If you listen to the earnings calls — and I mean really listen, not just read the headlines — there's a theme emerging from every major AI player.
The cost of running AI is collapsing.
The Race to Zero
Two years ago, a single GPT-4 query cost real money. Enough that you had to think about it. Today, you can run hundreds of queries for pennies. By next year? The cost will be a rounding error.
The floor isn't zero. It's the cost of electricity to run the chips. But that's close enough to zero that it changes everything.
And here's what most people miss: the AI providers want it this way.
Think about it. Free refills at Burger King aren't charity — they get you in the door to buy the Whopper. The soda is a loss leader.
AI inference is becoming the same thing. Microsoft, Google, Amazon — they're not racing to sell you tokens. They're racing to host your data. To be your cloud. To own the platform where your business runs.
The AI is the free refill. Your data is the Whopper.
Once your documents are in OneDrive, your code is in GitHub, your databases are in Azure — switching costs are enormous. The AI that sits on top? That's just the sweetener that keeps you locked in.
Token $0 isn't a race to the bottom. It's a strategy.
What Happens When AI Is Free?
Here's the question nobody in the boardroom is asking yet:
When everyone has essentially free access to the same AI capabilities, what's the competitive advantage?
It's not the model. You'll be able to run GPT-6 or Claude 5 or whatever comes next for fractions of a cent. So will your competitor. So will the startup that launches next week.
It's not the interface. Chat interfaces are trivial to build. Every vendor will have one.
It's not the prompt engineering. That knowledge spreads in weeks. There are no durable secrets in prompting.
So what's left?
The context. Your data. Your processes. Your history. Your institutional knowledge. The stuff the model can't know unless you tell it — and the stuff your competitor can't copy because it's yours.
The 90/10 Split
Think of AI usefulness as two components:
- The model (10%) — the raw intelligence, the language understanding, the reasoning capability
- The context (90%) — the specific knowledge about YOUR business that makes the model's output actually useful
Right now, most companies are obsessing over the 10%. Which model should we use? Should we switch to the new one? What about fine-tuning?
Meanwhile, the 90% sits untouched. Scattered across databases, documents, email threads, and the heads of employees who've been there for a decade.
When the 10% becomes free, the 90% is all that's left.
What the Mag-7 Already Know
Listen to the earnings calls from Microsoft, Google, Amazon, Meta. They're not betting on model superiority. They're betting on data gravity.
Satya Nadella doesn't stay up at night worrying about whether GPT beats Claude. He sleeps great knowing your company's files are in SharePoint, your code is in GitHub, your databases are in Azure. The AI is just the cherry on top that makes you never want to leave.
Google isn't betting on Gemini being the best model. They're betting on it being connected to your email, your documents, your search history. You're not the customer — you're the context.
The giants have figured this out. They're not selling AI. They're selling the ecosystem where AI happens to be free. The model is the loss leader. Your data is the product.
And here's the kicker: it doesn't even matter who has the best model.
Google might have the best TPUs. OpenAI might have the smartest model. Doesn't matter. If your data lives in Azure, your inference is running on Microsoft's GPUs. Period. You're not moving petabytes of data to save a few cents per token. Nobody is.
Data gravity always wins. The model follows the data, not the other way around.
This is why token costs are racing to zero. Not because they have to — because they want them to.
What This Means for You
If you're a mid-sized business, you have a choice:
Option 1: Keep renting. Pay for ChatGPT Enterprise. Use Copilot. Hope the generic AI gets smart enough to be useful for your specific problems. It won't, but you can hope.
Option 2: Build the context layer. Map your business — the entities, relationships, processes, and rules that make you you. Create the knowledge structure that any AI can plug into. Own the 90%.
Option 1 is easier. Option 2 is a moat.
When token costs hit zero, the companies with context will use AI to answer questions their competitors can't even ask. Not because they have better AI. Because they've built the foundation that makes AI useful.
Walking the Talk
This isn't theoretical for us. We're building zeros on zeros.
Every decision we make gets captured as a node in the graph. Every user story. Every commit. Every bug fix. Every "why did we do it this way?" moment. We stuff it all into the graph and let the edges sort it out.
When we start a coding session, the system injects relevant decisions automatically. When we investigate a bug, it surfaces similar issues from the past. When we're about to make a choice that contradicts something we decided six months ago, it tells us.
The graph gets denser every day. Hundreds of nodes. Thousands of edges. Institutional knowledge that doesn't walk out the door, doesn't retire, doesn't forget.
That's the 90%. That's what we're building. And that's what we help our clients build.
The Window
Here's the thing about building context: it takes time. You can't buy institutional knowledge. You can't download it. You have to map it, structure it, refine it.
The companies that start now will have a 2-3 year head start by the time token costs hit zero. That's the window.
Your competitors are still arguing about which ChatGPT tier to subscribe to. They're debating prompt templates. They're waiting for the next model to solve their problems.
They're focused on the 10%.
The question is: are you?
A ChatGPT subscription isn't an AI strategy. It's a tool purchase. Strategy is building the context that makes tools useful. The model is becoming free. The context never will be.
