Cosmic Wire · Enterprise research

Target verticals research plan

Localized LLM token economics for Mark’s four priority verticals — preliminary package for Tuesday, then iterate with Keli.

Deadline Tuesday Owner Mia → Keli / Mark Depth Data centers first Source Target Verticals PDF + sync notes

Scope lock — this week

In Build now

  • Data centers
  • Critical infrastructure & utilities
  • Manufacturing
  • Telecom
  • Token spend today vs localized LLM + occasional frontier calls
  • Public benchmarks + explicit Cosmic open questions
  • Preliminary for Tuesday → iterate

Out Until later

  • Retail (explicitly deferred)
  • Medical, banking, sports & hospitality, law enforcement
  • Full competitor matrix / comparison database
  • Pricing POC (next week with Mark)
  • Claiming Cosmic cost deltas without Jonathan inputs
  • “Done” as final sales truth
Thesis to prove: a localized, customer-business LLM cuts token spend and strengthens data sovereignty vs feeding ops data into frontier models (Palantir-style rented LLM layer) — while Cosmic turns first-party data into actionable intelligence.

Vertical map

From the Cosmic Wire Target Verticals deck. Priority four are Mark’s start list; the rest stay on the map only.

P1
Data centers
P1
Critical infra
P1
Manufacturing
P1
Telecom
Later
Retail
Later
Medical
Later
Banking
Later
Sports / LE

Shared framework

Same eight fields for every vertical so Tuesday decks stay comparable:

  1. Buyer & use cases — who pays; which jobs burn tokens
  2. Data gravity — what stays private vs must hit a frontier model
  3. Token economics baseline — list vs realized spend; note provider subsidy
  4. Localized LLM delta — share that can stay local; residual frontier %
  5. Sovereignty / compliance — why local matters here
  6. Cosmic wedge — actionable intelligence on first-party ops data
  7. Enhance vs replace — incumbents that supply data but not relativity (e.g. Samsara)
  8. Open questions for Cosmic — only Jonathan / product can answer

Label every number:

Verified vendor / docs / filings · Observed public benchmarks · Inferred scenario · Blocked needs Cosmic

Sequence — deep → thin

1. Data centers first Mark meeting

Three buyers, not one market — Cosmic can get paid across the chain:

Construction / GC

RFIs, schedule, supply chain, QA. High doc/RAG volume. McCarthy / Palantir-style comps.

Operators / colo

Health, anomalies, capacity, tickets. Continuous telemetry → local model; rare frontier escalation.

Hyperscalers

Capex/ops at scale. Highest willingness to pay; sovereignty + spend leverage.

Deliverable: one vertical brief + scenario table (e.g. 10k / 100k / 1M tokens-day → all-frontier vs 80–95% local).

2. Critical infrastructure & utilities

Remote grid diagnostics, early anomaly detection, disruption prevention. Overlaps data-center power. Reuse CenterPoint / Chain Reaction public framing as competitor context — not Cosmic claims.

3. Manufacturing

IoT diagnostics, uptime, throughput. High-frequency local inference; rare complex reasoning to frontier.

4. Telecom

Network diagnostics, eSIM, continuous device signal. High event volume → strongest “local first” cost story if usage proxies exist.

Research method

Pass Time Output
A Map ~0.5 day 5–8 primary sources per vertical (pricing pages, filings, industry AI spend). Prefer primary over blogs.
B Model ~0.5 day One spreadsheet tab per vertical: workload mix, tokens/job, $/M tokens (list vs effective), local share, frontier residual, annualized delta + sensitivity.
C Narrative ~0.5 day 1-pager per vertical: sovereignty + token savings + Cosmic relativity. End with blocked items for Jonathan.
D Tuesday pack Wrap 4 briefs (or 1 deep DC + 3 thinner) · shared scenario model · open-questions list · what Mark can say this week vs needs Cosmic proof.
Product questions to carry (do not invent answers): Cosmic BFT real or planned? Own chain vs existing? Do customers buy the ledger or is it plumbing? Which models power engine/dashboard? On-prem / air-gap readiness vs Palantir.

Definition of done — Tuesday

Per Keli: preliminary + iterable, not final. Done means:

Dependencies — do not wait

Need Action now
Verticals list Have it — four locked from Mark via Keli
Mark PowerPoint Proceed with these four; update if email differs
Jonathan tech facts Public pass first; questions in every brief
Cosmic unit costs / margins Note ~80%+ margin target from Keli; don’t invent unit costs
Pricing POC Out of scope this week
Info routing Go through Keli only — not Rob

Suggested calendar

Today
Lock template + data-center three-buyer outline + source list (Pass A start)
Next
Data-center model + first draft brief (Pass B/C)
Following
Utilities + manufacturing thin passes
Before Tue
Telecom thin pass + Tuesday package + open questions for Keli → Jared / Jonathan