Who Do You Want To Be

Upper Right or Nowhere

Higgsfield did a billion dollars in ARR in eighteen months. Faster than anyone not named OpenAI or Anthropic. They spend four million dollars a month on inference, ten thousand per employee every month, and not a dollar of it trains a model. It runs one. They point it at content: studio-grade video, a machine of professional creators feeding the distribution beast. Ten grand a head a month isn't R&D you amortize once. It's rent. And it's buying the one thing everyone else in AI is scrambling for and can't find — a moat.

a16z drew the map this weekend without quite reading it back. Four ways to build something durable: create a new behavior, own distribution, price better, lock people in. They buried two of them in a single paragraph, correctly. You can't out-price a commodity, and you can't lock anyone into a general-purpose model when swapping one for the next is a task-by-task afterthought. Report-writing AI has switching costs of zero. That's most of what shipped this year.

Which leaves distribution — we've said it since April, and it's still true — and the lever underneath it that a16z waved off too fast. Switching costs aren't dead. They're re-earned. The second the behavior you unlock is one only you can do well, the lock-in comes back for free, because there's nothing to switch to. Yet.

Here's the trap most founders walk into: being the best is not the same as being the only. I play tennis. So do Alcaraz and Sinner. Nobody's paying to watch me, and nobody's switching costs are protecting the twelfth-best model at anything. "Best" is a ranking in a crowded field, and the field reshuffles every quarter. "Only" is a category of one. Higgsfield isn't the best at a thing lots of people do. It's the sole road to a thing you couldn't do at all last year.

So the question a16z hands every operator, every investor, every parent reading this, isn't "is my model better." It's the Oingo Boingo one. Who do you want to be? Draw the quadrant. One axis: did you unlock a behavior that didn't exist yesterday, or just do an old chore faster? The other: can only you do it, or can anyone? Upper right — new behavior, and you're the only one — is the whole game. Everywhere else you're renting GPUs at eight bucks an hour to sell a token somebody undercuts tomorrow.

Stop asking if your model is better. Ask what your customer can suddenly do — and whether anyone else can let them do it.

At COAI today: the full Signal/Noise — the four-lever teardown, the turn distribution just took toward being readable by agents instead of seen by people, and the quadrant drawn out with names in every box — is live at getcoai.com.

Outsider Labs Advisory →

$10,000 per employee, per month. That's what Higgsfield burns on inference — not training, running — to make studio-grade video at a scale nobody can match, on its way to a billion in ARR in eighteen months. Read the number as rent, not R&D: it's the monthly cost of being the only one who can do a thing. Commodities don't spend like that, because commodities can't charge like that. Who do you want to be — the toll road, or the traffic?

For the investor — price the moat, not the model. Put an AI company on the quadrant before you look at the multiple. A behavior only they deliver — Higgsfield's video — earns pricing power and real switching costs. "Best-in-class" at a commodity task earns a race to zero: Opus 5.5 runs 40% cheaper than the last one, and Sonnet 5.5 shipped this morning. The model line isn't the asset. Underwrite what the customer can't get anywhere else.

For the business owner — find the behavior you alone unlock. Don't ask which tool makes your team faster at what they already do; that's table stakes your competitor buys the same afternoon. Ask what your customers can suddenly do because of you that they couldn't last year, and whether you're the only door to it. If the honest answer is "anyone could offer this," you don't have a product. You have a feature waiting to get bundled.

For the parent — coach for "only," not "best." Your kid is growing up where every general skill — write the essay, build the spreadsheet, summarize the reading — is free and infinite. Being best in the class at a commodity task is the tennis trap: somewhere, someone is Alcaraz. Push them toward what's theirs — the weird combination, the taste, the question nobody else asks. Uniqueness is the only switching cost a human has left.

Tuesday, September 29

— Harry and Anthony

Sources: