1.2 The IoT, IIoT, and Physical AI landscape

Connected embedded systems operate across vastly different environments, from a €30 battery-powered radiator valve in an apartment to a safety-certified vibration controller on a wind turbine or an autonomous mobile robot navigating a hospital corridor. Although the underlying silicon and communication protocols overlap, the engineering requirements differ sharply across three major domains: consumer and smart-home IoT, the Industrial Internet of Things (IIoT), and Physical AI (autonomous embodied systems).

Origins and generations of the Internet of Things

Before the phrase “Internet of Things” existed, engineers built precursors under several names: embedded Internet, machine-to-machine (M2M) telemetry, wireless sensor networks (WSNs), ambient intelligence, and Mark Weiser’s ubiquitous computing (1991)¹. While the underlying communication networks evolved from early academic packet-switched networks (ARPANET in 1969, transitioning to TCP/IP in 1983) to the World Wide Web (created by Tim Berners-Lee at CERN in 1989–1991), those networks were designed around human beings typing at terminals and reading documents.

In 1999, British technology pioneer Kevin Ashton coined the term Internet of Things during a presentation at Procter & Gamble to describe linking Radio-Frequency Identification (RFID) tags in a corporate supply chain to the Internet². Ashton’s observation was straightforward: almost all data available on the Internet at the end of the 20th century had been captured by humans pressing keys, taking photographs, or scanning bar codes. Because humans have limited time, attention, and accuracy, computers knew far more about ideas than about physical things. Giving computers their own means of sensing the physical world removed the human bottleneck.

Key milestones in the transition from early RFID experiments to modern connected embedded infrastructure include:

• 1999–2005 (Early consumer and RFID concepts): Ashton introduces the term IoT (1999)²; LG demonstrates the first Internet-connected refrigerator (2000); the International Telecommunication Union (ITU) publishes its first comprehensive IoT report (2005)³.
• 2008–2009 (The connectivity inflection point): Analysts at the Cisco Internet Business Solutions Group (IBSG), led by Dave Evans (2011)⁴, estimate that between 2008 and 2009 the number of connected physical devices surpassed the human population of the planet (~6.8 billion at the time).
• 2012 (World IPv6 Launch): Originally specified by the IETF in 1995–1998 (RFC 1883 / RFC 2460, elevated to full Internet Standard RFC 8200 in 2017)⁵, IPv6 expands the IP address space from 32 bits (2³² ≈ 4.29 × 10⁹ addresses) to 128 bits (2¹²⁸ ≈ 3.4 × 10³⁸ addresses). Following the 24-hour World IPv6 Day test on 8 June 2011, major network operators permanently enabled IPv6 on 6 June 2012, providing the addressing foundation required for low-power mesh protocols such as 6LoWPAN and Thread (Chapters 15 and 17).
• 2014–2020 (First-generation cloud IoT and fragmentation): Rapid expansion of consumer smart-home appliances and proprietary cloud platforms. Competing wireless stacks (Zigbee, Z-Wave, BLE, proprietary sub-GHz) force users to deploy multiple incompatible bridge hubs.
• 2022–2027 (Edge intelligence, Matter, and mandatory cybersecurity regulation): The Connectivity Standards Alliance (CSA, formerly the Zigbee Alliance) releases Matter (2022, reaching Matter 1.5 in November 2025) over Thread and Wi-Fi to unify application-layer interoperability. Simultaneously, Connected Devices cross 20 billion active nodes in 2025 (Hanafy et al., 2026)⁶, while European regulation shifts connected security from voluntary best practice to law through the Radio Equipment Directive (RED) cybersecurity delegated act (applicable from August 2025), the EU Data Act (applicable from September 2025), and the EU Cyber Resilience Act (CRA) (vulnerability reporting mandatory from September 2026, full compliance from December 2027; Chapter 19)⁷.

Three architectural generations

Across these two and a half decades, the system architecture of connected devices has progressed through three distinct generations:

Characteristic	Generation 1: Cloud-Centric IoT	Generation 2: Edge and Fog IoT	Generation 3: Collaborative & Physical AI (IoCT)
Primary data path	Leaf devices stream raw sensor data directly (or via a transparent bridge) to a remote cloud server.	Leaf nodes talk over local low-power meshes to an Embedded Linux edge gateway that filters, aggregates, and acts locally.	Heterogeneous devices pool sensing, compute, and actuation horizontally across local trust domains (Hanafy et al., 2026)6.
Where intelligence lives	Centralised cloud data centres.	Split between microcontroller TinyML/DSP filters and gateway containers.	Distributed across physics-informed on-device models, gateways, and peer autonomous agents.
Failure mode on WAN outage	Device becomes inoperable (“bricked” thermostat or smart lock).	Local control loops, alarms, and HMIs continue operating autonomously.	Devices dynamically re-orchestrate tasks across surviving local peers.
Security model	Perimeter-based (often weak or hardcoded credentials; unencrypted local buses).	Hardware root of trust, secure boot, mutual TLS (mTLS), local data entitlement.	Hardware attestation plus dynamic, time-bounded capability tokens across domains.

Consumer IoT and the smart home: lessons in architecture

The smart home illustrates both the practical value of connected embedded systems and the architectural mistakes of first-generation cloud-centric design.

Consider the evolution of a domestic heating thermostat:

1. Mechanical bimetallic strip: A purely physical feedback loop. Two bonded metals with different coefficients of thermal expansion bend to open or close a contact. It has zero network attack surface, costs a few euros, and lasts decades, but offers wide hysteresis and no scheduling.
2. Programmable digital thermostat: Adds a thermistor, a small microcontroller, an LCD, and a 7-day timer. It reduces energy waste only if the occupant manually programs schedule changes, and the interface makes that hard: a typical week of four periods a day is 28 times and 28 temperatures, entered one value at a time through a handful of buttons on a small LCD that never shows the whole week. Many of these thermostats were left on their factory schedule or held at a fixed temperature, and field studies found that they saved little energy in practice; in 2009 the US Environmental Protection Agency stopped awarding them its Energy Star label for this reason.
3. Connected learning thermostat (e.g., Google Nest): Combines temperature, humidity, ambient light, and passive infrared (PIR) occupancy sensing with Wi-Fi connectivity, cloud weather feeds, and schedule-learning algorithms. (Google acquired Nest Labs for $3.2 billion in early 2014.) Similarly, connected smoke and carbon-monoxide alarms such as the Nest Protect (discontinued by Google in 2025) link multiple ceiling units across a home so that if smoke is detected in a basement workshop, every alarm in the house announces the specific room in spoken audio, sends a smartphone alert, and signals the heating controller to shut down the gas boiler upon detecting carbon monoxide.

Figure 1.3 puts the three generations side by side. Turn the dial on the mechanical thermostat, try to change one period of the programmable thermostat’s schedule (and count the button presses), then turn the ring on the connected thermostat and see what still works when its internet link fails.

Figure 1.3

Three generations of heating thermostat

derekmolloy.ie

Choose a generation, then operate the sketch.

1. Mechanical2. Programmable3. Connected learning

Try it: change Wednesday's Wake period from 06:00, 21 °C to 07:00, 20 °C. Press SET to start editing (SET moves to the next field, ▲ and ▼ change it) and RUN to save.

Button presses: 0 (fewest possible: 12)Reset

What is stored (the LCD shows one value at a time)

WAKE	06:00·21°	06:00·21°	06:00·21°	06:00·21°	06:00·21°	06:00·21°	06:00·21°
LEAVE	08:00·16°	08:00·16°	08:00·16°	08:00·16°	08:00·16°	08:00·16°	08:00·16°
RETURN	18:00·21°	18:00·21°	18:00·21°	18:00·21°	18:00·21°	18:00·21°	18:00·21°
SLEEP	22:00·16°	22:00·16°	22:00·16°	22:00·16°	22:00·16°	22:00·16°	22:00·16°
	MON	TUE	WED	THU	FRI	SAT	SUN

A full week is 7 days × 4 periods, each with a time and a temperature: 56 values entered through five buttons, with no view of the whole week. Many of these thermostats were left on their factory schedule, or switched to HOLD, and saved little energy.

Programmable 7-day thermostat

Senses

Room temperature, with a thermistor

Decides

A small microcontroller follows a 7-day schedule of set points

Connects

Nothing

Acts

A relay switches the boiler

Cost and life

Tens of euros

Limits

Saves energy only if someone programs the schedule, through five buttons and a small LCD

Figure 1.3. Three generations of domestic heating thermostat, drawn as generic sketches. The decision moves from the physics of a bimetallic strip, to a schedule that the occupant must program, to a schedule learned from the occupant’s adjustments and occupancy sensing, with remote features that depend on the vendor’s cloud.

How connected alarms actually talk when Wi-Fi fails

In consumer marketing, interconnected alarms are sometimes loosely described as using “ad-hoc Wi-Fi”. In reality, the alarms must keep talking to each other without depending on a domestic Wi-Fi router. The Nest Protect therefore includes a dedicated IEEE 802.15.4 low-power radio alongside its Wi-Fi interface, forming a direct battery-backed mesh network between alarm units using Nest’s 802.15.4 mesh stack (which became the Thread protocol covered in Chapter 17) to carry Nest Weave, an application-layer protocol that later fed into Matter.

When consumer IoT goes wrong: cloud dependency, fragmentation, and privacy

Despite clear benefits in energy management (such as coordinating heat pumps and immersion heaters with local solar photovoltaic generation and time-of-use electricity tariffs), water-leak detection, and assisted living for older adults, consumer IoT has exposed four recurring engineering failures:

1. Cloud tethering and forced obsolescence: When a device relies on a vendor’s cloud server to interpret button presses or schedules, the hardware’s lifespan is tied to the vendor’s commercial priorities rather than the durability of its electronics. For example, in October 2025 Google ended cloud support and app connectivity for its 1st- and 2nd-generation Nest Learning Thermostats, rendering their connected features inoperable even though the physical sensors and relays remained intact.
2. Protocol fragmentation: Until recently, a homeowner installing smart lighting, radiator valves, and door locks frequently encountered incompatible radio and application layers: Zigbee (2.4 GHz mesh), Z-Wave (868 MHz sub-GHz mesh in Europe), Bluetooth Low Energy, Wi-Fi, and legacy proprietary protocols such as LightwaveRF or powerline X10. Bridging these protocols without vendor lock-in is one of the primary reasons to deploy an open-source Embedded Linux gateway running Home Assistant and Node-RED (Chapter 16).
3. The “Internet of Creepy Things” and default passwords: Consumer cameras, doorbells, vehicle telematics boxes, and appliances collect intimate behavioural telemetry. When shipped with default credentials or unpatched firmware, they become mass surveillance liabilities. In one widely publicised early incident, websites indexed live feeds from more than 73,000 commercial IP cameras that still used their factory-default vendor passwords. By April 2022, Raspberry Pi OS removed the historical default pi / raspberry account for the exact same reason, and European regulation now forbids universal default passwords on connected radio products: the RED Delegated Regulation (EU) 2022/30 has applied since August 2025 (assessed against EN 18031), and the Cyber Resilience Act extends this to all connected products from December 2027 (Chapter 19)⁷.
4. Feature excess (“Internet of Useless Things”): Adding a camera and Wi-Fi SoC to a refrigerator or a toaster adds €100 to €400 in hardware cost and expands the home’s network attack surface while solving no meaningful physical problem. Good connected embedded engineering starts by asking whether a network link and an operating system are justified by the Information Value Loop.

The last two failures came together in November 2026, when connected refrigerators were reported to have been hacked around the world. This BBC News report explains what happened:

Play

“Why are fridges being hacked around the world?”. Credit to @BBC News on YouTube.

In a well-architected smart building, leaf sensors publish structured payloads over a local broker on an Embedded Linux gateway rather than routing every state change through a distant continent. A typical multi-sensor node payload routed through a local Mosquitto MQTT broker and processed by Node-RED or Home Assistant looks like this:

{
  "device_id": "env-node-04",
  "timestamp": "2026-11-07T18:20:00Z",
  "temperature_c": 21.4,
  "pressure_hpa": 1013.2,
  "humidity_pct": 48.5,
  "gas_resistance_ohm": 45200,
  "ds18b20_flow_c": 54.2,
  "ds18b20_return_c": 43.8,
  "load_kg": 12.6
}

We will build the complete hardware-to-dashboard pipeline for this class of local telemetry in Chapters 8, 9, 14, and 16.

The Industrial Internet of Things (IIoT) and the outcome economy

While a failed smart light bulb in a living room is a minor annoyance, a missed deadline or security breach in an industrial plant can halt a production line, damage multi-million-euro machinery, or injure workers. The Industrial Internet of Things (IIoT) applies connected embedded systems to manufacturing, energy grids, rail, aviation, agriculture, oil and gas, and clinical healthcare, bridging Operational Technology (OT) (programmable logic controllers, motor drives, fieldbuses, and safety interlocks) with Information Technology (IT) (IP networks, time-series databases, and enterprise analytics).

According to market analysis in the BCC Research AI & IoT Technology Research Review (2025, Report IFT317A / IFT299A)⁸, the global IIoT market grew from $124.7 billion in 2022 toward a projected $391.8 billion by 2028 (a compound annual growth rate of 21.6%). Crucially for embedded engineers, hardware (sensors, actuators, industrial gateways, and network interfaces) accounts for more than 50% of total IIoT revenue, reflecting the physical reality that software analytics cannot exist without ruggedised field instrumentation. Broader enterprise IoT spending reached about $324 billion in 2025, with forecasts approaching $649 billion by 2030 (IoT Analytics, 2025)⁹.

The four phases of Industrial Internet evolution

In their foundational report Industrial Internet of Things: Unleashing the Potential of Connected Products and Services, the World Economic Forum and Accenture (2015)¹⁰ mapped the trajectory of industrial connectivity across four progressive phases (Figure 1.4):

Figure 1.4

Four phases of the Industrial Internet

derekmolloy.ie

Click a phase, or a group label, to see what it covers.

Operational payback

Phase 1Asset monitoring, predictive maintenance, energy reduction

Phase 2Software-differentiated products, remote diagnostics and updates

New business models

Phase 3Sell guaranteed, measured results instead of hardware

Phase 4Autonomous coordination and a human-machine workforce

Select a phase to see its description and examples.

Figure 1.4. The four phases of the Industrial Internet evolution, adapted from World Economic Forum & Accenture (2015)¹⁰. Phases 1 and 2 deliver immediate operational payback, while Phases 3 and 4 restructure industrial business models and physical autonomy.

1. Phase 1: Operational efficiency (Immediate payback): Retrofitting existing industrial assets with vibration, acoustic, thermal, and current sensors to reduce unplanned downtime and energy consumption.
2. Phase 2: New connected products and software-driven services: Embedding compute and telemetry inside commercial equipment so that physical products are continuously improved and differentiated by software in the field.
3. Phase 3: The outcome economy: Shifting from selling physical machinery to selling quantifiable, guaranteed operational outcomes measured in real time by embedded sensors (for example, selling guaranteed megawatt-hours of wind generation, cubic metres of compressed air, or zero-downtime elevator availability rather than selling a physical compressor or lift).
4. Phase 4: The autonomous pull economy and integrated human-machine workforce: Connecting demand signals directly to flexible, autonomous manufacturing cells and mobile robotics, while augmenting human technicians with real-time spatial and diagnostic data.

Phase 4 is already visible in large logistics centres. This Euronews report goes inside Amazon’s busiest European warehouse, where mobile robots, laser scanning and human workers handle orders together (Amazon Robotics, formerly Kiva Systems, is one of the deployments listed below):

Play

“Inside Amazon’s busiest European warehouse, where robots, lasers and humans deliver the future”. Credit to @Euronews Next on YouTube.

Real-world industrial deployments

Across both the WEF (2015, 2025, 2026)¹⁰¹¹¹² reports and the BCC Research (2025)⁸ industrial survey, several concrete deployments show how these four phases work on the factory floor and in the field:

• Zero-downtime robotics and predictive maintenance (Fanuc, ThyssenKrupp, and Shell): Unplanned equipment failure is the single largest cost driver in continuous manufacturing. Fanuc instruments its industrial robots with vibration, current, and thermal telemetry to forecast reducer and servo failures before a line stops, establishing a “zero-downtime” manufacturing benchmark (BCC Research, 2025)⁸. ThyssenKrupp connects elevator controllers and door-drive sensors to identify mechanical wear patterns, separating faults that require immediate technician dispatch from those that can wait for scheduled maintenance (WEF, 2015)¹⁰. In energy production, Shell deploys downhole and surface sensor networks to monitor drilling vibration and pipeline integrity in real time (BCC Research, 2025)⁸.
• Software-differentiated machinery and autonomous field vehicles (ZF Friedrichshafen and John Deere): Automotive supplier ZF Friedrichshafen combines driver-behaviour data with topographical map data in commercial truck transmissions to anticipate road gradients and select gear shifts automatically, reducing fuel consumption and mechanical wear (WEF, 2015)¹⁰. John Deere combines centimetre-accurate RTK-GNSS, stereo computer vision, and inertial measurement units on autonomous tractors to till and spray fields with minimal chemical waste (BCC Research, 2025)⁸.
• Connected assembly, logistics, and clinical monitoring (Airbus, Bosch, Amazon Robotics, and Philips): In its Factory of the Future initiative, Airbus instruments assembly tools and equips technicians with connected torque wrenches and smart glasses to verify fastener specifications across millions of aircraft components (BCC Research, 2025)⁸. Bosch embeds localisation sensors in factory tooling under its Track and Trace programme to eliminate hundreds of person-hours wasted searching for calibration equipment. In warehouse fulfilment, Kiva Systems (acquired by Amazon for $775 million in 2012, now Amazon Robotics) pioneered fleets of autonomous drive units that bring shelving pods directly to human pickers. In healthcare, Philips networks multi-parameter bedside monitors so that clinical alarms cross-check multiple physiological signals before alerting staff, reducing false-alarm fatigue and medication errors (WEF, 2015; BCC Research, 2025)¹⁰⁸.

From Industry 4.0 to Industry 5.0 and decarbonisation

While Industry 4.0 (a term introduced in Germany in 2011) focused on digitising factories through cyber-physical systems, IIoT, and cloud analytics, the European Commission and the World Economic Forum’s Fourth Industrial Revolution (4IR) Network Impact Report (2026)¹³ emphasise a complementary shift toward Industry 5.0: placing human workers, resilience, and environmental sustainability at the centre of industrial automation. For example, the WEF Transitioning Industrial Clusters initiative uses real-time industrial telemetry to coordinate shared energy grids, track carbon intensity, and optimise circular material flows across industrial ports.

Why industrial AI pilots stall: lessons from the WEF MINDS cohort (2026)

As industrial organisations move from basic telemetry to deploying artificial intelligence across their operations, many early pilot projects fail to reach production. In two landmark industry studies, AI in Action: Beyond Experimentation to Transform Industry (WEF & Accenture, January 2025)¹¹ and Proof over Promise: Insights on Real-World AI Adoption from 2025 MINDS Organizations (WEF & Accenture, January 2026)¹², the World Economic Forum analysed hundreds of global enterprise AI deployments.

Two findings from the 2026 MINDS (Meaningful, Intelligent, Novel, Deployable Solutions) cohort point directly to the role of connected embedded engineering:

1. “Data quality” is the number-one barrier to real-world AI impact: Enterprise AI models trained purely on historical IT records fail when confronted with physical plant dynamics. Successful industrial organisations overcome this by building differentiated data foundations that combine structured records with real-time edge telemetry, synthetic edge scenarios, and physics-based data (WEF, 2026)¹². High-impact AI depends on clean, calibrated, accurately timestamped sensor streams from the field.
2. “Limitations of technical infrastructure” is a top-three barrier: Cloud-only infrastructure cannot meet the latency, bandwidth, or reliability demands of factories, hospitals, and logistics hubs. The 2026 MINDS report notes that leading organisations are systematically modernising their technology stacks by deploying edge computing devices that process information directly where it is generated, replacing fragmented point solutions with unified edge-to-cloud platforms¹². At the same time, connecting legacy OT equipment introduces severe cybersecurity risk: according to Trend Micro data cited by BCC Research (2025)⁸, industrial control system (ICS/OT) security breaches cost an average of nearly $3 million per incident, making hardware secure boot, network segmentation, and encrypted telemetry mandatory (Chapter 19).

Physical AI and next-generation physical autonomy

The convergence of connected embedded systems, multimodal sensing, and on-device machine learning has given rise to Physical AI (also termed embodied AI or physical autonomous systems). In its April 2026 briefing paper Next-Generation Physical Autonomy: Advancing Intelligent Systems for a Thriving and Resilient Society¹⁴, the World Economic Forum Global Future Council on Autonomous Systems defines physical autonomous systems as systems that “integrate artificial intelligence, information processing, sensing and electro-mechanical capability to perceive, decide and operate in complex environments”, moving out of caged factory cells into semi-structured and unstructured human spaces (public roads, hospitals, construction sites, agriculture, and domestic homes).

Physical AI on prime-time television

Robots are now part of mainstream entertainment. In this broadcast from the 2026 Spring Festival Gala, robots perform a martial-arts routine on stage:

Play

“Martial arts robots dazzle at 2026 Spring Festival Gala”. Credit to @CGTN on YouTube.

What the video shows, and what it does not

This is a rehearsed stage performance, filmed and edited for broadcast. It shows how far motor control, balance and coordination between robots have come. It does not show robots deciding what to do from what they sense on the night; a routine like this is choreographed in advance. Keep that distinction in mind as you watch, and consider: which parts of the routine need sensing and real-time control on the robot itself (keeping balance, recovering from a slip, avoiding a neighbour that is out of position), and which parts could be scripted beforehand? The first group is the subject of Chapter 11; the split between what a device does locally and what it is told is the two-tier architecture of Section 1.3.

Why Physical AI is different from digital AI

A Large Language Model (LLM) running in a data centre operates purely on discrete digital tokens. If a cloud server takes an extra 800 ms to generate a paragraph of text, nobody gets hurt; if the model hallucinates a word, the user can regenerate the prompt.

As the WEF Next-Generation Physical Autonomy (2026)¹⁴ paper and Chen et al. (AI+HW 2035, 2026)¹⁵ emphasise, Physical AI must contend with the continuous, uncertain, and unforgiving dynamics of the physical world:

• Irreversible physical consequences and forces: A bipedal robot, surgical manipulator, or delivery drone applies real mechanical forces to its surroundings. It must respect Newton’s laws, friction, inertia, and actuator torque saturation at every millisecond.
• Strict real-time loops and offline survival: An autonomous system cannot wait for a cloud round-trip when a pedestrian steps into its path or a gust of wind disturbs its attitude, and it must maintain safe control when wireless connectivity drops completely inside a tunnel, basement, or rural field.
• Physics-informed world models: Rather than relying solely on correlation over text or pixels, next-generation Physical AI couples data-driven learning with physical priors and differentiable physics simulators. Chen et al. (2026)¹⁵ highlight the emergence of latent world models, such as Joint Embedding Predictive Architectures (JEPA), which learn structured internal representations of how physical objects move and interact over time without wasting compute reconstructing every background pixel.

Four scenarios for physical autonomy by 2031

Because breakthroughs in actuators, sensors, and embodied models can occur rapidly while safety standards and public trust evolve slowly, the WEF Global Future Council on Autonomous Systems (2026)¹⁴ maps four plausible trajectories for physical autonomy over the 2026–2031 horizon along two axes: the rate of technological innovation and the level of societal acceptance:

Scenario (2031 horizon)	Rate of innovation	Societal acceptance	Engineering and deployment reality
1. Proven deployment	Low / steady	High	Gradual, reliable expansion in structured environments (ports, warehouses, mining, energy inspection). Robots are trusted as dependable industrial infrastructure, though general-purpose autonomy remains limited.
2. Integrated progress	High	High	Breakthroughs in perception, manipulation, and edge compute align with transparent safety verification, clear regulation, circular manufacturing, and workforce reskilling. Autonomous systems safely augment human work across healthcare, construction, and logistics.
3. Tech disillusionment	Low	Low	Overpromised capabilities, high total cost of ownership, supply-chain bottlenecks in sensors/actuators, or visible safety failures erode public and investor confidence, stalling deployment.
4. Divided deployment	High	Low	Rapid technical capability advances without shared safety governance, cybersecurity-by-design, or public legitimacy, leading to fragmented regional technology stacks, dual-use geopolitical friction, and public pushback.

Moving the field toward Proven deployment and Integrated progress rather than Tech disillusionment or Divided deployment is not a marketing challenge; it is an embedded systems engineering challenge. It requires deterministic real-time control (Chapters 11 and 12), verifiable sensor fusion (Chapter 9), low-power edge inference (Chapters 13 and 21), resilient local networking (Chapters 15 to 18), and hardware-anchored cybersecurity and regulatory compliance (Chapters 19 and 20).

Scale at the far extreme: micro- and nano-scale IoT

While most of this book focuses on centimetre-scale microcontrollers and single-board computers, researchers are also pushing connected sensing down to the micrometre and nanometre scales. Early concepts of Smart Dust (pioneered at UC Berkeley with DARPA and RAND support) envisioned cubic-millimetre autonomous sensing motes powered by micro-energy harvesting. At an even smaller scale, Akyildiz and Jornet (2010)¹⁶ formulated the Internet of Nano Things (IoNT), exploring graphene and carbon-nanotube antennas, molecular communication, and terahertz-band (0.1 THz to 10 THz) electromagnetic channels inside the human body or industrial materials (where molecular absorption limits the range). Although nano-scale nodes use specialised physical channels, they obey the exact same Information Value Loop and energy-per-bit trade-offs as the macro-scale leaf nodes in our lab kit.

🧩 Knowledge Check

Concept Match

Match the phase of Industrial Internet evolution (WEF, 2015)

derekmolloy.ie

Quiz

Select 0/1

Connected Nest Protect alarms keep alerting each other when the home's Wi-Fi router fails. How?

derekmolloy.ie

Quiz

Select 0/3

Which of the following are recurring failures of consumer IoT described in this section?

derekmolloy.ie

Quiz

Select 0/1

In a Generation 2 (edge and fog) architecture, what happens when the WAN link to the cloud fails?

derekmolloy.ie

Footnotes

1. M. Weiser, “The Computer for the 21st Century,” Scientific American, vol. 265, no. 3, pp. 94–104, Sept. 1991. ↩

2. K. Ashton, “That ‘Internet of Things’ Thing,” RFID Journal, 22 June 2009 (recounting his 1999 presentation at Procter & Gamble). ↩ ↩²

3. International Telecommunication Union (ITU), ITU Internet Reports 2005: The Internet of Things, Geneva, Nov. 2005. ↩

4. D. Evans, The Internet of Things: How the Next Evolution of the Internet Is Changing Everything, Cisco Internet Business Solutions Group (IBSG) White Paper, Apr. 2011. ↩

5. S. Deering and R. Hinden, Internet Protocol, Version 6 (IPv6) Specification, IETF Internet Standard RFC 8200 (obsoleting RFC 2460), July 2017. ↩

6. W. A. Hanafy, N. Sehatbakhsh, D. Irwin, M. Srivastava, and P. Shenoy, “The Internet of Collaborating Things: Agentic Edge AI for Autonomous Cross-Domain Collaboration,” arXiv preprint arXiv:2609.10542v1, June 2026. ↩ ↩²

7. European Parliament and Council of the European Union, Regulation (EU) 2024/2847 on horizontal cybersecurity requirements for products with digital elements (Cyber Resilience Act), Official Journal of the European Union, Nov. 2024; and Commission Delegated Regulation (EU) 2022/30 supplementing Directive 2014/53/EU (Radio Equipment Directive), Jan. 2022. ↩ ↩²

8. BCC Research, AI & IoT Technology Research Review (Reports IFT317A and IFT299A: Global Markets for Industrial IoT), 2025. ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶ ↩⁷ ↩⁸

9. IoT Analytics, State of IoT — Enterprise IoT Market Report & Forecast 2025–2030, 2025. ↩

10. World Economic Forum in collaboration with Accenture, Industrial Internet of Things: Unleashing the Potential of Connected Products and Services, Geneva, Jan. 2015. ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶

11. World Economic Forum in collaboration with Accenture, AI in Action: Beyond Experimentation to Transform Industry, Geneva, Jan. 2025. ↩ ↩²

12. World Economic Forum in collaboration with Accenture, Proof over Promise: Insights on Real-World AI Adoption from 2025 MINDS Organizations, Geneva, Jan. 2026. ↩ ↩² ↩³ ↩⁴

13. World Economic Forum, Fourth Industrial Revolution (4IR) Network Impact Report, Geneva, 2026. ↩

14. World Economic Forum Global Future Council on Autonomous Systems, Next-Generation Physical Autonomy: Advancing Intelligent Systems for a Thriving and Resilient Society, Briefing Paper, Apr. 2026. ↩ ↩² ↩³

15. Y. Chen, J. Cong, A. Mirhoseini, C. Kozyrakis, S. Mitra, Y. LeCun, R. Puri, et al., AI+HW 2035: Shaping the Next Decade, 2026. ↩ ↩²

16. I. F. Akyildiz and J. M. Jornet, “The Internet of Nano-Things,” IEEE Wireless Communications, vol. 17, no. 6, pp. 58–63, Dec. 2010, doi: 10.1109/MWC.2010.5675779. ↩