The humanoid wave is the biggest upside in the QNX story — and the earliest. Here’s what’s real today, and what’s a 2030s bet.
By The 🍌🐀 (The Banana Rat)
Scope & disclosure. A forward-looking editorial thesis on the robotics safety-software opportunity, not investment advice. As-of date: 2026-07-01. Factual claims are sourced in the endnotes; inferences are labeled as thesis. Conflict-of-interest disclosure: the author holds a position in BlackBerry (BB), whose QNX division is discussed below. Do your own research.
Part of the 🍌🐀 Physical AI & Edge Compute field guide → the hub for the whole thesis.
Everyone is watching the robot’s hands. The humanoid waving from the keynote stage, the backflip, the folding of laundry. Same mistake the crowd made with the car. The hands and the brain behind them are the show. The business is the boring thing underneath: the certified reflex that guarantees the machine stops before it crushes a foot, even when the brilliant AI brain has confidently decided the foot isn’t there. That reflex is the exact slot QNX already fills in 275 million cars — and it’s now being wired into robots by the one company that owns the robot brain.
Let’s separate what’s real today from what’s a bet on the 2030s, because both are in this story and the honest version needs both.
First, the honest size of it
The forecasts for humanoid robots are enormous and wildly inconsistent — and that spread is the point. Goldman Sachs sees a ~$38 billion market by 2035 (about 1.4 million units). Morgan Stanley sees a ~$5 trillion ecosystem and over a billion humanoids by 2050, with ~13 million in service by 2035. Bank of America’s 2026 forecast runs from ~90,000 units this year to 10 million a year by 2035 and 3 billion installed by 2060 [1]. When credible banks disagree by two orders of magnitude, the honest read is: nobody knows, the direction is up, and the near-term number is small. Today’s humanoid market is roughly $3 billion [1]. The real robotics revenue right now is in unglamorous autonomous mobile robots (warehouse and factory movers) — a ~$5 billion market growing around 20% a year [2].
So: gigantic someday, modest today. Anyone selling you the $5 trillion number as if it arrives next quarter is selling.
Why a robot needs a certified reflex at all
Here’s the structural argument, and it’s the same one that holds in the car. A modern robot’s intelligence, the vision-language-action model that lets it improvise, is probabilistic. It guesses the best next action. That is exactly why it is brilliant, and exactly why it cannot be certified to a safety integrity level: you can’t prove a bound on a system that’s allowed to surprise you. So when a machine shares space with a human, functional-safety practice requires a separate, deterministic channel (emergency stop, speed and force limiting, collision boundaries) that can override the AI the instant it misbehaves [4].
The standards are catching up to exactly this shape. ISO 10218 for industrial robots was refreshed in 2025 with safety functions up to SIL 3; ISO 13482 covers personal-care robots [4]. The tell for how early the humanoid wave really is: there is no finalized safety standard for a dynamically walking humanoid yet — the relevant standard (ISO 25785) is still in draft, and certifying a robot to work unfenced, around people, remains the open deployment barrier [4]. The reflex layer isn’t optional. It’s just early.
NVIDIA hands QNX the slot
This is the load-bearing part, because it isn’t QNX marketing itself — it’s the industry’s compute kingmaker doing it. NVIDIA has consolidated the robot brain: its Jetson Thor module (a 2,070-TFLOP machine with an integrated functional-safety processor) anchors a roster of humanoid makers — Figure, Agility, Boston Dynamics, 1X, Apptronik, Unitree and more [3]. Then, in June 2026, NVIDIA shipped Halos for Robotics, billed as the industry’s first full-stack safety system for physical AI — and it named QNX.
Halos ships in two flavors: Linux-only, or Linux + QNX, where NVIDIA’s hypervisor partitions the compute into a Linux VM running the AI and a QNX VM running the safety-critical functions. NVIDIA’s own words: QNX is “a real-time operating system with a long pedigree in certified safety systems… its inclusion enables stronger software partitioning for higher safety integrity use cases” [5]. That is the entire thesis of this cluster (the probabilistic brain in one partition, the certified reflex in another) reproduced in robots, by NVIDIA, on the record. (It’s distinct from, and builds on, the April 2026 QNX–IGX Thor integration [6].)
What QNX has actually landed — beyond the car
The robotics revenue isn’t here yet, but the proof that QNX travels beyond automotive very much is. Management now says the General Embedded Market is roughly 20% of QNX revenue, with a total addressable market it believes could eventually exceed automotive — on the back of a record QNX quarter ($78.7M) and a ~$950M royalty backlog [7]. The named non-automotive wins are concrete:
- Medical robotics: Kinova’s KIMA surgical-capable arm is built natively on QNX OS 8.0, both developed to IEC 62304 Class C; QNX and Kinova say a pre-certified OS shaves 12–18 months off device development [8].
- Medical devices: Johnson & Johnson selected QNX OS for Safety to power a new AI-driven heart pump [9].
- Rail: Medha’s communications-based train control for India’s metros runs on QNX OS for Safety, targeting the highest rail-signaling integrity level [10].
- Industrial: a General Embedded Development Platform win with an (unnamed) major North American OEM [9].
- Ecosystem standing: Arm named QNX a foundational software partner at its physical-AI launch, and ABI Research ranked QNX a top Leader in RTOS for robotics functional safety; Apex.AI’s ROS 2-based autonomy framework is now compatible with QNX SDP 8.0 [11].
That’s a pipeline, credentialed by the right names — not yet a robotics revenue line.
The honest hedge
Sell this as credible optionality with proof points, not a robotics revenue engine. Three things keep it calibrated: (1) timelines are hype-prone: Morgan Stanley itself calls humanoid adoption “relatively slow” until 2035, and the bank forecasts diverge fivefold-plus; (2) most robot safety software shipping today is bespoke ROS 2/Linux, not a certified RTOS: the certified-safety-layer architecture is endorsed by NVIDIA and QNX but early in real deployment; and (3) QNX’s landed non-automotive wins are industrial, medical and rail, and mostly pre-shipment: they feed backlog rather than current revenue, and the NVIDIA Halos/IGX and Apex.AI items are non-exclusive enablement where QNX is one of several partners, with no disclosed robotics revenue.
Actionable Takeaway: the robotics/humanoid leg is where the automotive playbook says the puck is going (the same certified-reflex-under-probabilistic-brain pattern, now blessed by NVIDIA) — but on a multi-year, not multi-quarter, clock. Watch two tells: GEM revenue disclosure finally breaking out non-automotive dollars, and a named humanoid or robot shipping QNX in the safety VM (not just an integration or a demo). Until those land, treat robotics as the highest-ceiling, longest-fuse part of the thesis.
The full BlackBerry/QNX thesis (the growth models, the certification moat, and the honest “10× bigger” test) lives in the pillar: QNX: The Quiet Operating System Powering the AI Age. Why that layer is so hard to dislodge once it’s in: The Moat Is Made of Paperwork.
The 🍌🐀 has spoken. 🍌🐀
Sources
[1] Humanoid market forecasts (wide dispersion): Goldman Sachs ~$38B TAM by 2035 (~1.4M units); Morgan Stanley ~$5T ecosystem and >1B humanoids by 2050 (~13M in service by 2035), adoption “relatively slow” until 2035; Bank of America (Physical AI pt.2, Mar 2026) ~90K units 2026 → ~1.2M 2030 → ~10M 2035, ~3B installed by 2060; current humanoid market ~$2.9–3.1B in 2025 (MarketsandMarkets). Goldman Sachs Insights; Morgan Stanley / CNBC, Apr 2025; BofA Institute / Fortune, Mar 13 2026; MarketsandMarkets. https://www.morganstanley.com/insights/articles/humanoid-robot-market-5-trillion-by-2050
[2] Autonomous mobile robot (AMR) market ~$4.5–5.3B in 2025, ~17–22% CAGR to 2030–35 (Grand View, Mordor, Intel Market Research, Business Research Insights). The nearer-term, real-revenue robotics category. Grand View Research; Mordor Intelligence, 2025–26. https://www.grandviewresearch.com/industry-analysis/autonomous-mobile-robots-market
[3] NVIDIA robot-brain consolidation: Jetson Thor (2,070 FP4 TFLOPS, 128GB, 40–130W, Blackwell GPU, integrated functional-safety processor); Isaac GR00T partner roster includes 1X, Agility Robotics, Apptronik, Boston Dynamics, Figure AI, Fourier, Sanctuary AI, Unitree, XPENG (Tesla runs its own compute). NVIDIA Newsroom + Jetson Thor page, 2025–26. https://www.nvidia.com/en-us/autonomous-machines/embedded-systems/jetson-thor/
[4] Robot functional-safety standards: ISO 10218-1/-2:2025 (industrial robots/cells; application of IEC 61508, safety functions up to SIL 3), ISO 13482 (personal-care robots, 2014, revision in progress), IEC 61508 (umbrella, SIL 1–4), ISO/TS 15066 (collaborative-robot speed/force limiting). No finalized standard yet for dynamically walking humanoids (ISO 25785 in development); fenceless/unfenced certification remains the open deployment barrier. ISO; A3/Automate; 2025–26. https://www.iso.org/standard/73934.html
[5] NVIDIA Halos for Robotics, announced Jun 22 2026 (Automate, Chicago): “the industry’s first full-stack safety system for physical AI,” with a dedicated functional-safety island (isolated I/O/power/clocks) partitioned from main compute. Ships Linux-only or Linux + QNX, where the NV Hypervisor partitions IGX into a Linux VM (AI/apps) and a QNX VM (safety-critical). NVIDIA: QNX is “a real-time operating system with a long pedigree in certified safety systems… its inclusion enables stronger software partitioning for higher safety integrity use cases.” Agility Robotics is the first named adopter; QNX is one of several software partners. NVIDIA Newsroom + developer blog, Jun 22 2026. https://developer.nvidia.com/blog/inside-nvidia-halos-for-robotics-a-full-stack-functional-safety-system-for-physical-ai/
[6] QNX OS for Safety 8.0 integrated with NVIDIA IGX Thor for safety-critical edge AI (robotics, medical, industrial); Early Access; non-exclusive; no disclosed design wins or revenue. QNX/NVIDIA press release, Apr 20 2026. https://www.stocktitan.net/news/BB/qnx-and-nvidia-deepen-collaboration-to-advance-safety-critical-edge-2fpnzowdrzx3.html
[7] GEM ≈ 20% of QNX revenue with a TAM management believes could exceed automotive (CEO John Giamatteo, Q4 FY2026 call, Jun 2 2026); QNX Q4 FY26 record revenue $78.7M (+20% YoY), royalty backlog ~$950M; FY27 QNX guidance $290–307M. BlackBerry not breaking out GEM as a separate reported line (management commentary). Motley Fool / StockTitan, Jun 2 2026. https://www.stocktitan.net/news/BB/black-berry-reports-fourth-quarter-and-full-fiscal-year-2026-na3inuqf5na8.html
[8] Kinova “KIMA” medical robotic arm: Robot Control Library natively compatible with QNX OS 8.0, both to IEC 62304 Class C; QNX/Kinova report a 12–18-month development-time saving from reusing pre-certification. KIMA launched Jun 9 2026 (Society of Robotic Surgery). Kinova Robotics / PR Newswire, Jun 2026. https://www.kinovarobotics.com/medical/resources/building-the-future-of-medical-robotics-together-how-kinova-kima-and-qnx-are-accelerating-medical-robotics-time-to-market
[9] Johnson & Johnson selected QNX OS for Safety to power a new AI-driven heart pump (a landed design win; “will power,” not yet shipping); plus a QNX General Embedded Development Platform (GEDP) win with a major North American industrial OEM (unnamed). BlackBerry Q4 FY2026 earnings call, Jun 2 2026 (Motley Fool transcript).
[10] Medha Servo Drives’ communications-based train control for India’s metro/monorail runs on QNX OS for Safety (pre-certified IEC 61508 SIL 3, supporting an EN 50128 SIL 4 signaling target). QNX knowledge hub; CIOL, 2026. https://qnx.software/en/resources-knowledge-hub/insights/how-qnx-tech-powers-advanced-rail-safety-systems-in-india
[11] Ecosystem credentials: Arm named QNX a foundational software partner at its physical-AI CPU launch (CEO Rene Haas); ABI Research “Commercial RTOS for Robotics Functional Safety” ranking (Apr 21 2026) named QNX, Wind River, SYSGO and Green Hills “Leaders,” with QNX top-ranked overall; Apex.AI’s ROS 2-based framework compatible with QNX SDP 8.0 (Dec 2025). ABI Research via GlobeNewswire, Apr 21 2026; Arm; Apex.AI. https://www.globenewswire.com/news-release/2026/04/21/3277943/0/en/QNX-Wind-River-SYSGO-and-Green-Hills-Software-Named-Leaders-in-ABI-Research-s-Commercial-Real-Time-Operating-System-RTOS-for-Robotics-Functional-Safety-Ranking.html
0 comments