Author
You were handed pricing on top of the job you already had. Every week a vendor you buy from or compete with changes a price, and someone asks what it means.
TL;DR: The AI Pricing Observatory is a weekly log of verified pricing moves in AI software, each with a read written by a person on the mechanic under the vendor’s label. Read it for two jobs: suppliers, whose unit is your cost, and competitors, whose moves show the condition they answered. This is how to read the AI pricing observatory each week.
What a verified move is, and the gate it passes
A change on a pricing page and a pricing move are different objects. A move is the log’s noun: a pricing change a vendor made, dated and verified against the vendor’s page. Roadmap teasers and executive quotes are announcements, and announcements never enter it.
Every entry passes the citability gate and is evidenced as a transition: the dated event plus what the vendor’s terms said before it. A page that reads differently this month than last may be a repricing, a rewrite or a correction, and only the before-state says which.
Who writes the read, and what it says
A pricing architect writes each read, and Chris Mele approves each read before it publishes to the log and the weekly email.
The label is what a vendor calls its model; the mechanic is what the invoice turns on. Every read names the mechanic, so you judge a competitor by what is counted and the label stops doing the work.
A read states what changed, what did not, and the sequence, and leaves the vendor’s motive unstated, so the conclusion stays yours and the log stays defensible when the vendor reads it. That is why the log covers a bounded set of vendors: breadth is traded for a person’s read on every entry.
The three decision pages
Every move is filed under the decision that moved: licensing, packaging or pricing. Most are announced in pricing language, so the filing itself is the first read you take from an entry. The filing follows pricing architecture, the three decisions of licensing, packaging and pricing (licensing model, packaging model, pricing model) that determine how a software company captures value, made in the decision layer.
The licensing page files the moves that changed who may use the software and on what unit, so the value metric is decided here. A credit is a surrogate unit, a vendor-defined billing unit that stands in for value the company cannot meter directly; a pass-through keeps the supplier’s unit on your customer’s invoice. Both are filed here. The page’s question: which unit did each vendor move to, and does it measure what the customer values or what the vendor spends?
The packaging page files the moves where what ships together changed. Its question: what ships together, and who decided?
The pricing page files the moves where the rulebook moved: what the units cost, on what rate card, under what rules. Its question: did the rulebook change, or did the unit underneath it change?
The vendor arcs
Each admitted vendor has its own arc page with its moves in order; OpenAI and GitHub are two. A band at the end lists the vendor’s changes our collection saw and did not admit to the log, and it says nothing about the vendor.
Admission is by hand, on four tests: the vendor’s own software prices AI capability, and the move is a step on the arc from selling tokens to selling outcomes that the log traces across its vendors. One move is verifiable to the vendor’s own page, and buyers weigh the vendor, or the move names a mechanism nobody else has used.
Three recent moves, read live
The table renders the three most recently verified moves from the log, each with its teaser line; the full read is on the Observatory. Take three questions into each row: the decision filed under, the mechanic named, and the terms before the move.
The weekly email and the Slack digest
The weekly email, the Observatory Weekly Register, lists the moves the week they are verified: a dated headline and a one-line read for each. The Slack digest is the same summary in a channel you choose, on Mondays. Both are silent in a week with no verified move: a quiet Monday means the Observatory held still, and a habit paced by the digest is paced by verified change in the market.
What the Observatory cannot see
The log records what vendors announce. It cannot see what anyone signed. What vendors announce and what buyers sign are two halves of the market, and the log holds the announced half. Pricing intel holds what buyers signed: a contribution goes into its own store under a self-reported provenance, no contributed figure enters the log, and what comes back is the aggregate. The two halves stay trustworthy for their own purposes because they never mix.
The Observatory shows what vendors announce. Does your pricing survive what buyers sign?
Public announcements cover only the vendor half of the market. Score your signed half: see whether your licensing, packaging, and pricing hold once discounts, carve-outs, and negotiated terms reshape what buyers actually commit to.
How to read the AI Pricing Observatory each week: the two jobs
Suppliers, for cost: their unit is your cost
The model providers’ rate cards set the floor: if your product runs on a supplier’s model, your cost moves when its price moves. If you passed the supplier’s unit through, you adopted a unit chosen upstream, which the licensing page calls a licensing decision made by default.
Each week, read the rate-card and pass-through sections for your suppliers. Ask of each change: if inference spend doubled next quarter, which part of our architecture responds, and who chose our unit? The decision that follows is in pass-through or recast, and the moves are evidence for the three decisions in AI software pricing.
Competitors, for condition: the four-part read
Read a competitor’s move in four parts. Their shape: which decision moved, and whether the label and the mechanic agree. Their condition: the pressure the sequence of moves points to; our condition: whether the same pressure shows in your own won and lost deals. The gap: what remains once those three are named. The Observatory supplies the first two, your deal record the third, and the gap is yours to close.
Competitive analysis for B2B software argues for the choice set buyers weighed. Competitor pricing page analysis argues that the page shows the published list prices and hides the architecture. Pricing benchmarks vs ground truth argues that a benchmark strips the conditions that produced it, so your deal record outranks any median.
What a competitor’s customers paid is learned from those customers, by people, openly, from public and permitted sources, and the Observatory is the public record you read before and alongside that work. The record is the market’s, and the decision is yours: what you do with the same moves depends on your own architecture, and that read does not fit in a Monday message. Talk to an expert, and a pricing architect reads it with you. Next on the Start Here path: continuous monetization.
