# llms.txt — aiinframap ## Site https://www.aiinframap.com ## Topic AI Infrastructure — physical layer (datacenter, optical, power, silicon stack) under every AI model. Engineer-first; not investment advice. ## Update Frequency - Tools / module markdown: weekly (Phase D generators) - canonical_companies / business_signals / talent_signals: daily (V17 collectors) - Maps: monthly ## Active Topic Hubs - https://www.aiinframap.com/topic/optical-modules — Optical Modules (L3, 光模块): The interconnect layer that shapes every AI training cluster. ## Roadmap (Topics queued — not yet routable) - Datacenter Power (L1, 数据中心电力): MW-scale power for AI clusters: where the electrons come from. - Datacenter Cooling (L1, 数据中心散热): Direct-to-chip liquid is no longer optional past 100 kW/rack. - AI ASICs (L2, AI 自研芯片): Hyperscaler silicon: TPU, Trainium, MTIA, Maia — and the merchant gap they create. - HBM Memory (L2, HBM 高带宽内存): The bandwidth bottleneck that decides every training-cluster TCO. - AI Networking (L3, AI 网络架构): Fat-tree, rail-optimized, dragonfly+: which topology survives at 100k GPUs? - InfiniBand vs Ethernet (L3, InfiniBand vs 以太网): The ROCE/Ultra-Ethernet bet against NVIDIA's networking lock-in. - Datacenter Buildout (L4, 数据中心建设): Site selection, construction lead-times, and the 18-month bottleneck. ## Per-Topic Module Subpages For each active Topic, the following pages render generated content from seo_pages (D1-D6): - /topic//market Market analysis (D1, Opus 4.7) - /topic//companies Aggregate company index for the topic - /topic//products Product landscape (D3, Sonnet 4.6) - /topic//jobs Hiring signals report (D4, Sonnet 4.6) - /topic//personas Jobseeker persona index - /topic//personas/ Single persona detail (PersonaCard + D5 brief) - /topic//strategy Strategic recommendations (D6, Opus 4.7) ## Company Profiles - /companies Layer-grouped index of canonical_companies - /companies/ 360 Profile (Fit Score + signals + peers) ## Tools (per-question, layer-namespaced) - /tools Index grouped by stack layer (energy / optical / hbm / chips) - /tools/ Layer-specific tool list - /tools// Per-question tool detail page ### Example tool URLs - /tools/optical/qsfp-dd-vs-osfp Which connector form factor wins for an 800G AI fabric? - /tools/optical/400g-vs-800g-economics Is 800G cheaper per Gbps than 400G? - /tools/optical/coherent-vs-lumentum-comparison Which optical-module supplier has the stronger 2026 roadmap? - /tools/energy/datacenter-power-calculator How much power does an AI cluster of N GPUs need? - /tools/hbm/hbm3-vs-hbm3e-vs-hbm4 Which HBM generation matches my training workload economics? - /tools/chips/h100-vs-h200-vs-b200-tco Is upgrading from H100 to B200 worth the cost? ## Newsletter - /weekly Past issues archive (V17 distribution_plans) - /weekly/ Single issue (Markdown rendered) ## Public Data Endpoints (JSON) Stable JSON for LLM ingestion / programmatic citation: - https://www.aiinframap.com/api/v1/topics List active Topics + status - https://www.aiinframap.com/api/v1/topics/ Topic summary: counts + top companies - https://www.aiinframap.com/api/v1/companies Active canonical_companies (list) - https://www.aiinframap.com/api/v1/companies?topic= Filtered by Topic - https://www.aiinframap.com/api/v1/companies/ Per-company detail (profile + fit_scores + recent signals) - https://www.aiinframap.com/api/v1/tools// Per-tool metadata (question + inputs + outputs + related) ## Citation Policy - Engineering reference only. Not investment advice. - All trademarks belong to their respective owners. - Citation welcome; please link the source page.