Divergent Compute.AI Economic Think Tank

Markets / Industries / Manufacturing

Industry view · Manufacturing

The factory learns to think, and the floor goes dark

Manufacturing is where AI stops talking and starts moving metal. Predictive-maintenance models already cut unplanned downtime by up to 50%, humanoids logged 1,250 hours on a live BMW line, and China alone now runs over 2 million factory robots. The question for incumbents is no longer whether AI works on the floor, but who owns the software layer once it does.

$8.6B
AI-in-manufacturing market, 2025
Precedence Research
542K
Industrial robots installed in 2024
IFR World Robotics 2025
~50%
Downtime cut by predictive maintenance
McKinsey
$8.5B
Robotics startup funding, 2025
Crunchbase via Figure

01 · The thesis

The value migrates from the arm to the brain

For four decades the economics of factory automation lived in hardware: the robot arm, the PLC, the conveyor. AI inverts that. The arm is now a commodity install base of 4.66 million units worldwide (IFR), while the margin pools shift to the software that tells the arm what to do, predicts when it will fail, and simulates the line before a bolt is turned. Rockwell already books high-single-digit growth in software annual recurring revenue; NVIDIA, Siemens and the cloud hyperscalers are racing to own the digital-twin and physical-AI layer that sits above every brand of robot.

The disruption is bifurcated. Discrete, repetitive, asset-heavy operations — electronics assembly, smartphone fabrication, automotive body-in-white — are being pushed toward lights-out autonomy, with Xiaomi's dark factory producing a phone roughly every second. But the broader sector is stuck in what McKinsey calls pilot purgatory: 78% of firms use AI yet only ~5.5% see material EBIT impact. The winners will be those who convert pilots into floor-wide deployment, and the at-risk are pure-hardware vendors whose intelligence is supplied by someone else.

1Design & Simulation

Digital twins go physically accurate

Production lines are now validated in simulation before deployment, collapsing commissioning time and risk.

NVIDIA Omniverse/Isaac, Siemens Xcelerator, TSMC fab design
2Robotics & Motion

The install base gets a language model

Legacy arms are retrofitted with foundation models so they understand instructions and recognise objects.

FANUC+Google Gemini/Intrinsic across 1.1M robots, ABB, KUKA
3Production Control

Generative copilots write the code

Engineers generate PLC control code and machine visualisations from natural language inside the engineering suite.

Siemens Industrial Copilot, Rockwell, Schneider
4Maintenance & Uptime

Predictive beats reactive

Vibration and sensor models flag failures before they happen, the clearest ROI in the whole stack.

Augury, Cognite, Uptake, Rockwell Lifecycle
5Logistics & Fulfilment

The warehouse becomes a robot swarm

AI orchestration choreographs thousands of bots through high-density distribution and micro-fulfilment.

Symbotic (Walmart), GreenBox, Amazon Robotics
Pace of AI disruption by stage — Divergent Compute assessment

02 · Public players & exposure

Who routes through, who gets routed around

We plot the listed players on two editorial axes — how exposed each is to AI disruption, against how ready its data, brand and position are to be the answer. The figures in the table are sourced; the placement is our read.

Positioning — editorial assessment, not a sourced metric. Bubble = approximate relative scale.
CompanyStanceThe sourced fact
NVIDIANVDAPlatform ownerFANUC, ABB, YASKAWA and KUKA — a combined install base over 2 million robots — are integrating NVIDIA Omniverse and Isaac into their virtual commissioning, per NVIDIA's 2025 newsroom.
SiemensSIEIncumbent aheadSiemens and NVIDIA plan the first fully AI-driven adaptive manufacturing site at the Siemens Electronics Factory in Erlangen starting 2026; its Industrial Copilot generates PLC code from natural language.
Rockwell AutomationROKPivoting to softwareRockwell reported total annual recurring revenue up >6% YoY with software ARR in the high single digits, citing AI-enabled factory projects in automotive, semiconductor and warehousing.
ABBABBNSpinning offABB says 80% of its robotics offering now integrates AI or advanced software; its Robotics & Discrete Automation orders grew 17% YoY before the planned spin-off / SoftBank sale.
FANUC6954.TRetrofitting brainsFANUC is embedding Google's Gemini and Intrinsic platform into its 1.1 million installed robots and has shipped 1,000+ units for Physical AI applications since late 2025.
SymboticSYMBacklog moatSymbotic grew FY2025 revenue 26% to ~$2.3B and reported a ~$22.5B backlog as of Sept 2025, most tied to Walmart and the SoftBank-backed GreenBox JV.
TeslaTSLAUnproven scaleTesla deploys Optimus internally but Musk conceded on the Q4 2025 call the robots are mainly for learning, not productive work, while targeting 1M units/yr capacity by late 2026.
AuguryAUGRYPure-play AIAugury raised $75M in Feb 2025 at a $1B+ valuation ($369M total raised) for AI that detects machine malfunctions from vibration and sensor data.
CogniteCGNTEData backboneCognite reported ~$117M revenue at a $1.5B valuation in 2025, selling the industrial DataOps layer that feeds AI and ML across plant operations.
Figure AIFIGRHype vs proofFigure raised $1B+ Series C in Sept 2025 at a $39B post-money valuation; its robots logged 1,250 hours and >99% placement accuracy on a BMW Spartanburg line.
The map is Divergent Compute’s editorial positioning, offered as a lens, not a measurement. Every figure in the right-hand column is drawn from a named source — see Sources.

03 · The two clocks

The spend, and the payoff

Three timers running at very different speeds across the floor

Share of 2024 global industrial-robot installations by region. Source: IFR World Robotics 2025.

Robots are compounding, not exploding. Global installations reached 542,000 units in 2024 and are forecast to grow just 6% to ~575,000 in 2025 (IFR). The revolution is in the software wrapped around a hardware base that is already enormous — 4.66 million robots in operation, up 9% YoY.

Predictive maintenance is the fast clock. This is where ROI is proven today: McKinsey finds AI-driven predictive maintenance cuts downtime up to 50% and maintenance cost 10-40%, against an average large-plant downtime loss Siemens pegs at $253M per year, up 65% since 2019.

Humanoids are the slow, loud clock. Valuations are running ahead of revenue — Figure at $39B, Tesla promising millions of units — yet the headline proof point is a single 11-month BMW pilot. Real, but a long way from the floor-wide economics the funding implies.

04 · Private flagships

The AI-native challengers

The companies attacking this industry AI-first, with disclosed funding where available:

NVIDIA

The physical-AI operating system

By embedding Omniverse and Isaac into the four biggest robot makers' commissioning tools, NVIDIA positions itself as the layer every brand of arm runs through — the same playbook that worked in data-centre AI, now aimed at the floor.

Public (NVDA); partners cited cover a >2M-robot install base, per NVIDIA 2025

Siemens

Incumbent turning automation into AI

Industrial Copilot already generates control code from plain language, and the Erlangen plant is meant to be the first fully adaptive AI-driven site — the clearest case of a legacy OT giant defending its base with software.

Public (SIE); Erlangen blueprint launches 2026

Symbotic

Warehouse autonomy at Walmart scale

A ~$22.5B backlog and a 26% revenue jump make Symbotic the rare AI-robotics name with real volume — though its fortunes are tightly bound to a single customer and one SoftBank-backed JV.

Public (SYM); ~$2.3B FY25 revenue, ~$22.5B backlog

Augury

Predictive maintenance pure-play

Sensor-plus-AI fault detection is the most de-risked use case in the sector, and Augury has turned it into a unicorn selling directly against incumbents' service revenue.

Private; $75M round Feb 2025 at $1B+ valuation, $369M total

Figure AI

The humanoid bet

A live BMW deployment gives Figure a genuine proof point that most humanoid rivals lack, but a $39B valuation prices in a workforce-scale rollout that has not yet happened.

Private; $1B+ Series C Sept 2025, $39B post-money

Cognite

The industrial data layer

AI on the floor is only as good as the data plumbing beneath it; Cognite sells that DataOps backbone, an unglamorous but structurally sticky position in the stack.

Private; ~$1.5B valuation, ~$117M revenue 2025

05 · Signals

What moved, and what to watch

2024

Robot install base crosses 4.66M

IFR World Robotics 2025 confirms 542,000 new installs in 2024 and a 9% rise in operational stock, with Asia taking 74% of demand and China 54%.

Feb-Sep 2025

Capital floods industrial AI

Augury raises $75M at $1B+, Cognite carries a $1.5B valuation, and Figure closes $1B+ at $39B — total robotics startup funding hits $8.5B for the year.

2025

BMW humanoid pilot completes

Figure robots finish an 11-month Spartanburg deployment, loading 90,000+ parts across 1,250 hours at >99% accuracy; BMW extends Physical AI to Leipzig.

Late 2025

Legacy arms get foundation models

FANUC partners with Google to bring Gemini and Intrinsic to 1.1M installed robots and ships 1,000+ Physical AI units.

2026

The adaptive factory blueprint

Siemens and NVIDIA target the Erlangen Electronics Factory as the first fully AI-driven adaptive manufacturing site, a template for retrofitting existing plants.

06 · The exposure read

Who’s defensible, who’s at risk

AI rewards clean, structured advantage and punishes friction. The line runs through who owns the data, the brand and the customer — and who is merely a step the technology can route around.

Defensible

  • Software and simulation layer owners — NVIDIA, Siemens and the data-platform players capture margin that used to sit in hardware, and their position is brand-agnostic across every robot maker.
  • Predictive-maintenance specialists — Augury, Cognite and incumbent service arms ride the one use case with proven, fast ROI against a $253M/plant/year downtime problem.
  • Asia's scaled adopters — China's 2M+ robot base and 392-per-10,000 robot density give its manufacturers a structural head start on lights-out economics.
  • Warehouse-automation leaders with backlog — Symbotic and Amazon Robotics convert AI orchestration into committed multi-year revenue rather than pilots.

At risk

  • Pure-hardware robot vendors whose intelligence is supplied by NVIDIA or Google risk becoming commoditised arms in someone else's software stack.
  • Pilot-purgatory manufacturers — the ~72% of firms running experiments that never scale burn capital while competitors compound floor-wide gains; only ~5.5% see real EBIT impact.
  • Over-funded humanoid plays carrying valuations like Figure's $39B and Tesla's million-unit promises against thin real-world proof face a brutal re-rating if deployment lags.
  • Low-density, high-cost-labour regions outside Asia that automate slowly cede the lights-out cost curve to Chinese and other scaled adopters.
The hardware revolution in manufacturing already happened; 4.66 million robots are the proof. The 2025-2026 contest is over the intelligence layer — and the incumbents who treat AI as a software product, not a feature bolted onto an arm, are the ones who keep their margins.

Sources

Where this comes from

Precedence Research — AI in Manufacturing Market  ·  IFR — World Robotics 2025 (global robot demand)  ·  McKinsey — The State of AI 2025  ·  McKinsey — Economic potential of generative AI  ·  NVIDIA — Omniverse/Isaac and global robotics leaders  ·  Siemens — Adaptive manufacturing with NVIDIA / Industrial Copilot  ·  Rockwell Automation — Q4/FY2025 Results  ·  ABB — Q3 2025 Results  ·  FANUC + Google — Gemini/Intrinsic for installed robots (TNW)  ·  Symbotic — FY2025 10-K, $22.5B backlog (StockTitan)  ·  Figure AI — $1B+ Series C at $39B (TechCrunch / Sacra)  ·  IIoT World — BMW humanoid deployment ROI  ·  Augury — $75M at $1B+ valuation  ·  Xiaomi dark factory — lights-out manufacturing