AI Data Protocols

How data moves from sensor to intelligence. Five layers, each with its own protocol patterns.

The Data Pipeline

COLLECT → CONNECT → STORE → COMPUTE → APPLY
   ↓          ↓         ↓        ↓         ↓
 Sensors   Networks   Persist   Train    Deploy
 (DePIN)   (Helium)   (IPFS)   (GPU)   (Models)

Each layer has centralized incumbents and DePIN challengers. The protocols define how data flows between layers.

---

Collection Protocols

How raw data is gathered from the physical world.

DePIN Sensor Deployment

Step	Action	Protocol	Verification
1	Deploy device	Physical installation	Location proof
2	Calibrate	Sensor initialization	Quality attestation
3	Collect	Continuous measurement	Timestamp + signature
4	Transmit	Push to aggregator	Delivery confirmation
5	Reward	Token distribution	Proof of contribution

Collection Categories

Category	What's Collected	DePIN Protocol	Precision
Positioning	RTK corrections	GEODNET	Centimeter
Mapping	Street-level imagery	Hivemapper	Visual
Weather	Temperature, humidity, pressure	WeatherXM	Station-grade
Wireless	Coverage attestations	Helium	Signal strength
Web data	Internet content	Grass	Page-level
Energy	Grid measurements	Daylight Energy	Meter-grade

Proof of Collection

Every data point needs provenance. The protocol pattern:

Device Identity → Timestamp → Location → Measurement → Signature → Chain

Why it matters: Unverified data is commodity. Cryptographically attested data is premium. The attestation layer is where DePIN creates defensible value.

---

Connectivity Protocols

How data moves from collection point to processing.

Protocol	Use Case	Throughput	Range
LoRaWAN	IoT sensors, low bandwidth	Low	Long (km)
Helium 5G	Mobile, high bandwidth	High	Medium
WiFi/CBRS	Dense urban coverage	High	Short
Satellite	Remote, global coverage	Medium	Global

The Three Flows

Same architecture as telecom:

DATA INTENT → ROUTE → INFRASTRUCTURE → SETTLE → FEEDBACK
     ↓           ↓           ↓               ↓         ↓
  Request     Path AI    Network link     Payment    Quality

Flow Stage	Data Implementation	Provider
Intent	Data request (query, stream)	Consumer application
Route	Optimal path selection	Network AI
Infrastructure	Physical connectivity	DePIN operators
Settle	Micropayment for delivery	Blockchain
Feedback	Quality metrics, latency	Protocol oracle

---

Storage Protocols

How data persists for training and retrieval.

Protocol	Model	Best For	Trade-off
Filecoin	Incentivized IPFS	Large datasets, cold storage	Retrieval speed
Arweave	Permanent storage	Immutable records, proofs	Cost per MB
Ceramic	Mutable data streams	User profiles, session data	Complexity
IPFS	Content-addressed	Deduplication, sharing	No incentive layer

Storage Workflow

Raw Data → Preprocess → Deduplicate → Store → Index → Serve
              ↓                          ↓        ↓
         Quality gate              CID/proof   Discovery

The cost curve: Decentralized storage is already price-competitive with AWS S3 for cold storage. Hot storage and retrieval remain centralized advantages.

---

Compute Protocols

How data becomes intelligence through training and inference.

Distributed GPU Networks

Protocol	Focus	GPU Count	Revenue Model
io.net	General GPU compute	500K+	Marketplace fees
Render	Graphics + AI rendering	100K+	Burn-mint equilibrium
Akash	General cloud compute	Growing	Reverse auction
Gensyn	ML training verification	Early	Proof of training

Training Workflow

Dataset → Preprocess → Distribute → Train → Verify → Aggregate → Model
             ↓              ↓          ↓        ↓
        Quality check   GPU selection  Epochs   Proof of training

Inference Workflow

Query → Route → Edge/Cloud → Process → Return → Settle
           ↓                     ↓          ↓
      Latency optimize     Model select   Micropayment

The thesis: Training is batch — price matters most. Inference is real-time — latency matters most. Distributed networks win on price (training) while edge networks win on latency (inference).

---

Application Protocols

How trained models reach users and generate value.

Pattern	Description	Example
API marketplace	Models served via API, pay-per-query	Replicate, Together
Edge deployment	Models run on device, zero latency	On-device inference
Agent protocols	AI agents discover and use models	MCP, A2A
Data marketplace	Raw and processed data traded	Ocean Protocol

The Full Loop

Collection → Storage → Compute → Application
     ↑                                    ↓
     └──── Application generates ────────┘
            new data for collection

This is the VVFL in protocol form. Each layer feeds the next. The loop accelerates with scale.

---

Protocol Economics

Layer	Revenue Capture	Token Mechanism
Collection	Data sale fees	Proof of contribution rewards
Connectivity	Transfer fees	Data credit burn
Storage	Storage fees	Capacity staking
Compute	Processing fees	GPU staking + burn
Application	Query/API fees	Usage-based burn

The integration thesis: Protocols that span multiple layers capture more value. A network that collects AND stores AND computes has structural advantages over single-layer plays.

---

Context

• AI Data Overview — The transformation thesis
• Platform — ABCD stack for each layer
• Players — Who operates at each layer
• Telecom Protocols — Parallel protocol architecture