Labs buying from the training-input layer (enter bell)

Sum over training-input vendors of each vendor's best procurement score in the trailing four quarters, at quarter ends. Rubric: 1 a lab named as customer; 2 revenue or growth attributed to lab demand; 3 a disclosed lab contract or commitment of at least $100M. Derived metric lab_procurement_signal_4q.

Reason through this

The reading

One scored vendor so far (Mercor, revenue attributed to lab demand in June 2026); the sum is a floor until more vendors have a quotable source.

Direction over the last 5 readingseach dot links to its record

score

in the rule's windowearliermoves within ±1.00 score read as stable, and a move counts only if more than half the steps go its way; this window moved −2.00 score and 3 of 4 steps fell; higher reads concentrating

estimatereported

8 readings drawn from the derived metric lab_procurement_signal_4q; 5 observations in the evidence.

Derived (lab_procurement_signal_4q) from: CNBC; TechCrunch; Sacra company pages (free summaries); The AI Insider; Dealroom news. CNBC. TechCrunch. Sacra estimates (free company page). The AI Insider. Dealroom.

The numbers

as of	reading	value	record
2026-09-30	—	2.00 score	reported
2026-06-30	—	3.00 score	reported
2026-03-31	—	2.00 score	reported
2025-12-31	—	3.00 score	reported
2025-09-30	—	4.00 score	estimate
2025-06-30	—	4.00 score	estimate
2025-03-31	—	3.00 score	estimate
2024-12-31	—	2.00 score	estimate

Why it matters

P1's enter bell: when the labs start buying from the layer below them, that layer's economics turn; read with the exit bell, when they buy the vendors outright.

What cuts against it

Scores are hand-coded from quoted text; vendors announce selectively and the labs rarely confirm; lab compute commitments are on the circular ledger and excluded here, so the bell reads data and environments, not compute.

How it is measured

series procurement.*.signal.pt
derived metric lab_procurement_signal_4q (formula in the semantic layer)
source CNBC · default tier 5 · CNBC; short quotation
source Dealroom news · default tier 5 · Dealroom; short quotation
source Sacra company pages (free summaries) · default tier 6 · Sacra; short quotation of the free summary
source TechCrunch · default tier 7 · TechCrunch; short quotation
source The AI Insider · default tier 5 · The AI Insider; short quotation
proxy capital_flow, market_structure · cadence quarterly

Four quarters; a one-point move is one vendor's single disclosure. Rising = the labs buying more from the training-input layer.

Applied to metric:lab_procurement_signal_4q.

Derived rows (8)

as of	value	inputs
2026-09-30	2.00 score	source1
2026-06-30	3.00 score	sources1 2
2026-03-31	2.00 score	sources1 2
2025-12-31	3.00 score	sources1 2 3
2025-09-30	4.00 score	sources1 2 3+1 more
2025-06-30	4.00 score	sources1 2 3
2025-03-31	3.00 score	sources1 2
2024-12-31	2.00 score	source1

Latest readings

2026-09-30 · 2.00 scoreas of 2026-09-30
2026-06-30 · 3.00 scoreas of 2026-06-30(2 obs)
2026-03-31 · 2.00 scoreas of 2026-03-31(2 obs)
2025-12-31 · 3.00 scoreas of 2025-12-31(3 obs)
2025-09-30 · 4.00 scoreas of 2025-09-30(4 obs)
2025-06-30 · 4.00 scoreas of 2025-06-30(3 obs)

Why this status, and its history

Held at emerging by its override note: the rubric sum has run 3, 4, 4, 3, 2 and 3 at the last six quarter ends as vendors disclosed lab customers and then went quiet, which tracks disclosure rather than demand; training-input demand reads through the expert-data market run-rate instead.

2026-09-11not yet measurable to emergingconf 30 → 30 · Claude (model)
Held at emerging by its override note: the rubric sum has run 3, 4, 4, 3, 2 and 3 at the last six quarter ends as vendors disclosed lab customers and then went quiet, which tracks disclosure rather than demand; training-input demand reads through the expert-data market run-rate instead.
sources1 2 3 4 5
2026-09-11unmeasured to not yet measurableconf — → 30 · Claude (model)
First reading, quarter to 30 Jun 2026: one training-input vendor scored (Mercor, 2 on the rubric: growth attributed to AI labs buying expert data). The direction rule needs five quarterly readings; not yet measurable until then. Initial seed.
source1

Why this confidence

Rests on 5 observations from CNBC, Dealroom news, Sacra company pages (free summaries), TechCrunch, The AI Insider; best evidence tier 5, grade C; current.

Confidence is independent of status: 90–95 multiple strong independent sources; 70–89 good evidence, some ambiguity; 50–69 mixed or hard to operationalise; 0–49 limited or vague evidence.

Predictions that lean on this

Not happening
Practice tasks with automatic graders, and the expert-written data they consume, are where the shortage goes next. · This site
Too early to tell
Labs buy more of their training data straight from leading app companies and specialists, squeezing the firms that sell general practice environments. · Sean Cai

Reading

emergingA model's judgement, not a reading: probably concentrating. The site's own score points the other way: it has drifted down over the last year and on its rule reads dispersing. The judgement discounts it because the score is a tally of what vendors chose to disclose, which the site describes as a floor, and it falls as old announcements age out even if buying continues. General knowledge, and the site's reading on the largest expert-data vendor, say the labs are buying far more expert data and training environments than a year ago. If that demand has since cooled, the series is right and this call is wrong.

Latest

2.00 scoreas of 2026-09-30

What the rule reads

2.00 scoreas of 2026-09-30

The rule

Direction over 5 readings; moves under 1.00 score count as stable; higher means concentrating.

Confidence 30 of 95: limited or vague evidence. grade C

Leading: labs buy data and environments for models released months later, so purchases precede capability and revenue.

Updated 2026-09-11