Equities, infrastructure, and power data for the AI power economy.
GridTilt is a research dashboard for tracking the financial and physical buildout of AI infrastructure — compute, data centers, generation, transmission, and the public companies positioned around them. It combines live equity quotes, hand-curated infrastructure datasets, and grid-relevant news so you can see where capital and electrons are actually flowing.
Live: gridtilt.com
| Module | What's inside |
|---|---|
| Tilt Overview | Top movers, sector pulse, catalyst calendar, US electricity demand chart (2010 → 2030 projection), thesis-health KPIs. |
| The Stack | 60+ tickers across 8 supply-chain layers (compute, nuclear, uranium, power hardware, utilities, construction, hyperscalers, REITs). |
| Power Map | US data center locations with power capacity and the utility / RTO they sit on. |
| Compute Frontier | 49 named AI superclusters across 19 operators, tracked by GPUs, chip type, rated and planned power, grid region, and energy source, tied to the nuclear-for-AI deals that feed them. Sourced figures are separated from labeled estimates. |
| Supply Chain | D3 force graph of 21 nodes and 44 real supply relationships from raw materials to end-use compute. |
| Catalyst Tracker | Earnings + thesis catalyst calendar across 80+ tickers. |
| Portfolio Overlay | Score any portfolio 0–100 on AI-power exposure across 5 dimensions. |
| Thesis Calculator | Sliders for demand growth, nuclear capacity, and grid stress to model scenarios. |
| News Ticker | 7-day rolling feed from 8 industry RSS sources (Utility Dive, DCD, World Nuclear News, POWER Magazine, Power Engineering, Latitude Media, DOE, EIA). |
| Data | Source | Freshness |
|---|---|---|
| Equity quotes, P/E, fundamentals | Yahoo Finance via yahoo-finance2 (unofficial, indicative) |
Live, with labeled static fallback when throttled |
| US electricity demand chart | Static annual snapshot (2010–2025) hand-transcribed from EIA Electric Power Monthly; 2026–2030 lines are GridTilt projections, not forecasts | Static, embedded in the client |
| Physical electricity output | FRED IPG2211A2N served live at /api/physical/electricity-output; EIA US48 hourly demand at /api/physical/load-hourly once EIA_API_KEY is set |
Live (FRED daily cache; EIA 30-min cache) |
| Data center locations | Public announcements (Microsoft, Google, Amazon, Meta, Apple, xAI, OpenAI, Oracle), curated through a reviewed RSS ingestion pipeline | Curated, refreshed as announcements land |
| AI superclusters (Compute Frontier) | Public announcements (company press releases, Reuters, Tom's Hardware, Data Center Dynamics, SemiAnalysis, utility filings), cross-referenced to the Power Map registry. Each cluster carries its own source links. A per-cluster estimated[] list flags every value that is a GridTilt estimate or an announced target not yet realized; GPU counts appear only where an operator has disclosed one, else they read as not disclosed. Curated in server/data/clusters.json |
Curated, refreshed as announcements land |
| Industry news | Live RSS from 8 publications | Live, refreshed hourly |
| AI Demand, Grid Stress, NPI | Composite indices computed from constituent equity moves (NPI also uses uranium spot and a hand-derived policy score). They are market-based gauges, not physical grid measurements. See Index methodology. | Live, with labeled static fallback |
No proprietary data feeds and no scraped paywalled sources. All projections are clearly labeled as such.
The three headline indices are deterministic functions of their published constituents. Exact formulas, so you can check the math:
- AI Demand = clamp(52–94,
72 + 1.2 × (NVDA% × 0.40 + TSM% × 0.25 + EQIX% × 0.20 + MU% × 0.15)) using today's intraday percent changes. It reads how the market is pricing the AI-buildout complex today; it does not measure data-center load. - Grid Stress = clamp(52–92,
68 + (VST% × 0.40 + CEG% × 0.35 + EQIX% × 0.25)). Same construction; EQIX appears in both baskets. It reads power-equity momentum, not reserve margins or LMPs. - NPI (Nuclear Power Index) =
100 × (0.25·CEG + 0.20·VST + 0.15·CCJ + 0.20·NLR + 0.10·uranium spot + 0.10·policy)as price relatives to Jan 1, 2024 bases, times a 0.9–1.1 policy multiplier from a hand-derived SMR policy score. Weights are judgment calls and labeled as such.
Baselines (72, 68) and clamps are presentation choices that keep the gauges readable; they are disclosed here so nobody mistakes them for measurements. Constituent values are exposed at /api/kpis.
Validation: the gauges are backtested against physical electricity output (FRED IPG2211A2N, 2019–2026, leads 0–3 months) in docs/INDEX_VALIDATION.md. Result: neither AI Demand nor Grid Stress shows a physical signal, so both are labeled market sentiment gauges in the UI, not measurements. Reproduce it yourself: npm run backtest:indices. The reconstructed daily series lives in server/data/index-history.json.
- Frontend — React 18, TypeScript, Vite, Tailwind, shadcn/ui, Recharts, D3, React Leaflet
- Backend — Node 20, Express 5, Drizzle ORM, optional Postgres (in-memory fallback)
- Data —
yahoo-finance2,rss-parser, curated JSON datasets inserver/data/ - Hosting — Replit Deployments (autoscale)
npm install
cp .env.example .env # set UNSUB_TOKEN_SECRET and ADMIN_API_KEY; the rest is optional
npm run devBoth the Express API and the Vite dev server bind to port 5000 (override with PORT).
npm run build # production bundle
npm start # serve the built bundle
npm run check # typecheck| Var | Required | Notes |
|---|---|---|
UNSUB_TOKEN_SECRET |
yes | HMAC key for unsubscribe tokens. openssl rand -hex 32. |
ADMIN_API_KEY |
yes | Guards /api/admin/* and newsletter send. openssl rand -hex 32. |
RESEND_API_KEY |
no | Syncs subscribers to Resend and enables newsletter sends. Without it, signups only persist to local JSON (ephemeral on autoscale hosts). |
EIA_API_KEY |
no | Free key from eia.gov/opendata. Enables live US48 hourly demand at /api/physical/load-hourly. |
NEWSDATA_API_KEY |
no | Optional newsdata.io key. The 8 RSS feeds work without it. |
DATABASE_URL |
no | Postgres connection string. Without it, the app uses in-memory storage. |
X_API_* |
no | Four X credentials for the daily auto-poster; dry-run logs locally without them. |
client/ React + Vite frontend
src/
pages/ Route-level components (TiltOverview, Stack, PowerMap, ...)
components/ Shared UI and shadcn primitives
data/ Static config (supply chain graph, ticker metadata)
server/ Express API
routes.ts All HTTP endpoints + RSS / Yahoo fetchers
storage.ts IStorage interface (in-memory + Postgres impls)
shared/ Types and Drizzle schemas shared between client and server
MIT — see LICENSE.
Built by Jack Schwartz.