Reliance AI & Physical AI — Research Compendium

CompendiumAll research, in one place

This is the overarching hub for the Reliance AI-Devices & Physical-AI research programme. It pulls every workstream into one navigable document, carries the headline evidence from each, reuses the two signature visuals (the 2026 adoption chart and the capability×geography matrix), and ends with the complete Source Library — every link we collected, organised by theme.

The programme in one paragraph

Two linked questions ran through the work: (1) what AI devices should Reliance/Jio pursue, and (2) where in the world do the Physical-AI capabilities behind them actually concentrate. The consistent answer: the smartphone stays primary, AI reaches people through devices they already buy (glasses are the one new object with traction), the real strategic value is a cross-device AI layer rather than a new gadget — and the hard capabilities (robot "bodies", compute, sensing) are layered across geographies that Reliance should access by partnering and investing, not by building a national full stack.

Workstreams covered

AI Devices independent study — glasses · hardware · materials · adoption · journey · propositions
2026 device-adoption data (linear, actual-vs-forecast)
Cross-Area-of-Life customer journey + derived architecture
Global Physical-AI ecosystems + capability×geography matrix
India capability (independent) and Reliance implications
Earlier Reliance scouting: company options · strategic hardware decisions · benchmarks

Evidence discipline (applied throughout)

Every load-bearing claim is tagged FACTSYNTHESISFRONTIERCOUNTER-EVIDENCE and sourced. Never equated: funding = validation · valuation = capability · announcement = product · prototype = product · pilot = scale · shipment = active use · demo = deployment · research-leadership = commercial-leadership · patent = adoption. Counter-evidence was actively sought for every major conclusion.

Full detail lives in the standalone deliverables (see the Deliverables index): the AI-Devices study, the Physical-AI study, two Excel evidence masters, and the master workbook of all tabular research.

Thread 1AI Devices — the independent answer

The smartphone is not being replaced this decade. The high-volume AI story is AI poured into devices people already buy; the one new object that works is camera-AI glasses. The strongest proposition is not another gadget — it is a cross-device personal-AI layer (identity · context · permissions · orchestration), plus a few hardware bets that each attach to an existing behaviour. A standalone "AI companion / pin" does not survive the evidence.

Where AI glasses are going SYNTHESIS — a smartphone companion now, one component of a distributed personal-AI system over five years, not a replacement. Every product ships phone-tethered; the useful loop is capture + open-ear audio + look-and-ask + translation. Displayless camera-AI glasses are the real growth tier (Meta 2M→7M units 2024→2025, ~94% of display-less); display/HUD emerging (Ray-Ban Display $799); true AR remains frontier. EssilorLuxottica

Which AI hardware is worth building — ranked by evidence: (i) ambient home-sensing + care on the gateway; (ii) health/ambient-health sensing (CGM/RPM — strongest India demand); (iii) AI earbuds differentiated by Indic translation-in-noise + hearing-assist (the only resilient India wearable category); (iv) camera-AI glasses (integrate global, prototype affordable). A general-purpose screenless companion/pin fails the test (Humane/Rabbit/Friend). India TWS

What changed in materials × AI FACT — not a wonder-material but a shipping stack: a tolerable waveguide display + a surface-EMG neural band + silicon-anode batteries (15–25% density), converged in the $799 Meta Ray-Ban Display. Generative material discovery, solid-state/structural batteries, e-skin, microLED-AR, non-invasive glucose are gated by yield/durability/cost — options, not dependencies. Meta

The five propositions

P1 — Personal-AI context & permission layer ("one journey, many devices"): the spine, built on India's DPI rails (Aadhaar/UPI/Account Aggregator/ABHA) + MCP/A2A. P2 — Gateway ambient-sensing & care node. P3 — Indic AI hearable (translation-in-noise + hearing-assist). P4 — Health-sensing distribution + RPM. P5 — AI glasses: integrate now + prototype affordable. Not pursued: standalone companion/pin, smart charging case as hero, mass entry smartwatch, near-term owned AR, in-house silicon/phone-OS, non-invasive glucose, consumer BCI/humanoids.

Thread 1 · dataDevice adoption — the 2026 scale reality

Primary metric is annual unit shipments/sales, not market value — to separate technological attention from real adoption. Two things changed the picture in 2026: the smartphone benchmark is now shrinking (−13% on a memory-cost shock), while the AI-on-existing-device wave accelerated (GenAI phones 45% of 2026 shipments; AI PCs ~55%).

Linear scale (not log): glasses/XR/rings are correctly tiny next to the ~1,120M smartphone bar. Metrics are not perfectly comparable — shipments vs sales vs share-derived vs run-rate; see the 2026 dataset for each cell's label and source.

Category	Latest 2026	YoY	Nature
Smartphones (total)	~1,120M FY (−11% Q2)	−13% to −14%	Existing mass purchase (now shrinking)
GenAI smartphones	45% of shipments (~500M)	+9 pts share	AI added to existing device
AI PCs (new)	~55% of PCs (~135M)	~+24 pts	AI added to existing device
Hearables / TWS	82.1M Q2 (~330M run-rate)	−0.7%	Existing purchase (AI subset unmeasured)
Smartwatches	~150M FY	−4%	Existing purchase (~nil GenAI)
Smart speakers	~140M FY	−6%	Existing purchase (GenAI retrofit, no rebound)
Smart glasses (total)	~14M FY (+263% H1)	+263%	New AI-native — small + fast-growing
XR headsets	~8.8M FY	declining (VR/MR −43% '25)	New object — contracting
Smart rings	~6M FY	+53%	New object (fastest-growing wearable)
AI pins / companions	~0 — dead	collapsed	New AI-native — failed

FACT The hypothesis holds and strengthens: AI scales through devices people already own (GenAI phones + AI PCs rising even as the phone base contracts); the one new object with traction (glasses) is still ~80× smaller than smartphones. India diverges on volume (its market is contracting in 2026) while converging on premiumisation. Counterpoint · glasses +263%

Thread 1 · journeyCross-Area-of-Life customer journey & architecture

A researched home→travel→health journey crosses seven Areas of Life across TV, gateway, phone, earbuds and a glucose monitor. It branches (Plan → booking · health prep ★ · shopping · home prep), and proves why the value is a shared layer, not a new device.

What the journey teaches: services exist but are un-integrated (The Ken calls Reliance's health assets a "jigsaw"); travel booking, airport/ground mobility and forex are hard gaps (Adani, not Reliance, runs the airports); telemedicine is thin. Jio does have verified home security cameras + automation (JioThings/JioHome) and international roaming/eSIM. Clinical decisions must route to a licensed professional (Telemedicine Guidelines 2020: "AI cannot counsel or prescribe"). The Ken · guidelines

The architecture it forces: five capabilities must be common across devices and businesses — identity · context/state · permissions/consent · orchestration · audit/trust — while hardware drivers, clinical logic and business data stay specialised. India's DPI rails make the consented version uniquely buildable here.

The AI-companion test (verdict): COUNTER-EVIDENCE NO UNIQUE CUSTOMER JOB IDENTIFIED for a standalone companion. At every moment, phone + earbuds (with TV, gateway, CGM) already does the job; the one flicker — affordable voice-first access for non-smartphone users — collapses into the earbuds themselves (the Indic hearable), not a separate gadget. Consistent with adoption evidence (AI pins ~0, dead).

Thread 2Global Physical AI — where capabilities concentrate

There is no "Physical AI country." Its centre of gravity is split by layer: the US/UK own intelligence (models, compute, simulation, surgical robotics); China owns deployment, volume manufacturing, volume sensing (LiDAR) and consumer robotics; Japan/Germany/Switzerland own the precision-component body and engineering IP; Taiwan/Korea own leading-edge fab + memory; Israel over-indexes on sensing/edge; India owns applied CV + integration + cost, but imports the mechatronic body.

Where to look for what

CAPABILITY — read across to see WHERE	US	China	Japan	S.Kor	Taiwan	Ger	Swiss	France	UK	Israel	India
Embodied AI / robot foundation models	D	S	L	L	·	L	·	·	S	L	L
Humanoid robots — development	D	S	L	S	·	E	·	·	S	S	E
Humanoid robots — manufacturing at scale	E	D	S	S	·	L	·	·	·	·	L
Industrial robotics (arms / automation)	S	S	D	S	·	D	D	L	L	·	L
Industrial deployment scale (installs)	S	D	S	D	·	S	S	L	L	·	L
Autonomous inspection & maintenance	D	S	L	L	·	S	D	L	S	L	E
Autonomous mobility (robotaxi)	D	D	S	S	·	S	·	S	S	D	L
Drones / aerial autonomy	S	D	L	S	·	L	S	L	L	S	E
Computer vision / spatial intelligence	D	S	S	S	·	S	L	S	S	D	S
Sensors — LiDAR / volume sensing	S	D	S	L	·	L	L	L	·	S	L
Sensors — optics / MEMS / precision	S	S	D	S	L	D	D	S	L	S	L
Robotic components (reducers/actuators)	L	E	D	S	·	S	S	L	·	L	L
Edge AI / embedded compute	D	S	S	S	S	S	L	L	L	D	E
Semiconductors — leading-edge fab	E	S	E	S	D	S	·	L	L	L	E
Semiconductors — memory / HBM	S	S	S	D	S	·	·	·	·	·	L
Simulation / digital twin / CAE	D	S	S	S	·	D	D	L	L	L	S
Materials AI (frontier)	D	S	S	S	·	S	L	L	D	L	L
Medical / surgical robotics	D	S	S	S	·	S	S	L	S	L	E
Consumer physical AI	L	D	S	S	·	S	S	L	L	E	L

DemonstratedSignificantEmergingLimited· insufficient

The hard, scarce chokepoints

FACT The Japanese precision-reducer duopoly (Nabtesco ~60% of RV reducers) gates every robot joint — robotics sovereignty is a reducer/actuator-supply problem before an AI one. And the compute stack is function-split: design (US/Nvidia), leading-edge fab (Taiwan/TSMC ~70% of foundry revenue), memory/HBM (Korea/SK Hynix), EUV (ASML, sole).

The validated vs the hyped

COUNTER-EVIDENCE Robotaxi is the only scaled, independently-validated deployment (Waymo 271M+ miles). Humanoids are pre-commercial with funding detached from revenue (Tesla halted Optimus over dexterity; Bain: 8-hr unsupervised shift ~10yrs out). China took 93% of automotive LiDAR as Western LiDAR collapsed. Materials AI is pre-product (A-Lab "41 materials" disputed).

China installed 295,000 industrial robots in 2024 = 54% of the world (IFR); Western installs declined. IFR · reducers · LiDAR

Thread 2 · IndiaIndia capability — independent read

India is strong where value is in code, data and systems engineering; weak where it is in metallurgy, precision mechanics and fab-scale electronics. SYNTHESIS

Genuine strengths

Applied computer vision as deployed product — Qure.ai (18 FDA clearances), Netradyne (450k subs), Detect, Remidio
Warehouse/agri robotics + robotic vision — Addverb (Reliance-owned), Ati Motors, CynLr, Unbox, Niqo
A hard-tech proof — SS Innovations surgical robot (~4,000 surgeries, ⅓ da Vinci cost, telesurgery)
Chip design + AI talent — SiMa.ai (India-origin), Mindgrove; ~416k AI professionals, ~28% of global STEM workforce

Genuine gaps

Precision electromechanics — servo motors, harmonic/cycloidal reducers, actuators (near-total import dependence)
Rare-earth magnets (80%+ China); advanced sensors/LiDAR; leading-edge fab (mature-node only)
Industrial robot density ~30/10k (near world bottom); no indigenous industrial-arm maker
Humanoid hardware (announcement-stage); marquee firms tilt US-domiciled

Complementary partners by gap: Japan (reducers/servos/arms) · Germany/Switzerland (motion/sensing/inspection) · Korea/Taiwan (fab/memory) · US (models/compute/sim) · Israel (sensing/edge). India's realistic 3–5-yr edge: vision-rich, cost-sensitive, software-defined physical systems (medical, warehouse, agri, inspection, fleet) — not the mechatronic substrate. IFR India

SynthesisReliance implications — across both threads

The two threads converge on one posture. On devices: own the connective AI layer, ride existing devices, add only hardware that attaches to a behaviour. On Physical AI: own the perception/AI/deployment layer, partner-and-secure the body and compute, invest selectively in India deep-tech, monitor — not fund — the frontier.

Action	Where it applies	What / who
BUILD	The connective AI layer + applied-CV deployment on own operations	Personal-AI context/permission layer on DPI rails (P1); gateway sensing (P2); deploy on owned Addverb / Netradyne-linked assets
PARTNER / SUPPLY-SECURE	The mechatronic body + compute/sim + components	Japan (Nabtesco/Harmonic Drive reducers), US (Nvidia), DE/FR (Siemens/Dassault); glasses optics (Lumus); chipset partners
PARTNER / DEPLOY	Autonomous inspection on own energy assets; health devices	Gecko, ANYbotics, Detect (industrial); AliveCor/Butterfly (health, distribute)
INVEST	India deep-tech + edge silicon + sensing + Indic AI	CynLr, Ati Motors, SS Innovations, SiMa.ai, Hailo, Dozee, BeatO, Sarvam; Boult (hearables, clean cap table)
ACQUIRE / GROW STAKE	Proven Reliance-linked / India-native assets	Addverb (owned), Netradyne (early backer), Detect, Dozee (distressed RPM) — never merely because "interesting"
MONITOR, not fund	Frontier / benchmark	Humanoids (Figure/Unitree), materials AI (GNoME/MatterGen), robotaxi (Waymo/Baidu), consumer robotics; standalone AI companion/pin (dead)

The single structural insight: SYNTHESIS the capability Reliance most lacks and most needs is the mechatronic body (reducers, actuators, precision motors, advanced sensors) — a Japan/Germany partnership-and-supply problem, not a build problem. On devices, the parallel insight is that the scarce asset is the connective layer, not a new gadget. In both, the winning move is to own the data/integration/deployment layer, partner for the hard physical substrate, and avoid funding the frontier until unit economics are proven.

Earlier scouting detail (company-level BUILD/PARTNER/INVEST/ACQUIRE calls per category) is in the Reliance company-options and strategic-hardware-decisions workbooks; see the Deliverables index and the Source Library §Companies.

IndexAll deliverables produced

The standalone files behind this compendium (delivered earlier in the chat):

🟣

Global-Physical-AI-Strategic-Study.html
Physical-AI study: 10 findings, capability×geography matrix, capability map, ~100-company landscape, India + Reliance implications, 4 slides.

📊

Global-Physical-AI-Evidence-Master.xlsx
10 tabs: ecosystem benchmark, matrix, 101-company universe, deployments, research institutions, funding/M&A, India, diligence candidates, sources.

🟣

AI-Devices-Independent-Strategy-Study.html
AI-Devices study: 12 parts — glasses, hardware, materials, 2026 adoption, customer journey, architecture, 5 propositions, Barbara brief, 4 slides.

📊

AI-Device-Adoption-Dataset-2026.xlsx
2026 device-adoption data: master table + India table + sources, every figure labelled actual/forecast/derived with metric type + date.

🟣

AI-Devices-Customer-Journey-DeepDive.html
Branched home→travel→health journey, exception flow, cross-area transitions, device classification, AI-companion test.

📊

Reliance-AI-Device-Company-Options.xlsx
Concrete company/startup options per AI-device category with BUILD/PARTNER/INVEST/ACQUIRE/MONITOR calls + executive shortlists.

📊

Reliance-AI-Devices-Strategic-Hardware-Decisions.xlsx
Jio hardware track-record evidence, decision rule, device-by-device recommendations, counter-evidence, slide-ready outputs.

📊

Reliance-AI-Devices-MASTER-Workbook.xlsx
All tabular research combined (102 tabs) with a contents index — benchmarks, outlook, org-model, company options, decisions.

Also filed as project docs the synthesis + stream dossiers for both the AI-Devices and Physical-AI studies.

AppendixSource Library — every link, by theme

All 164 links collected across the programme, grouped by theme. Primary sources (regulators, IFR, company filings, analyst houses) and the independent reporting that validates or disconfirms them. Each inline citation elsewhere in the compendium points here.

Device market data & adoption (IDC · Counterpoint · Omdia · Canalys · TechInsights) (19)

AI glasses (20)

Hearables & audio AI (5)

Materials × AI (12)

Home · gateway · Wi-Fi sensing (7)

Jio / Reliance services & India ecosystem (16)

Standards · architecture · health regulation (8)

Physical AI — ecosystems · robotics · chokepoints (9)

Humanoids & embodied AI (foundation models) (19)

Autonomous mobility (robotaxi) & drones (11)

Industrial robotics · inspection · components · edge · sim (7)

Medical / surgical robotics (6)

Consumer robotics (6)

India Physical AI (12)

Reliance device-scouting — company options (7)

Reliance AI & Physical-AI Research Compendium · consolidated 3 Oct 2026 · built from the independent AI-Devices and Physical-AI studies and the Reliance device-scouting workstreams. Evidence established first, Reliance implications derived last; counter-evidence actively sought throughout. Links verified where load-bearing.