Paper 1 of 5

The model is the pipe

Intelligence is cheap. Context is scarce.

Ranjit Kohli, YYZdata Inc. October 9, 2026, revised October 14, 2026. Paper 1 of 5.

What this paper argues

At any fixed level of capability, the price of a model falls by an order of magnitude a year or more. A new frontier is matched by an open-weight model in three to four months on the latest readings. Two of the three labs post the same price list and the third is close. The frontier itself is still a premium good and the labs will keep spending to hold it; the chip maker underneath them reports a 71% gross margin, and the labs do not publish one. For most of what a business needs, the frontier is not required. That last claim has the least measurement behind it of anything in this paper, and I say so when I reach it. The value moves up, to the layer that turns a business's own records into a trusted account of what happened. Not the data by itself: a restaurant can change its register, and the next vendor reads the same transactions. The scarce thing is access to the records plus the work of reading them against each other and making them mean something a decision can rest on. Section 8 shows four companies that are let in to read records they do not hold, at the same margins as the companies that hold them.

PDF, 17 pages.

About the author

Ranjit Kohli is founder and CEO of YYZdata Inc., whose lead product is Signals, an operational intelligence platform.

The series

Paper 1 of 5: The model is the pipe

Paper 2 of 5: The whole model, every time

Paper 3 of 5: Read first, guess last

Paper 4 of 5: Who answers the question

Paper 5 of 5: I watched the last pipe get overbuilt

© 2026 Ranjit Kohli. Published by YYZdata Inc. This paper may be shared and quoted with attribution.