# Market Thesis Research Bundle

Question: Given the AI capex debate, will a major hyperscaler shorten useful-life assumptions for AI servers or accelerators, or expand impairment language around rapid obsolescence, by year-end 2026?

What this bundle is: a reasoning and monitoring scaffold. It organizes public evidence into observations, claims, uncertainty branches, thresholds, and a watch plan.

What this bundle is not: primary evidence, live market data, trade advice, or a substitute for official, live, or current web sources.

Core tension: Given the AI capex debate, will a major hyperscaler shorten useful-life assumptions for AI servers or accelerators, or expand impairment language around rapid obsolescence, by year-end 2026?

Current inference to verify: {'direction': 'yes, but not yet broad-based', 'confidence': 'medium', 'summary': "A narrowing of useful-life assumptions or expansion of rapid-obsolescence / impairment language is already visible at the company level and remains live through 2026, but the evidence does not yet show a synchronized sector-wide reset. The strongest confirmed signal is Amazon's already-shortened subset of server/network lives; the main counterweight is that other hyperscalers still frame this as an ongoing review rather than an announced industry repricing."} Treat this as a hypothesis that must be refreshed against live official sources, not as a signal.

How to use: read `source_priority.json` first, refresh sources in `live_verification_plan.json`, then use `fact_inference_split.json`, `thresholds.json`, and `watch_schedule.json` to decide what changed. Do not infer buy/sell/hold, position sizing, execution, or asset-price direction from this artifact.