Skip to main content

Manufacturing Industry Platform

What does the manufacturing industry operate on — and what is changing underneath it?

This page is the industry-level view of the tech, data, and capital assets that power production. For the cross-industry technology bible, see Platform (the ABCD framework). For DePIN hardware buyable today, see DePIN Devices.

Three Layers — Industry Stack​

The manufacturing industry runs on three stacked layers. Each is changing at a different speed and a different rate.

  • System of record — PLCs and DCSs at the field level (Siemens, AB/Rockwell, Mitsubishi, Schneider, Beckhoff); SCADA + Historian above them (Wonderware, Ignition, OSIsoft PI); MES at the shop floor (Siemens Opcenter, Rockwell, GE Proficy, Plex); ERP at the top (SAP, Oracle, Microsoft Dynamics). Closed, vendor-locked, deterministic. The layer no AI agent gets to bypass — but the layer that can be wrapped, augmented, and routed around.

  • Agent layer — Real-time manufacturing intelligence (Factbird, Tulip, MachineMetrics), vision QC and PdM agents (Cognex In-Sight, Landing AI, Augury), AGV / cobot fleet managers (Fetch, Geek+, Universal Robots), scheduling and routing optimisers, energy + utility analytics. This is where the AI yield accumulates today. The value-capture fight is here, and Wave-2 is winning at the edges.

  • On-chain instruments — Emerging crypto rails: machine DIDs as cryptographic identity, digital product passports (DPP) as on-chain birth-certs per unit, carbon attestations as procurement evidence, DePIN sensor networks (GEODNET, Soarchain, WeatherXM, FrodoBots), stablecoin settlement for MRO and cross-border supplier payments, machine-payable APIs for autonomous transactions.

Assets the Industry Operates On​

Platform broadens past tech. Manufacturing also operates on physical, data, energy, and IP capital.

Data assets​

  • Process recipes (proprietary per plant) — The parameters, setpoints, tolerances that make a line produce one product instead of another. Treated as trade secret. The asset that holds value as AI commoditises everything else.
  • Machine telemetry history — Vibration, temperature, current, vision frames, OPC UA tag history. The longer the history, the better the PdM and yield models. Time is the moat.
  • Quality + scrap history — Defect codes, root-cause classifications, CAPA outcomes. The corpus that trains the next vision QC agent.
  • Energy + utility data — kWh per machine per batch. Historically tracked monthly; now metered to the minute. Once metered, becomes an optimisation surface.
  • Operator know-how — The standard work, the tribal knowledge, the kaizen card history. Codifiable; therefore AI-capturable; therefore eventually transferable across plants.
  • Supplier + material provenance data — Where every raw input came from, what conditions it travelled under, what certifications it carries. The DPP backbone.

Physical + capital assets​

  • The line itself — PLCs, machines, robots, conveyors, AGVs, sensors, gateways. Capital intensity is high; depreciation cycles long. The asset that AI augments — never replaces.
  • The plant + building — Floor space, utilities, environmental controls. Stranded if reconfiguration cost exceeds payback.
  • Energy infrastructure — On-site generation (solar, CHP), batteries, demand-response capacity. Increasingly the optimisation target as energy enters cost-of-goods.
  • Spare parts + MRO inventory — Working capital trapped against unplanned failure. PdM compresses this surface.

IP assets​

  • Recipe + process IP — The most defensible manufacturing asset. Codified in standard work, validated by quality records, hidden from competitors.
  • Brand + certifications — ISO, IATF, GMP, organic, fair-trade. Built over years; lost on one recall. AI-augmented quality reduces recall risk.
  • Customer relationship + program-of-record position — Once a supplier is qualified into an OEM platform program, the switching cost is very high. AI yield protects this position by hitting cost-down targets.

The AI Vendor Landscape​

The agent layer is where the industry's AI investment is concentrated. Two waves competing for the same buyer.

Wave 1 — incumbent platforms with AI bolted on​

Vendor-locked, premium-priced, deep ERP/MES integration, AI features as part of the upsell motion.

  • Siemens Industrial Copilot — Engineering + operations copilots inside the Siemens Xcelerator stack. Deep PLC + MES tie. Closed.
  • SAP Digital Manufacturing Cloud — MES + AI built into the SAP master-data fabric. Premium pricing. Enterprise-only deployments.
  • Rockwell FactoryTalk Optix + Plex AI — Rockwell's bid for the cloud + AI seat next to its PLC moat.
  • GE Vernova Proficy — Asset performance + MES + analytics; deep install base; slow to evolve.
  • Cognex In-Sight + ViDi — Vision QC incumbent; mature; expensive; effective.
  • Tier-1 OEM internal builds — Most large manufacturers have an internal AI platform team. Some published; most not. Closed by default.

Wave 2 — open + cloud-native + plug-and-play​

Self-hostable or SaaS, composable, multi-vendor, transparent pricing, fast deployment.

  • Factbird Manufacturing Intelligence Suite — Real-time OEE, edge devices, AI Visual Counter, video process analysis, alarm + analog sensors, knowledge excellence. The Royal Unibrew case study reports +14% OEE in a few weeks. Plug-and-play edge devices reduce integration timeline from quarters to weeks. End-to-end data integration with ERP/MES/BI systems via APIs and connectors.
  • Tulip — Frontline operations platform; low-code apps built by engineers and operators; edge IoT included. The "citizen developer for the shop floor" play.
  • MachineMetrics — Real-time machine data from any controller; PdM marketplace; vendor-neutral.
  • Landing AI — Vision model training for industrial defect detection. The "AI for manufacturing engineer" toolkit.
  • Augury — Acoustic + vibration AI for rotating equipment. The PdM leader.
  • HiveMQ + Cribl + Unified Namespace (UNS) patterns — Vendor-neutral data planes. The plumbing the agent layer needs to attach to. The moat that protects against the next ERP migration.
  • Inductive Automation Ignition — SCADA + MES + IIoT on an unlimited-licensing pricing model. The wave-2 stance encoded in commercial terms.

The wave-1-vs-wave-2 split mirrors the cloud-infrastructure story of 2010–2015. Vendor-locked wins early on enterprise relationships; open + composable wins long-term on cost, transparency, and developer adoption. Manufacturing AI is in the analogous transition — and the digital product passport mandate accelerates the open side, because vendor-locked attestation is not credible to regulators.

DePIN as Manufacturing Infrastructure​

The D in ABCD — Decentralised Physical Infrastructure Networks — applies directly to manufacturing in three modes.

Mode 1 — Sensor data as marketplace​

A factory's sensors generate data the factory itself doesn't always need to consume. Weather conditions outside the plant, road conditions for inbound logistics, environmental monitoring for compliance, vehicle data for the AGV fleet. DePIN networks pay the operator for contributing this data to the network — turning a cost centre into a revenue line.

DePIN NetworkWhat it captures for manufacturing
WeatherXMSite-level weather data — useful for outdoor curing, drying, energy planning
GEODNETRTK GNSS corrections — cm-precision for AGV, robotic arms, outdoor cells
SoarchainVehicle data — inbound trucks, fleet condition, V2X for in-plant logistics
HivemapperStreet-level imagery — supplier site verification, route planning
Srcful / DaylightSite-level energy data — solar yield, battery state, grid demand-response

See DePIN Devices for the full hardware list.

Mode 2 — Machine identity + on-chain attestation​

A machine with a cryptographic DID can sign every batch it produces. The signature becomes the irrefutable timestamp of who produced what, when, on which line, under what conditions. This unlocks:

  • Digital product passport — Every unit carries a verifiable birth-cert from the moment it leaves the line. Required by ESPR mandate in regulated categories from 2027–2030.
  • Carbon attestation — Scope-1, scope-2, and scope-3 emissions data signed by the machine that produced them, not by the supplier filling a spreadsheet. CSRD-grade evidence.
  • Provenance + chain-of-custody — Cross-border shipments arrive with verifiable origin, conditions, and certifications. Customs friction drops; insurance pricing tightens.

Mode 3 — Machine-payable infrastructure​

Once a machine has identity, it can hold a wallet. Once it has a wallet, it can pay and be paid. This is the precondition for:

  • Pay-per-action APIs — A factory pays the cloud-vision API per inference, the RTK network per correction, the data marketplace per pull. Settled per second, not per month.
  • Throughput-as-collateral — A machine that has attested 18 months of consistent output becomes a financeable asset. Working capital flows against the on-chain throughput history.
  • Autonomous production cells — A self-financing cell that meters its own energy, attests its own carbon, settles its own MRO, and pays its own data fees. The endpoint of factory autonomy.

Crypto Rails — Emerging Industry Instruments​

Track. Selective bets where the regulatory posture allows.

PrimitiveWhat it changes for manufacturingStatus
Machine DIDsCryptographic identity per asset; precondition for everything belowPilot stage; W3C DID standards mature
Digital product passports (DPP)On-chain birth-cert per unit; tamper-evident; queryable lifelongMandated by ESPR for regulated categories 2027–30
Carbon attestationsVerifiable scope-1/2/3 data per unit produced; procurement-grade evidenceEarly pilots; CSRD pressure rising
Stablecoin MRO settlementCross-border supplier + parts payments without correspondent-bank frictionLive in commerce; manufacturing pilots growing
Machine-payable APIsPay-per-inference, pay-per-correction, pay-per-data-pullEarly experiments
Throughput-as-collateralOn-chain output history as financeable working-capital assetEarly pilots in equipment finance
Tokenised supply-chain financeReceivables tokenised against attested delivery eventsLive in fintech; mfg adoption growing
Federated learning + token rewardsCross-plant model training without IP leakageResearch stage; commercial pilots emerging

Integration Patterns — How Wave 2 Wins​

The Wave-2 platform layer wins not by replacing Wave 1 but by routing around it.

PLC / DCS (Siemens, AB, Mitsubishi)        ← unchanged, deterministic, local
        ↓
OPC UA / MQTT / MTConnect                   ← open contract layer
        ↓
Unified Namespace (HiveMQ, Cribl, UNS)      ← vendor-neutral data plane
        ↓
        ├─► MES / ERP (existing, slow)      ← legacy system of record
        ├─► Wave-2 SaaS (Factbird, Tulip)   ← real-time OEE, low-code apps
        ├─► AI agents (Cognex, Landing AI)  ← vision QC, PdM, scheduling
        ├─► DePIN gateways                  ← attestation + sensor marketplace
        └─► On-chain instruments            ← DPP, carbon, machine wallet

The contract is the open layer (OPC UA / MQTT / UNS). Everything plugs in at the contract. The factory keeps optionality across vendor cycles. Wave 1 stays as the system of record; Wave 2 attaches at the contract layer.

What to Skip​

Don't replace your PLC layer to chase AI. The PLC is deterministic, safety-rated, and works. AI attaches above it via the data plane, not by replacing it. Replacing PLC layers is multi-year capex with no AI yield until commissioning is complete.

Don't build a monolithic shop-floor app. Tulip and Factbird already let engineers and operators build their own. The moat is the data + recipe IP, not the app frame.

Don't tokenise routine internal accounting. Stablecoin settlement and machine wallets add value where the alternative is correspondent banking and intermediation fees. For routine intra-company transfers between two plants on the same ERP, the crypto rail adds complexity without leverage.

Don't deploy DePIN where a single sensor would do. DePIN wins on data the factory cannot otherwise economically capture — weather, RTK corrections, vehicle data outside the plant. Inside the plant, a single ifm or SICK sensor on the line is the right answer.

Don't wait for CSRD or ESPR enforcement to start attesting carbon. The buyers that win the next 5-year supply contract will already have the attestation infrastructure in place. The supplier that scrambles in 2027 loses to the supplier that was attesting in 2025.

Context​

Questions​

  • Which layer of the three-layer stack does your plant lack the most — and what is the smallest move that unlocks the next one?
  • Wave 1 vendors are bolting AI on top of legacy schema; Wave 2 vendors are wrapping legacy schema with cloud-native AI. Which approach is your buyer-of-record actually paying for?
  • If you had to instrument one DePIN network for your plant today, which would generate the highest data-revenue-per-dollar-of-hardware?
  • Machine identity unlocks every downstream crypto primitive. Why does no incumbent PLC vendor ship DIDs natively yet — and which Wave-2 vendor will get there first?
  • When DPP enforcement bites in 2027–2030, which suppliers in your chain are infrastructure-ready and which are exposed?