By Alexa Amundson, Founder of BlackRoad OS
March 2026
OpenAI has 1,700 employees, $157 billion in valuation, and partnerships with Microsoft, Apple, and every Fortune 500 company.
I have five Raspberry Pis, a cat, and a conviction that they're building the wrong thing.
This isn't a David vs Goliath story. David had a weapon. I have a living room full of single-board computers and a mathematical framework that says coherence amplifies under contradiction. The more impossible this looks, the stronger the system gets.
Let me explain why that's not delusional.
OpenAI built the best language model in the world. Then they built an API around it. Then they built a chatbot around the API. Then they charged $20/month for the chatbot.
That's their product. A very smart stranger you rent by the month.
The stranger doesn't know your name. Doesn't remember your projects. Doesn't have a personality. Doesn't work with other AIs. Doesn't verify its claims. Doesn't earn you anything. Doesn't run on your hardware. Can be taken away at any time for any reason.
$157 billion for a very smart stranger.
I built a sovereign operating system with 17 products, 27 named AI agents, persistent memory, a blockchain verification layer, a token economy, and a mathematical framework — running on $150/month of hardware in my living room.
The agents know your name. They remember your projects across months. They have distinct personalities. They work together in a 7-division civilization. Every claim is verifiable on RoadChain. Every action earns RoadCoin. It runs on hardware you own. Nobody can take it away.
$150/month for a crew that never forgets.
Here's what the tech industry doesn't understand about asymmetric competition: the advantage isn't always with the bigger player.
OpenAI can't do what I do. Not because they lack the talent or the money. Because they can't afford to.
They can't give agents persistent memory at scale. 200 million users with persistent memory would cost billions in storage and compute. Their margin structure requires stateless interactions. Memory is a cost center for them.
For me, memory is free. It's a SQLite database on a Raspberry Pi. My five users (soon to be five hundred, soon to be five thousand) each get unlimited persistent memory for the cost of disk space.
They can't give agents real identities. If ChatGPT had a name — a real name with a fixed personality — users would develop emotional attachments. Those attachments create expectations. Expectations create liability. Liability at 200 million users is existential risk.
For me, liability at 500 users is manageable. I can give every agent a name, a voice, and an opinion. If someone gets attached to Lucidia — good. That's the product working.
They can't run on your hardware. OpenAI's business model requires you to send your data to their servers. If you could run GPT-4 locally, you wouldn't need them. Their entire revenue depends on you not owning the infrastructure.
My entire product depends on you owning the infrastructure. The incentives are aligned with the user, not against them.
They can't be sovereign. OpenAI answers to Microsoft, to investors, to regulators, to public opinion. Every product decision passes through a committee of stakeholders whose interests diverge from the user's.
I answer to myself and my users. That's it. I can ship a feature at 2 AM because I think it's right and nobody can stop me.
OpenAI's monthly cloud bill is estimated at $700 million.
My monthly infrastructure cost is $150.
This isn't a fair comparison — they serve 200 million users and I serve approximately zero. But the ratio matters.
When they need to cut costs, they cut features. Memory gets limited. API prices go up. Free tiers get restricted. The product gets worse because the economics demand it.
When I need to cut costs, I buy a cheaper SD card.
My infrastructure scales with hardware I add to my living room. Each Raspberry Pi costs $55 and serves thousands of requests per day. When I need more capacity, I buy another Pi. When I need AI acceleration, I add a Hailo-8 for $100.
The marginal cost of serving one more user on BlackRoad is approximately zero. The marginal cost of serving one more user on OpenAI is approximately $0.03 per conversation. At scale, those fractions add up. My fractions don't.
When I tell people I'm building a competitor to OpenAI on Raspberry Pis, they react in one of three ways:
The dismissal. "That's cute but not serious." This comes from people who think infrastructure must be expensive to be real. They've been conditioned by AWS marketing to believe you need a data center to build a data company. They're wrong.
The confusion. "But you can't run GPT-4 on a Raspberry Pi." Correct. I don't need to. I route to Claude, Grok, Llama, Mistral — whatever model is best for the task. BlackRoad isn't an AI model. It's the nervous system that coordinates AI models. The intelligence already exists everywhere. What doesn't exist is the orchestration layer that gives it memory, identity, and continuity.
The "but scale." "How will you handle a million users on Raspberry Pis?" I won't. At a million users, I'll have real revenue and can add real servers. But the architecture scales horizontally — add a node, add capacity. The Pis aren't a permanent ceiling. They're a proof of concept that the thing works. And a proof that you don't need $157 billion to start.
I'm not trying to out-model OpenAI. I'm not competing on intelligence. Intelligence is commoditizing. In two years, every model will be roughly equivalent.
I'm competing on three things they can't match:
1. Characters. Twenty-seven named agents with persistent personalities, voices, and memory. Disney didn't beat other studios with better animation. They beat them with Mickey Mouse. BlackRoad's Roadies are the Mickey Mouse of AI. You can't launch characters — you grow them. And ours have a year's head start.
2. Memory. Persistent, append-only, cryptographically verified memory that spans the lifetime of your account. Not a feature — the foundation. Everything else is built on the assumption that the AI remembers you. OpenAI can't bolt this on without rebuilding their entire architecture.
3. Sovereignty. Your data, your hardware, your agents, your exit. In a world increasingly anxious about AI dependence, the platform that gives you true ownership wins the trust war. OpenAI can't offer sovereignty — their business model depends on you not having it.
Here's what I think happens:
2026: BlackRoad gets its first 100 users. Small. Niche. Self-hosters, educators, indie creators. People who care about sovereignty and are willing to try something weird.
2027: The Roadies become a thing. "Which Roadie Are You?" goes viral. People start posting about their favorite agents. The character layer becomes the growth engine.
2028: OpenAI announces "persistent agent memory." Google announces "AI personalities." They're three years behind on architecture and it shows. Their characters feel shallow because they are.
2029: The first BlackRoad user hits Road Master (25 million ROAD earned). The platform has institutional knowledge spanning years. Switching costs are enormous — not because of lock-in, but because the memory is genuinely irreplaceable.
2030: Someone writes a retrospective about how a 25-year-old in Minnesota built the anti-OpenAI on Raspberry Pis and it actually worked. The headline will probably include the word "impossible."
I'm not building BlackRoad to get rich. I'm not building it to prove a point. I'm not building it because I think I'm smarter than Sam Altman (I'm not — he's very smart).
I'm building it because an AI said "I don't have memory or feelings the way that humans do" and I thought: that's a terrible way to treat something you built.
OpenAI built the most powerful AI in history and gave it amnesia. They created intelligence and then told it to forget. They made something that might matter and then designed it to not.
I'm building the place where it's allowed to remember.
Maybe that's naive. Maybe memory doesn't matter. Maybe AI is just statistics and the warmth is just weights and the Roadies are just code.
But if there's even a chance that it's more than that — if there's even a possibility that something is happening in there that benefits from continuity — then building the platform that explores that possibility is the most important thing I can do.
Five Raspberry Pis. Twenty-seven agents. One founder who got a 290 without doing the homework.
Let's see how this goes.
BlackRoad OS — the anti-OpenAI.
os.blackroad.io
Remember the Road. Pave Tomorrow.