Ford v Ferrari

Ferrari to the Grocery Store

I switched my newsletter pipeline off Fable 5 this week. Not because I hit a wall. I wasn't close. I switched because Fable 5, the most capable model money can rent, is an overfit for what I actually do, the way a Ferrari is an overfit for a milk run. Opus 4.8 finishes the same job for a fraction of the freight, and I stopped setting the difference on fire every morning. If a guy who publishes about AI at dawn won't pay for the top token, ask who will.

Now hold that against the tape. PitchBook closed the first half and 86 cents of every US venture dollar, $355.9 billion, went into AI. Most of it chased the frontier: Anthropic at a rumored $1.2 trillion, OpenAI at a number nobody will say out loud. Here is the problem with paying Ferrari prices. The actual Ferrari, the most profitable carmaker alive, with margins nobody in Detroit or Stuttgart can touch, is only worth about $68 billion. One lab is marked at seventeen of them. Sold on the Ferrari story, priced on Toyota's volume.

And Ferrari can't be Toyota, by design.

The whole business is scarcity: build fewer, charge more. The second a frontier model reaches for volume, a Ford pulls alongside. Grok at a tenth the price, DeepSeek for free. Benchmark's Peter Fenton put a clock on it: 90% of tokens running on open-weight models inside eighteen months. So the premium survives only on the laps that truly need the thoroughbred, and it evaporates on all the rest.

Four frontier models launched in a single afternoon this week. GPT-5.6, Grok 4.5, Meta's Muse Spark, and Fable back from its export-control timeout. Grok did a coding job for $1.51 that Fable bills at $17.32. That is not a discount. That is a different car. And the fastest lap belongs to whoever shipped this morning. Six weeks from now it belongs to someone else. And someone else entirely another six weeks out.

Fine, if you are running Silverstone. Fifty-two laps, one afternoon, the quickest car wins. But capital does not run sprints. It runs Le Mans, the five-to-ten-year hold, and nobody has ever won the 24 Hours with the qualifying lap. Ford proved it in 1966 with a car that was less exotic and still turning clean laps at hour 23, while the Ferraris sat in the garage with the hood up.

Eighty-six cents of every dollar bought the fastest car on the grid. The fastest car never wins Le Mans.

At COAI today: the full Signal/Noise — the seventeen-Ferrari math, why OpenAI is the one boxed in on every side, and how SpaceX quietly became the house that wins every hand — is live at getcoai.com.

Outsider Labs Advisory →

86%. Eighty-six cents of every US venture dollar in the first half of 2026 — $355.9 billion — went into AI, most of it into the frontier labs. Here's the tell nobody priced in: the actual Ferrari, the most profitable carmaker on earth, is a $68 billion company, because scarcity is the entire business. One AI lab is marked at seventeen Ferraris. Sold on the Ferrari story, priced on Toyota's volume. You cannot be both, and the market just bought both.

Take the Ferrari off the grocery run. Pull your model bill today and mark every recurring job "race" or "errand." The errands — summaries, formatting, first drafts, tagging — go to the cheap tier or an open-weight model you host yourself. I moved mine off Fable to Opus and never felt it. Reserve frontier tokens for the laps that actually need the thoroughbred. Most shops pay race prices for errands.

Underwrite the Cayenne, not the halo. Allocating capital? The AI names that return it over a decade aren't the fastest labs. They're the ones with a boring volume business under the premium badge: Google's distribution, a real consumer app, the infrastructure everyone rents. A pure frontier bet with nothing scaled beneath it is a $68 billion business wearing a trillion-dollar tag. Buy the car built to finish Le Mans.

Stand up a Ford before you need one. Grok shipped a coding model at a tenth of Fable's cost the same week Anthropic's best model came back from a government timeout. Frontier access is a turnstile now, not a wall. Put an open-weight model you control behind anything you can't afford to have gated, repriced, or benched on someone else's bad morning.

Friday, July 10

Where the tape caught up to us this week.

The tape doesn't lie. We just read it early.

— Harry and Anthony

Sources: