What the World Wants to Serve

Twelve categories of static datasets that Safecloud can host, encrypt, and monetize through OCP delegation — no API keys, no user accounts, no rate-limit anxiety.

How Safecloud Serves Datasets

1

Pack

Provider runs safecloud-pack to create a .safecloud archive with HKDF subtree keys

2

Host

Archive lives on jets — or on a CDN as a static file with range-request support

3

Delegate

Provider creates OCP delegations: per-domain, rate-limited, geo-restricted, revocable

4

Serve

App's domain gets chunks via Origin header — no API key, no user signup, no SDK

The archive's cleartext tree descriptor means clients can browse what's available without decrypting. subtreeRange() enables HTTP range requests — a client fetches one byte range for exactly the subtree it needs. Encryption is per-subtree via HKDF, so tiered access (free tier gets country-level, paid tier gets city-level) is a key-management decision, not a server-side filter.

🗺️
Map Tiles$17B geospatial market

z/x/yrange

Vector tiles (MVT/PBF), raster tiles (PNG/JPEG), terrain tiles (Mapbox Terrain RGB, Cesium quantized mesh). The OpenStreetMap planet is ~80GB as vector tiles — 300 million files that OpenFreeMap serves off a Btrfs image with nginx. A Safecloud archive turns those 300M files into one packed file where "all tiles for Europe at zoom 14" is a single contiguous byte range. Encryption and delegation add what OpenFreeMap can't offer: paid tiers, geographic licensing, and access revocation.

Satellite and aerial imagery follows the same z/x/y pattern. Cloud-Optimized GeoTIFFs (COG), WMTS tile pyramids, Sentinel-2 and Landsat scenes. The Earth observation data market alone is $5B+.

Who pays today: Mapbox ($150–$1500/mo), Google Maps Platform ($7/1000 loads), HERE ($100–$900/mo), Maptiler ($25–$490/mo). All API-key-gated with per-request billing.

🏗️
3D Terrain & Buildings$1.8B 3D mapping market

lod/x/yrange

Cesium 3D Tiles (b3dm, i3dm, pnts), CityGML, Mapbox terrain-DEM. Used for urban planning, flight simulators, autonomous driving simulation, and Google Earth-style globes. The tree-structured LOD (level of detail) format maps directly to a Safecloud archive tree — coarse geometry at the root, detail tiles deeper. A city government packs its building models once; architecture firms buy delegations to stream them into their tools.

Who pays today: Cesium ion ($150–$1000/mo), Google 3D Tiles API (per-request), Nearmap ($5K+/yr for 3D city models).

⛓️
Blockchain & Crypto Data$2.3B analytics market

chain/blockrange

Full node chain data (Ethereum ~1TB, Bitcoin ~600GB), transaction receipts indexed by contract/topic/block range, state snapshots (Erigon .seg files), UTXO sets. The data is public and append-only — the block hash IS the CID. A Safecloud archive organized as chain/contract/topic/blockRange lets a DeFi app fetch just the USDC Transfer events for 2024 without scanning the full chain. The crypto analytics market is $2.3B because indexing and serving this public data is hard. Safecloud makes it a hosting problem.

Historical price data: OHLCV candles for every token pair. Binance alone has thousands of pairs at 1-minute granularity going back years. Tree: exchange/pair/year/month.parquet. Order book snapshots, MEV/mempool data, and wallet labels (Arkham/Nansen-style) are even more commercially valuable.

Who pays today: Chainalysis ($200M+ ARR), Dune ($350–$2500/mo), The Graph (per-query GRT), Alchemy ($50–$400/mo), Kaiko market data ($1000+/mo), Nansen ($150–$2500/mo).

🧠
AI Model Weights & Training Data$9.6B training data + $5B+ model hosting

model/layerrange

Safetensor shards, GGUF quantizations, LoRA adapters, embedding models. This is Infrastructure's original use case for the archive format: a model provider packs weights into a .safecloud archive with HKDF subtree keys. Tier 1 gets the full-precision key. Tier 2 gets the quantized subtree key. Tier 3 gets individual expert layers. One archive, tiered access. The AI training dataset market is $4.5B in 2025 heading to $9.6B by 2029.

Training datasets (ImageNet, LAION, Common Crawl subsets, The Stack, RedPajama) are massive static corpora that researchers and companies need. Licensing restrictions make delegation a natural fit — the data owner delegates access to approved institutions with geographic and rate constraints.

Who pays today: HuggingFace Pro ($9–$20/mo per user), Replicate ($0.0001/sec inference), AWS/GCP model hosting ($100s–$1000s/mo), Scale AI ($millions/yr for labeled data).

📚
Encyclopedias & Knowledge Bases$5B+ reference data market

lang/articlestream

Wikipedia dumps (~22GB compressed for English), Wiktionary, Wikidata (~100GB structured facts), Wikimedia Commons. Apps like Kiwix already package Wikipedia for offline use as ZIM files. A Safecloud archive lets you serve individual articles by path without unpacking the whole dump, and gate access by language subtree. A mobile app fetches articles on demand; an AI pipeline fetches the full English corpus as one subtree range.

Also: Legal databases (CourtListener — millions of case opinions), patent databases (USPTO full text), drug references (DrugBank, FDA NDC), ICD-10/ICD-11 code sets, PubMed abstracts (35M+ records). All structured, versioned, commercially licensed.

Who pays today: LexisNexis/Westlaw ($100–$500/mo per seat), UpToDate ($500+/yr), Elsevier ScienceDirect ($10K+/yr institutional).

🌤️
Weather & Climate$3.4B weather data services

time/regionrange

NOAA GFS forecasts, ERA5 reanalysis (petabytes of hourly global weather since 1940), radar composites, GOES satellite imagery. Each timestep is a tile pyramid. Historical weather data is a time-series archive where each day/hour is a subtree — a climate researcher fetches 1950–1970 European winters as one range request. The weather data services market is $3.4B and growing.

Who pays today: Tomorrow.io ($100–$2000/mo), OpenWeatherMap ($40–$2000/mo), DTN ($expensive, enterprise), Climacell/Tomorrow (acquired).

🧬
Genomics & Bioinformatics$22B bioinformatics, $4B genomic platforms

genome/chromrange

Reference genomes (GRCh38 — 3.1 billion base pairs), variant databases (gnomAD — 76K genomes), protein structures (AlphaFold DB — 200M+ predicted structures), gene expression atlases. The bioinformatics market is $22B heading to $43B. These datasets are enormous, structured by chromosome/gene, and access-controlled by institution. A Safecloud archive per reference genome, with HKDF subtree keys per chromosome, lets a hospital's bioinformatics pipeline fetch just chromosome 17 (BRCA genes) without downloading the full reference.

Who pays today: Illumina BaseSpace ($100–$10K/mo), DNAnexus ($enterprise), Seven Bridges ($enterprise), NCBI SRA (free but slow).

📦
Package Registries & Container Images$1.7B container registry + $B+ package hosting

name/versionstream

npm (~3M packages), PyPI, crates.io, Maven Central, Homebrew bottles, Linux distro mirrors (apt, yum). Every package version is an immutable tarball — content-addressed by nature. A Safecloud archive of the npm registry is a CDN-ready offline mirror that any CI system can pull from. OCI container image layers are the same pattern — content-addressed blobs that a registry serves by digest. The container registry market is $1.7B.

Who pays today: Docker Hub ($7–$24/mo per seat), GitHub Packages (included in $4–$21/mo), JFrog Artifactory ($98–$750/mo), AWS ECR ($0.10/GB/mo).

📈
Financial Market Data$50B+ market data industry

exchange/symbol/daterange

Historical equities (OHLCV for every stock on every exchange), options chains, futures, FX tick data, SEC filings (EDGAR full text), earnings transcripts. The financial market data industry tops $50B — and most of that is selling access to static historical records. Order book reconstructions, trade-and-quote (TAQ) data, and alternative data (satellite imagery of parking lots, credit card transaction panels) are the premium tier. The alternative data market alone is $15B+.

Who pays today: Bloomberg Terminal ($24K/yr), Refinitiv ($22K/yr), Quandl/Nasdaq ($500–$50K/yr), Polygon.io ($30–$500/mo), Alpha Vantage (free–$50/mo).

🎬
Media Libraries$43B CDN market (serving media)

hls/dashrange

HLS segments (.ts), DASH segments (.m4s) — already what Safecloud does for video streaming. But also: royalty-free music libraries (Epidemic Sound, Artlist), podcast archives, audiobook catalogs, stock photo/video libraries, font collections (Google Fonts, Noto). All immutable, all served by path or ID, all commercially licensed. The CDN market is $43B and most of that bandwidth is media delivery.

Delegation maps perfectly to media licensing: a publisher delegates "10,000 streams/month from these 500 tracks" to a podcast app's domain. The app never handles license keys.

Who pays today: Epidemic Sound ($12–$500/mo), Shutterstock ($29–$199/mo), Getty ($175–$500/mo), Cloudflare Stream ($1/1000 min stored + $5/1000 min delivered).

🔬
Scientific Datasets$3B+ research data market

dataset/variable/slicerange

Zarr arrays (chunked N-dimensional arrays for climate, oceanography, materials science), HDF5 archives, astronomy catalogs (Gaia — 1.8 billion stars, SDSS — 4.6M spectra), particle physics (CERN Open Data — petabytes). Cloud-native formats like Zarr are designed for partial reads — each chunk is a file at a known path. A Safecloud archive of a Zarr dataset lets researchers fetch exactly the spatial/temporal slice they need via subtreeRange().

Who pays today: AWS Open Data (free egress for listed datasets, but $0.09/GB otherwise), Google Earth Engine ($enterprise), CMIP6 data access (free but slow from ESGF nodes).

🛡️
Threat Intelligence & Blocklists$17B threat intel market

feed/datestream

Malware signature databases, IP/domain blocklists, CVE databases, STIX/TAXII indicator feeds, phishing URL lists. Updated daily, versioned, consumed by millions of endpoints. EasyList and uBlock Origin filter lists alone serve hundreds of millions of users. Commercial threat feeds (CrowdStrike, Recorded Future, VirusTotal) charge per-seat or per-query. A Safecloud archive per feed version, with delegation-gated access, replaces the API-key model. The security product's domain gets the daily update via delegation; its deployed agents fetch indicators by path.

Who pays today: CrowdStrike Falcon ($8.99+/endpoint/mo), Recorded Future ($10K+/yr), VirusTotal ($600/mo), AlienVault OTX (free community tier).

Summary

CategoryMarket SizeAccess PatternTypical Archive SizeSafecloud Advantage
Map Tiles$17Bz/x/y coordinate80GB (OSM planet)Range requests, geo licensing
3D Terrain$1.8BLOD tree10–500GB per cityLOD = archive tree depth
Blockchain$2.3Bchain/block range1TB (Ethereum)Block hash = CID
AI Models$9.6Bmodel/layer path1–400GB per modelHKDF tiered access
Encyclopedias$5B+lang/article path22GB (Wikipedia EN)Per-article fetch
Weather$3.4Btime/region gridPB-scale (ERA5)Temporal subtree ranges
Genomics$22Bgenome/chromosome3–100GB per genomeInstitutional delegation
Packages$1.7B+name/versionTB-scale (npm)Offline CI mirrors
Financial$50B+symbol/dateTB-scale per exchangeReplaces API-key model
Media$43B CDNHLS/DASH segmentsGB–TB per libraryAlready Safecloud's core
Scientific$3B+variable/slicePB-scale (CMIP6)Zarr chunk = Safecloud chunk
Threat Intel$17Bfeed/date pathGB per daily snapshotMass endpoint distribution

Total addressable market across these twelve categories: north of $170 billion. Not all of it is accessible to Safecloud — much of it is value-added analytics, not raw data hosting. But the hosting and delivery layer alone, where providers charge for access to static datasets through API keys and per-request billing, is tens of billions. Safecloud's delegation model replaces the API-key infrastructure with a cryptographic grant — same access control, no server-side session state, no per-request metering overhead.