AI Signal Gauge · standalone page
Helix Compute Partners (HLCX)
Filings-based rating of whether a company’s AI is earning its cost · quarterly · two cohorts
AI Signal Gauge for HLCX
InfrastructureNeutral
Infrastructure cohort · quarterly · as of 10-Q filed 4 Aug 2026
Capped at Neutral. The factors alone read Earning. The veto applies because the useful-life extension supplies more than a quarter of operating income. It lifts when that share falls below 25% or when the company shows older hardware still earning.
Factor checklist
next input: 10-K, Mar 2027- Post-buildout free cash flowOCF − capex − leases − commitments−$120M · improving from −$410M over 8q · cohort median −$260M▲ improvingWhat would change this: Deterioration for two quarters after commitments. Next: 10-Q, Nov 2026.✓Supports
- Depreciation useful lifechange in accounting estimateServers extended 5 → 6 years · +$310M operating income · 38% of total▼ worseningWhat would change this: Disclosure that older generations retain paying workloads at acceptable margin. Veto lifts below a 25% share.✕Against
- Customer-funded revenue>10% customers, related partiesTwo customers above 10% · one is also a disclosed investor→ flatWhat would change this: Concentration below 10% or the related-party commitment expiring. Next: FY26 10-K.✕Against
- Debt & guaranteeson-balance-sheet onlyTake-or-pay commitments $1.9B · maturities matchedWhat would change this: Partial factor. Off-balance-sheet vehicles not scored.–Neutral
- Claim–reality divergenceclaims vs. financial movementModerate · specific claims, mixed delivery→ flatWhat would change this: Two quarters of agreement between claim and cash.–Neutral
- Capacity utilisationnot disclosed at company levelNot measuredNot measured
Reality Bar · AI Infrastructure & Software ETF
illustrative · 100 holdings
- Compounding
- 11
- Earning
- 19
- Neutral
- 23
- Leaking
- 8
- Talking
- 14
- Insufficient disclosure
- 25
Coverage, stated
41 of 60 demo names fully scored · 19 return insufficient disclosure, 17 of them in the application cohort. The number is published on the method page and moves as disclosure improves.
What this rating is not
It does not measure tokens, cost per task or hidden reasoning. It measures what public filings allow: disclosed cost against a provider price index, where AI sits in the business, and whether the claims match the numbers.