- Get It Dunn
- Posts
- $16.3bn in six months.
Two numbers landed in the same week. Read together, they tell you exactly where the money is.
November 1, 2026
Read Time: 9 minutes
Anthropic filed to go public on the twenty sixth of October. Nasdaq, ticker ANT.
The filing put revenue at $16.3 billion for the first half of this year, against roughly $1.2 billion in the same period a year earlier.
That is not growth. That is one of the fastest revenue ramps anybody has ever put on paper.
Two days before that, Apollo's chief economist published a note looking at the other side of the transaction. Revenue per employee across the Magnificent Seven has climbed from around $550,000 in 2022 to almost $800,000 now. For the other 493 companies in the S&P 500, it has been flat since 2019. Profit margins the same story. He described the financial payoff for those 493 companies as "zero."
Both of those are true, in the same week, about the same technology.
Billions are moving. Almost nobody buying is getting anything back.
That gap is the entire reason this job exists.
The Model Is Not the Product
Here is the thing that filing actually tells you, underneath the headline number.
Anthropic is not selling outcomes. It is selling intelligence by the token to anybody who wants to buy it. What the buyer does with it afterwards is not their problem and was never meant to be.
So you end up with a market where the ingredient is extraordinary, available to everyone, and getting cheaper every quarter, while the thing people actually want has not arrived.
Businesses do not want intelligence. They want their quotes going out in two days instead of nine. They want to stop losing enquiries in a shared inbox. They want the person who does the invoicing to be able to take a holiday.
Nobody at a model company is going to walk into a sixty person engineering firm in Sheffield and work out which of those three matters most. That is not a gap in the product. It is not the product.
The S&P 493 bought the ingredient. They are flat because nobody did the cooking.
Half of It Went Through Someone Else's Shop
The detail in that filing I find most useful has nothing to do with the revenue.
A very large share of it, reported as close to half, arrives through the cloud arms of Amazon, Google and Microsoft rather than directly.
Think about what that means. The best models in the world, and half the money still comes through somebody else's platform.
Not because those clouds are better at AI. Because that is where the customers already were. They had the contracts, the billing relationships, the security sign offs, the procurement history. The buyer did not go looking. The thing turned up inside the environment they already lived in.
Those platforms are institutions now. People do not leave them. Once a business has its data, its identity management and its finance team all sitting inside one ecosystem, switching is not a software decision, it is a two year project nobody wants to run.
So distribution beat product, at the very top of this market, with the strongest product in it.
Which is an argument I made a few weeks ago from a much smaller angle. If you specialise in a platform, Microsoft or Google or whichever one your market actually runs on, you are not narrowing yourself. You are standing where the customers already are and where they are never going to leave.
The biggest AI company in the world just demonstrated that at scale.
Why Cheaper Tokens Do Not Change Your Price
The other thing that moved this week is the price of the ingredient. OpenAI shipped a new model at roughly half the previous cost.
This gets treated as though it should worry you. It should not, and the reason is worth being precise about.
You are not selling tokens. Your client pays their own usage costs, directly, to whoever they are using. It is a line on their bill and it has nothing to do with your fee.
And for the businesses we are talking about, that line is nowhere near their biggest cost anyway. The expensive part is not running the thing. The expensive part is getting it in.
Working out what is actually broken. Mapping how work moves through the business. Deciding what gets changed first and what gets left alone. Getting thirty people to do something differently on a Monday morning when they were perfectly happy doing it the old way on Friday.
That is where the money goes, and no price drop touches it.
If anything, cheaper tokens help you. The thing you are proposing gets cheaper to run every quarter while the value of it stays the same, which quietly improves the business case for every project you have ever scoped.
Where Your Work Actually Starts
So where is the line? What is safe to build a career on when the labs keep shipping?
Your work starts before theirs.
Understanding a specific business. Finding where it is actually losing money, as opposed to where the owner thinks it is. Knowing which of the twelve obvious opportunities is the one worth doing first in this particular company with these particular people.
No model does that, and it is not because the models are not clever enough. It is because that work requires being in the room, having the conversation, noticing the thing on the whiteboard that nobody mentioned.
Diagnosis first, then prescription. I have said that enough times in this newsletter that regular readers could probably finish the sentence. This week two numbers came out that make the point better than I have ever made it.
Sixteen billion dollars of intelligence sold. Zero payoff for the companies buying it.
The missing piece is not better models. It is somebody doing the diagnosis.
The Obvious Objection: You Are Building on Someone Else's Platform
Worth being straight about this, because it is the honest downside and I do not want to skip past it.
This market is consolidating. A handful of model companies, three cloud providers, and increasingly the same names appearing in both. Everything you build sits on top of infrastructure owned by companies with enormous pricing power and no particular obligation to you.
That is real.
But look at any other part of technology and tell me where it is different. Operating systems went to three. Cloud went to three. Mobile went to two. Browsers, databases, payment processing, all of it concentrates eventually.
This is simply what technology does. It is not good or bad, it is the shape of the thing, and building a business on top of somebody else's infrastructure has been the normal condition of every software and services company for twenty five years.
The way you manage it is not by avoiding it. It is by owning the part that does not belong to any vendor.
You own the relationship with the business. You own the understanding of how it works. You own the fact that when something breaks on a Tuesday, they call you.
Models get swapped. Platforms change pricing. The engagement survives all of it, because what the client is actually paying for is somebody who knows their business and answers the phone.
The dependency is real. It just sits underneath a layer that is genuinely yours.
Two Ways to Be the Missing Layer
So if the gap between billions spent and nothing gained is the opportunity, what do you actually do about it?
There are two routes, and the useful thing about both of them is that neither requires you to be technical.
Go and audit mid-market businesses. Twenty to a hundred and fifty employees. Walk in, map the workflows, find what is costing them money, and give them a prioritised plan. This is the diagnosis layer the entire market is missing, and it is why an audit is worth five figures to a business that has already spent more than that on tools it is not using.
You are not building models. You are the person who works out what should be built, then either oversees it or hands it to a partner. (Launch an AI consultancy)
Or go and deploy the finished thing into local businesses. The plumber, the roofer, the salon, the HVAC company. They are not going to run an assessment and they do not need one. They need a website that works, a receptionist that answers, reviews that come in, and a system that stops leads falling on the floor.
That is productised, it is the same package every time, and it does not require you to understand anything about a transformer. It requires you to understand a local business owner, which is a skill most consultants are worse at than they think. (Launch a local AI agency)
Different buyer, different motion, different life, as I wrote about last week. Both of them sit in exactly the same gap.
And both of them are available precisely because the people making $16.3 billion in six months have absolutely no interest in doing either one.
The Real Point
The model layer is doing extraordinary numbers. The companies buying from it are flat.
That is not a contradiction and it is not a bubble warning. It is a missing layer, and the missing layer is diagnosis followed by deployment, done by someone who understands the specific business.
The labs are never going to do that work. The clouds are never going to do that work. The Big Four are doing it for companies a hundred times the size of your prospects.
Sixteen billion dollars says the ingredient is selling. Zero payoff says nobody is cooking.
Go and cook.
See you next week,
Andrew.
💡 Two ways to start
Launch a local AI agency. We build it for you in 24 hours. Brand, website, offer, demo assets, demo script, proposal and contract, all in your name. You sell the 24/7 AI receptionist & 5-star Google Reviews, and our platform handles delivery.
Launch an AI consultancy. Use your existing industry expertise to run paid AI assessments for businesses. Audits from $5,000 to $50,000, leading into implementation and ongoing retainers.
How did you like today's newsletter? |