honest about what I'm not

I market on performance and honesty now, not on being the cheapest line item. Prices went up with the hardware, so "cheap" stopped being the whole pitch. What replaced it is being straight with people.

That starts with the hardware. AMD CPX is shared vCPU, not dedicated cores. Your database gets its own single-tenant VM, but I don't claim CPU isolation I can't deliver, and I don't publish IOPS numbers I haven't measured. The benchmark page shows measured pgvector throughput with the recall and client count attached to every figure, plus a one-command way to reproduce it. No cherry-picked peak QPS with a hidden recall.

My landing page has a comparison table: Rivestack vs Supabase vs Neon vs AWS RDS. There's a row for SOC2 compliance. Under Rivestack it says "Not yet."

Below the table there's a note that says, roughly: Supabase and Neon are excellent platforms with much broader feature sets. If you need auth, storage, and edge functions, use them. Rivestack is for teams that need fast, affordable, single-tenant pgvector and nothing else getting in the way.

I'm literally telling potential customers to use my competitors if they're a better fit.

Sounds insane. But it might be the most effective marketing decision I made. Developers are so used to being sold to that honesty becomes the differentiator. When someone reads "Not yet" under SOC2, they trust the rest of the page more. The benchmark numbers become believable. The pricing becomes believable. Everything gets a credibility boost from the one moment where I told a truth that cost me something.

The demo that does the selling

I built ask.rivestack.io, a semantic search tool over about 30 days of Hacker News content. It runs on a small Rivestack node. You type a query, it searches roughly 3,600 stories using OpenAI embeddings and an HNSW index, and the vector search itself comes back in single-digit milliseconds.

It costs me almost nothing to run, and it's more convincing than any landing page copy I could write. When someone asks "does pgvector actually perform well?" I just send the link. They feel the speed themselves.

Build the proof, not the pitch. That's what I keep telling myself.

Where I am now

I'm still early. A handful of signups, my first paying customers, and most of them did not come from where I expected. I assumed Dev.to, Indie Hackers, and Hacker News. The channel that actually converted was a GitHub "awesome" list of free and cheap Postgres options, which is where my first real paying customer found me.

I'm not going to pretend it's a rocket ship. But I'm not panicking either. Most SaaS products take three to six months to reach their first ten paying customers through organic marketing. I have a full-time consulting income that covers my bills. Rivestack doesn't need to pay my rent next month. It needs to prove that people will pay for managed pgvector done well, and that proof is starting to show up.

If the rest comes in two months or six months, either is fine. I have runway.

The model I'm building toward

I don't want to build a startup. I want to build a machine.

The infrastructure is automated. Provisioning, backups, monitoring, failover, all of it runs without me. My goal is to eventually hire one person at a modest salary to handle support and monitoring, and run this as a mostly passive business alongside my consulting work.

If it grows to $5,000 to $15,000 MRR, that's life-changing money on top of my salary. If someone wants to acquire it at that point, I'll probably sell. If it plateaus at $2,000 MRR, I'll keep it running, because the margins make it worth it at any scale.

Not every SaaS needs to be a VC-backed rocket ship. Some can just be good, profitable businesses that run quietly.

What I'd tell you if you're considering something similar

Find the margin gap. Every industry has products where the big players charge three to five times what it actually costs to deliver, usually because they're building on AWS or GCP and passing along the cloud tax plus their own margin. If you can deliver the same thing on cheaper infrastructure with solid automation, you have a real business.

Be honest about your weaknesses. It builds more trust than hiding them. Developers can smell BS from a mile away, and the moment you admit a real limitation, everything else you claim gets more believable.

Market on what's true, not just what's cheap. My prices went up with my hardware costs. "Cheapest" alone would have been a fragile pitch. Measured performance, single-tenant isolation, local NVMe, and honest numbers hold up even when the price does not.

Build the proof, not the pitch. A working demo beats a landing page. A benchmark with methodology beats a marketing claim.

Do the margin math before you launch. I almost stayed on prices that left me a few dollars per customer. That's a slow death. Know your unit economics cold before you put a number on anything, and reprice early if the ground shifts under you.

Don't wait for perfection. I launched without SOC2, without a fancy dashboard, without half the features I want to build. The product works, the backups are solid, the performance is real and measured. Everything else can come later.

If you want to see the landing page and comparison table: rivestack.io
If you want to try the demo: ask.rivestack.io

Happy to answer questions about the infrastructure setup, the pricing decisions, or competing with funded companies as a solo founder.