AI: How to Save Us All
The case for structural separation in AI: three layers — infrastructure, models, and services — to prevent the most predictable monopoly in history and preserve the knowledge economy.
Everyone is asking the wrong question about AI and jobs. They ask: "How do we slow it down?" or "How do we retrain people?" or "What about UBI?" These are band-aids on a structural wound. The real question is: who is allowed to do what?
Here's the answer: model companies can't provide services. Service companies can't build foundation models. Infrastructure providers can't compete on their own platforms.
Three layers. Structurally separated. That's it.
This isn't about slowing innovation. It's about preventing the most predictable monopoly in human history from forming while we all watch it happen in real time.
We've Been Here Before
This idea isn't new. We've faced this exact problem multiple times and solved it the same way every time.
When commercial banks started gambling with depositors' money, we passed Glass-Steagall and separated banking from speculation. When AT&T owned every phone line and every phone in America, we broke them up and separated infrastructure from services. When electricity generation was bundled with distribution, we unbundled them so that the company running the wires couldn't crush independent power producers.
The pattern is always the same: when the entity that controls the platform also competes on the platform, competition dies. Innovation consolidates. And eventually, the public pays for it.
AI is the next iteration of this pattern. Except this time, the stakes aren't just phone bills or electricity rates. The stakes are the entire knowledge economy.
The Cloud Mistake We Already Made
If you want to see what happens without structural separation, look at cloud computing. We already ran this experiment and lost.
Amazon Web Services started as scalable infrastructure — compute, storage, networking. Simple. Beautiful. Then they noticed what their customers were building on top of their platform and started competing with them. ElastiCache gutted Redis Labs. DocumentDB went after MongoDB. Aurora undercut every managed database provider.
These companies built the open-source technology that made AWS valuable in the first place. AWS cloned it, offered it as a managed service with built-in distribution advantages, and there was nothing anyone could do about it. Redis and MongoDB had to change their entire licensing models just to survive. That's not competition. That's a landlord stealing your business plan from your mailbox and opening the same shop downstairs.
Google Cloud, Azure — they all did the same thing. They provide the infrastructure and compete on the services that run on top of it. We accepted this as normal. It was a mistake. And now we're about to repeat it at a scale that makes cloud computing look like a rehearsal.
What Happens If We Do Nothing
Without structural separation, the endgame is obvious. OpenAI builds GPT and launches an accounting service, a legal service, a consulting practice, a DevOps platform — all powered by their own model with cost advantages no independent firm can match. Google already owns the cloud (GCP), the model (Gemini), the distribution (Search, Android), and the consumer interface. That's four layers vertically integrated in one company.
Microsoft has Azure plus its OpenAI investment plus Copilot plus Office distribution. They don't just want a piece of every layer — they want to be every layer.
In this world, what happens to the consulting firm? The digital agency? The independent SaaS company? The freelance developer? They become tenants on someone else's platform, competing against their own landlord, paying rent to the company that's actively trying to replace them.
Every single knowledge-economy job — accounting, law, marketing, engineering, design, strategy — gets squeezed between the model company above and the automation pressure below. Not because AI is inherently destructive, but because we let the same entity control the tool and sell the work the tool produces.
The Three Layers
The proposal is simple enough to fit on a napkin:
Layer 1 — Infrastructure. AWS, Google Cloud, Azure provide compute, storage, and networking. Scalable infrastructure. Not managed databases. Not Redis-as-a-service. Not Kafka. Not AI model hosting with proprietary advantages. You rent the pipes. You don't run the plumbing business.
Layer 2 — Models. Anthropic builds Claude. OpenAI builds GPT. Google spins out Gemini as a separate entity. xAI operates Grok independently. They sell model access via API. That's it. No ChatGPT consumer product. No Claude.ai. No Gemini built into Search. The model company builds the best model it can and sells access to it. Period.
Layer 3 — Services. This is where the work happens. Consulting firms, agencies, SaaS companies, freelancers — they combine infrastructure and model access to build actual solutions for actual clients. This is where domain expertise matters. This is where human judgment matters. This is where relationships matter. This is where the jobs are.
Each layer does one thing. Each layer competes within its own lane. No vertical integration across layers.
The Price and Why It's Worth Paying
Yes, this means ChatGPT shuts down. Claude.ai shuts down. Gemini as a Google product shuts down. I know. I hear you.
But ask yourself: what are those products actually doing? They're the model company reaching past the model layer into the service layer. They're the foundation model provider competing with every company and every person who would otherwise build services on top of that model.
When ChatGPT answers a legal question, it's not just being a helpful tool — it's a law firm. When it writes marketing copy, it's a marketing agency. When it debugs code, it's a DevOps consultancy. The model company is providing the service and the engine simultaneously, and no independent firm can compete with free or near-free.
The price of structural separation is that consumers don't get free AI products from the model makers. The benefit is that thousands of companies build competing products on top of those models instead — with better specialization, more innovation, and actual market competition.
We didn't let AT&T provide free phone service to kill every independent telecom. We didn't let power generators offer free electricity to crush independent distributors. The logic is the same.
The China Paradox
Here's the part nobody wants to hear: China is already doing a version of this, and it's working.
The standard Western argument against regulation is "we can't afford to because China won't." But look at what's actually happening in China. DeepSeek released a frontier-competitive model at a fraction of the cost of Western equivalents. Qwen from Alibaba, models from Baidu — they're pushing open-weight models at an extraordinary pace. The model producers produce models. The application layer is separate. The infrastructure layer is largely commodity.
Now compare that to the American approach where Google owns the full stack from TPUs to Search results. Where Microsoft's investment in OpenAI gives it model access, cloud advantage, and product distribution in one integrated package. Where Meta builds the model, the platform, and the advertising service all under one roof.
The "free market" produced more concentration than China's state-managed approach. You can feel uncomfortable about that, but you can't argue with it.
DeepSeek built a frontier-competitive model for a fraction of the cost precisely because they weren't trying to own the whole stack. They focused on one layer and excelled. That's not an accident. That's the thesis proving itself in real time.
This Is Pro-Innovation
Let me be clear about what this proposal is and what it isn't.
It's not about slowing AI down. Model companies innovate as fast as they want. They compete ferociously with each other on model quality. That competition intensifies, not weakens, because models are their only product.
It's not about protecting jobs through artificial barriers. Service companies still need to adopt AI aggressively to compete. If your consulting firm isn't using AI, you lose to the one that is. But you're competing against other service firms, not against the model company itself.
It's not anti-market. It's the most pro-market proposal on the table. You're creating three thriving competitive markets instead of one integrated monopoly.
After the telecom unbundling, more ISPs emerged, not fewer. After Glass-Steagall, the financial sector grew, not shrank. Structural separation creates more economic activity, not less, because it creates more surface area for competition and entrepreneurship.
The Alternative Is Obvious
Without this, the trajectory is clear. Three to five companies will own the AI stack from infrastructure to consumer service. Every knowledge worker becomes dependent on platforms they don't control and can't compete with. The entire service economy — which is most of the economy — gets squeezed into irrelevance.
And then we'll have the political consequences. Displaced workers, concentrated power, democratic institutions unable to govern entities more powerful than most nations. We've seen this movie. We know how it ends.
Or we draw the lines now. Infrastructure stays infrastructure. Models stay models. Services stay services. Everyone grows. Everyone competes. Everyone innovates. But in their own lane.
The Three-Layer Rule. Simple enough to fit on a napkin. Important enough to save the economy.
It's a regulation, yes. But it's the kind of regulation that creates markets instead of constraining them. And right now, it might be the only idea on the table that actually works.
A Note on "Regulation"
If the word regulation made you flinch, consider this: you're already being regulated. You just didn't elect the regulators.
Apple decides what apps can exist on your phone — and takes 30% of every transaction for the privilege. Google controls which businesses get discovered online and buries the ones that don't pay. Amazon decides which products get visibility on the platform where half of all e-commerce happens. These aren't market outcomes. They're policies. Set by private companies. With no appeals process, no transparency, and no democratic accountability.
And then there's healthcare. HIPAA was supposed to protect patient data. In practice, it became a moat. Startups trying to innovate in healthcare spend years and millions navigating compliance — while the same companies selling you a phone, a search engine, and a smart speaker harvest your health data through fitness trackers, search queries, and voice assistants with virtually no oversight. The regulation protects incumbents. The data flows to the giants anyway.
So when someone says "regulation kills innovation," ask them: whose regulation? Because right now, the most consequential regulations in the economy aren't written by governments. They're written by platform companies, enforced unilaterally, and designed to protect their position — not yours.
The Three-Layer Rule doesn't add regulation to an unregulated market. It replaces corporate regulation with democratic regulation. It swaps rules written in boardrooms for rules written in public, with accountability, with the possibility of change.
You're going to be regulated either way. The only question is whether you get a vote.