"The Twitter/X discourse keeps surfacing this: Replit CEO admitted "we were paying $X per agent run" right before the new pricing landed." - that's the pattern this index is trying to make legible before the receipt drops, not after.- from the methodology doc
Wrapper status
Does it resell someone else's model at a markup? Pass-through compute = thin, subsidy-exposed margins.
Funding stage
Seed / Series A burns runway; Series B+ needs growth to raise. Both pressure pricing.
Observed cadence
How often it already files receipts. Recent + repeat changes forecast the next one.
Parent depth
A deep-pocketed parent (Microsoft, Alphabet, X Corp) absorbs losses a standalone can't.
23 vendors scored 0–100 · higher = more likely to silently downgrade · deterministic, no LLM in the loop · updated as receipts land.
◆Margin Stressed3 vendorsUnit economics force a silent downgrade as subsidies expire - migrate-ready. Score ≥ 50.
01
wrapper · pass-throughearly-stage · Seed
65
02
Cursor
Anysphere · Series B
wrapper · pass-throughgrowth pressure · Series Bprior critical · repeat
56
03
Replit Agent
Replit · Series B
wrapper · pass-throughgrowth pressure · Series Bprior critical · repeat
56
◆On Watch9 vendorsStructural pressure building, no near-term tell yet. Score 25–49.
01
Claude Code
Anthropic · n/a
wrapper · pass-throughrecent receipt · 90dprior critical · repeat
48
02
Bolt.new
StackBlitz · Series B
wrapper · pass-throughgrowth pressure · Series B
48
03
Perplexity
Perplexity · Series C
wrapper · pass-through
42
04
Windsurf
Codeium · Series C
wrapper · pass-through
42
05
wrapper · pass-throughprior critical · repeat
38
06
wrapper · pass-through
38
07
GitHub Copilot
Microsoft · n/a
wrapper · pass-throughrecent receipt · 90downed by Microsoft · cushioned
32
08
Claude (Anthropic)
Anthropic · Series F
first-party model3 receipts · 90d
28
09
Devin
Cognition Labs · Series A
early-stage · Series A
25
◆Stable11 vendorsSafe to depend on for at least the next quarter. Score < 25.
01
first-party modelrecent receipt · 90d
23
02
OpenAI API
OpenAI · private
first-party modelrecent receipt · 90d
23
03
growth pressure · Series B
18
04
Gemini (Google)
Alphabet · n/a
first-party model3 receipts · 90downed by Alphabet · cushioned
17
05
Notion AI
Notion Labs · Series C
first-party model
17
06
first-party modelrecent receipt · 90d
15
07
Midjourney
Midjourney · private (no VC)
prior critical · repeat
13
08
ElevenLabs
ElevenLabs · Series C
12
09
Grok (xAI)
X Corp · Series C
first-party modelowned by X · cushioned
9
10
Microsoft MAI
Microsoft · n/a
owned by Microsoft · cushioned
0
11
Kling AI
Kuaishou Technology · n/a
0
THE CONTRARIAN READ
Past Nerf Index tells you who has nerfed users. Subsidy Risk tells you who will.
Ranking by past behavior selects for vendors who got caught. Subsidy Risk is a forward-looking score using structural inputs - wrapper status, funding stage, runway pressure - that pre-date the receipt by 3-9 months.
Subsidy Risk is forward-looking by design: it scores structural margin pressure - wrapper status, funding stage, runway - that shows up 3-9 months before a critical or major receipt does. Every prediction and its outcome are logged publicly as they resolve, so the score is yours to audit, not ours to assert.
Score is rebuilt nightly from deterministic rules - no model in the loop. View source if you doubt it.
Methodology · v0.4 - 7 deterministic inputs, weighted:
wrapper (×20) ·
funding_stage (×15) ·
critical_receipts_90d (×15) ·
any_receipts_90d (×10) ·
parent_cushion (×−25) ·
prior_critical_history (×10) ·
forecast_pct_above_50 (×15). Capped at [0, 100]. Source:
lib/subsidy-risk.ts ↗