In one year, three private companies are going public at a combined $3.6 trillion. SpaceX is running its roadshow now at roughly $1.77 trillion. Anthropic sits at $965 billion after its Series H, a step ahead of OpenAI at $852 billion. Put together that is about the GDP of France, and nearly all of it is one bet: that AI capability is becoming infrastructure. For a marketing operator that is not a stock story. It is a signal about where your edge is about to move, and most people are going to read it backwards.
I am not writing this to tell you which IPO to buy. I am writing it because the same number that has Wall Street arguing about a bubble is quietly telling operators something useful about what to stop spending time on and what to compound instead. There are three layers where that judgment plays out, and the gap between the operators who compound and the ones who spin is decided at each one.
What does a $3.6 trillion year actually signal?
It signals that raw AI capability is finishing its move from moat to utility. The most valuable listings of the year are not consumer apps. They are the companies selling picks and shovels: launch capacity, compute, and frontier models. When the market prices that layer at trillions this fast, it is saying out loud that this is infrastructure everyone will build on, the way it once decided AWS was something you rent rather than rebuild.
Here is the shape of the year, with the figures as they stand.
| Company | Valuation | Timing | What it signals |
|---|---|---|---|
| SpaceX | ~$1.77 trillion | Roadshow now, pricing mid-June | Hard infrastructure is back in fashion at the top of the market |
| Anthropic | $965 billion | Listing expected this fall | The model layer is a trillion-dollar utility, not a feature |
| OpenAI | $852 billion | IPO as early as September | Frontier capability is now priced like a commodity input |
Read those three rows together and the message is not "AI is big." It is "AI capability is becoming a thing you buy, at a price the whole market pays, from companies worth more than most countries." That changes what counts as an advantage.
Why does this change the math for a marketing operator?
Because the input you were tempted to treat as your advantage is becoming a commodity everyone rents at the same price. When the model is a utility, having access to it is not an edge. Your competitor has the same Claude, the same GPT, the same generation quality, for the same per-token rate. The thing the $3.6 trillion is pricing is the floor, and a floor everyone stands on is not a differentiator.
So your edge moves up the stack, off the capability itself and onto three things you actually control: what you build on top of the model, how tightly your tools connect, and the judgment you apply that no model sells. Those are the three layers. The operators who compound work all three. The ones who spin keep trying to win at the one layer the labs just bought for trillions.
Layer one: are you building on the infrastructure or competing with it?
The operators who compound build on top of the labs, and the ones who spin quietly try to out-build them. You are not going to beat a $965 billion company at training a model, and you are not going to beat SpaceX at launch. The waste I see is teams pouring hours into owning capability the market just valued at trillions, when the entire point of infrastructure is that you rent it and put your energy somewhere it can actually pay off.
What building on it actually looks like
Building on it means treating the model as a utility and spending your scarce attention on the work above it: the offer, the funnel, the data, the relationships. Use the frontier model for the reasoning that is genuinely hard, route the mechanical volume to a cheaper tier, and stop treating "we use AI" as a position. Everyone uses AI now. The position is what you do with the layer you are not renting.
Layer two: is your stack connected or just collected?
The compounding advantage is not which tools you own, it is whether they talk to each other. Two operators can run the identical tool list and get completely different returns, because one wired them together and the other left them in silos. When the model layer is a commodity, the seams between your systems are where the real productivity now lives, and seams only exist where tools expose a real interface.
The silo tax, again
Every tool that cannot be reached programmatically charges you a tax you pay in manual work, every week, forever. It does not show on an invoice. It shows in the hour you lose moving data by hand from the tool that will not connect to the one that will. The connected stack compounds because each integration makes the next one more valuable. The collected stack just accumulates, and the tax grows with it.
If you are deciding where to put the next dollar, this is the order that compounds:
- Pick a system of record you own. One place every customer and every interaction lands, that you control and can query, not a marketplace that holds the relationship for you.
- Demand a real interface from every new tool. API, webhook, or MCP. If a tool cannot pass structured data downstream, it is a silo with a nice UI, and it taxes you forever.
- Wire the seams before you add features. Connecting the four tools you already have usually beats buying a fifth.
- Put the model at the center, not at the edge. The AI layer is most valuable when it can read and write across the connected stack, not when it lives in one app's sidebar.
Layer three: what can the $3.6 trillion not buy?
Judgment, taste, relationships, and an audience you own are not on any of these balance sheets. When everyone is building on the same models with the same connected tools, the only durable differentiator left is the human layer: knowing which campaign is worth running, which client to say no to, what your market actually wants under what it says it wants. The labs sell capability. They do not sell discernment, and discernment is the part that has gotten scarcer as capability got cheaper.
The $3.6 trillion buys you the same model your competitor has. It does not buy you the judgment to use it, the connected stack to act on it, or the audience to sell it to. That is the part you still have to own.
This is also why an owned audience matters more, not less, in an AI-saturated market. When generation is a commodity, attention is the scarce asset, and an audience you own outright is the one thing no lab can underprice you on. Rent the reach from the platforms, own the relationship in a system you control.
How do I position for this without chasing it?
You position by moving up the stack, not by buying more AI. The reflex when a number like $3.6 trillion lands is to feel behind and go shopping for tools. That is the spin. The compounding move is the opposite: spend less on capability you can rent and more on the layers the labs cannot sell you. Start here:
- Stop trying to own the model layer. Build on it. The capability is priced, settled, and not your moat.
- Count your silos. Every tool that will not connect is a standing tax. Replace the worst offender with something equally capable and actually connectable.
- Consolidate to one system of record you control, and route everything into it.
- Put your scarce hours on the human layer: offer, judgment, relationships, and an audience you own.
- Treat "we use AI" as table stakes, never as the pitch. The pitch is what you do on the layers above it.
The operators who come out of this ahead will not be the ones who spent the most on AI. They will be the ones who understood that when capability becomes infrastructure, the advantage moves to everything infrastructure cannot do for you.
Common questions
Should I buy the SpaceX, OpenAI, or Anthropic IPO?
That is a question for you and a financial advisor, and nothing here is investment advice. The operational takeaway is independent of the share price: whether or not you buy a single share, the signal that AI capability is now priced as infrastructure should change how you spend your time and your tool budget.
Does this mean AI is going to replace marketers?
It replaces the commodity parts of the work, not the judgment. The generation, the formatting, the first draft, those get cheaper and faster for everyone. What grows in value is the layer the models do not touch: deciding what to make, who to make it for, and which opportunities are worth your attention.
What if the AI bubble pops?
The infrastructure thesis holds for an operator either way. If valuations correct, the capability is still a commodity you build on. Do not time your stack to a market call. Build connected, own your audience, and apply judgment, and you are positioned whether the multiples hold or not.
What is the single move to make this week?
Pick the one tool in your stack that will not connect to anything and start replacing it. That single change attacks the silo tax, moves you up the stack, and compounds with every integration you add after it.
