The BlackBerry You Don’t See: Understanding QNX and Its Role in Physical AI
Notes by SDG ·
Most of us remember BlackBerry as the phone with the tiny keyboard that made every email feel urgent.
That business is gone. But buried inside BlackBerry is a 45-year-old software platform called QNX that now runs inside more than 275 million vehicles. It is also used in medical equipment, factories, robots, rail systems, and aerospace and defence applications.
So BlackBerry did not entirely disappear. It just moved somewhere less visible: underneath the dashboard.
I wanted to understand whether QNX is simply a durable automotive software business—or whether it could become an important layer in the emerging world of autonomous machines and physical AI.
The short version: QNX is not the intelligence inside an AI machine. It is the trusted operating foundation that helps the intelligence interact with the physical world safely and on time. That is potentially a valuable position. But it is not an uncontested one, and the financial opportunity needs more scrutiny than the phrase “275 million vehicles” suggests.
First: what exactly is QNX?
QNX is a real-time operating system, or RTOS, for embedded computers.
An embedded computer is a computer built into another product. Your laptop is a general-purpose computer. The computer controlling a car’s braking system, an industrial robot’s arm, or a medical pump is embedded inside the machine and has a specific job.
The operating system coordinates the hardware and software. But in safety-critical equipment, being generally fast is not enough. It must also be predictable.
If Netflix takes an extra half-second to load, civilization continues. If software takes an extra half-second to apply a brake or stop a surgical robot, the product team has a more complicated meeting.
That is what “real time” means here. It does not simply mean fast. It means that a task is guaranteed to run within a defined time limit. QNX calls this deterministic execution.
QNX also uses a microkernel architecture. In a traditional monolithic operating system, many services run together inside the kernel. A fault in one component can affect a much larger part of the system. QNX keeps only essential functions in the kernel and runs more services separately. If one service fails, the damage can be isolated instead of taking down the whole machine.
An imperfect but useful analogy:
● A monolithic operating system can resemble an open-plan office. One incident with the microwave affects everyone.
● A microkernel resembles a building with fire-rated rooms. A failure in one room is less likely to spread through the entire structure.
QNX’s newest platform, Software Development Platform 8.0, is designed for modern high-performance processors with two to 64 CPU cores while maintaining hard real-time behaviour. The product family includes:
● the QNX real-time operating system;
● QNX OS for Safety, with pre-certified components for regulated systems;
● the QNX Hypervisor, which separates multiple operating environments on one chip;
● development, debugging, security, audio, and middleware tools; and
● engineering and lifecycle support.
The certification matters. A carmaker or medical-device company still has to certify its complete product, but using a pre-certified operating system and qualified tools can reduce the work, cost, and risk involved. QNX is partly selling software and partly selling the answer to the question, “Do we really want to build and certify this layer ourselves?”
Usually, the answer is no. Particularly when steering is involved.
Where does QNX sit inside a modern car?
A modern vehicle may contain many electronic control units responsible for braking, steering, driver assistance, infotainment, the instrument cluster, battery management, audio, and other functions.
The industry is now consolidating some of those separate computers into fewer, more powerful centralized processors. This is part of the shift toward the software-defined vehicle, where more of a car’s features and behaviour are controlled—and updated—through software.
QNX can sit at the foundational layer, directly above the processor. Its hypervisor can then divide the hardware into isolated environments. For example:
● QNX can run the safety-critical driving or instrument-cluster functions;
● Android or Linux can run a richer infotainment experience; and
● a fault or restart in the entertainment system can be isolated from the safety system.
This is why comparing QNX with Android Automotive is sometimes misleading. They can compete for parts of the vehicle software stack, especially as Google expands Android Automotive beyond infotainment. But they can also coexist: QNX explicitly supports Android and Linux as isolated guest operating systems inside its hypervisor.
In other words, the car does not necessarily choose QNX or Android. It may use QNX underneath Android. The operating-system family tree inside a vehicle is apparently allowed to be as confusing as the trim packages.
How QNX makes money
The revenue model has two main stages.
During development, a manufacturer buys software-development licences, tools, support, maintenance, and engineering services. These sales arrive while a vehicle or device is being designed.
During production, BlackBerry earns royalties based on the number of units shipped, although some arrangements can include minimum-volume commitments or fixed royalties.
This creates a long delay between winning the business and recognizing much of the revenue. Automotive programs can take years to move from design to production, then remain in production for years. A design win today may therefore generate royalties well into the next decade.
That delay is both attractive and annoying:
● Attractive because being designed into a safety-critical platform creates switching costs and a long revenue tail.
● Annoying because investors must wait for the customer’s product to ship—and hope its sales forecast was not written during an unusually optimistic afternoon.
BlackBerry reports a QNX royalty backlog, which estimates future variable royalties from existing designs. It reached approximately $950 million at the end of fiscal 2026, up from $865 million a year earlier. But this is not contracted revenue in the same sense as a firm purchase order. BlackBerry explicitly says the estimate depends on projected production volumes and can change if programs or contracts are modified or cancelled.
The more useful signals to watch together are:
1. new design wins;
2. development-licence revenue, because engineers buy tools before products reach production;
3. royalty-backlog growth;
4. actual royalty revenue as customer products ship; and
5. revenue earned per vehicle or device as more QNX components are used.
The business today
QNX has become BlackBerry’s main growth engine.
In fiscal 2026, QNX generated $268 million in revenue, up 14%, with an adjusted gross margin of 83% and adjusted EBITDA of $71 million. In the first quarter of fiscal 2027, revenue rose 26% year over year to $72.3 million, adjusted gross margin reached 86%, and segment adjusted EBITDA was $19.3 million.
BlackBerry subsequently guided to $295 million to $312 million of QNX revenue for fiscal 2027. That would represent roughly 10% to 16% growth over fiscal 2026.
The latest quarter was helped by both higher royalties and an increase in development-licence revenue. The latter is worth watching because it can indicate that new programs are moving into development before generating production royalties.
QNX says its software is in more than 275 million vehicles, up 20 million in a year and 100 million since 2020. It works with more than 45 automakers, including the top ten global OEMs and 24 of the top 25 EV manufacturers. Recent disclosed activity includes BMW’s next-generation Neue Klasse architecture, Mercedes-Benz testing the Alloy Kore platform, NVIDIA DRIVE AGX Thor integration, Volvo’s EX60 audio platform, Leapmotor vehicles, and WeRide’s L2++ driver-assistance platform.
Those are strong signs of relevance. Still, the 275-million figure is an installed base accumulated over many years, not 275 million subscriptions paying BlackBerry every year. QNX is primarily monetized through development licences and per-unit production royalties. The number proves reach and credibility; it should not be multiplied by an invented annual subscription fee and presented as destiny.
So where does AI enter the story?
AI software can perceive, reason, predict, or decide. A camera model might identify a pedestrian. A planning model might decide that a robot should stop or move around an obstacle.
But an AI prediction is not a physical action.
Between “pedestrian detected” and “apply the brake,” a machine still needs software that schedules workloads, controls access to hardware, isolates failures, handles sensors and communications, and ensures that critical commands execute within deadlines.
QNX operates in this less glamorous but essential layer.
A simplified physical-AI stack looks like this:
1. Sensors collect images, sound, position, force, or temperature.
2. AI models interpret the environment and choose an action.
3. The operating system and middleware allocate compute, isolate workloads, and communicate with the hardware.
4. Actuators steer, brake, move a robotic arm, regulate a pump, or perform another physical action.
QNX is mainly in step three. It is not trying to out-reason an NVIDIA model. It is helping the machine execute software predictably on NVIDIA, Qualcomm, AMD, and other processors.
This role becomes more important as machines consolidate workloads onto powerful chips. A single processor may simultaneously run AI perception, safety controls, diagnostics, a user interface, and connectivity. The QNX Hypervisor can separate these workloads so that a failure in a non-critical application does not interfere with a safety-certified function.
BlackBerry is already positioning QNX beyond cars for industrial robots, warehouse systems, agricultural machines, medical devices, drones, aerospace, and defence. Approximately 20% of QNX revenue came from non-automotive embedded systems in fiscal 2026. That is enough to show the opportunity is real, but also small enough to show that the broader physical-AI thesis is still developing.
Why the future opportunity could be larger
There are four potential growth levers.
1. More software per vehicle
The simplest opportunity is not necessarily selling QNX into more cars. It is selling more QNX into each car.
Historically, QNX might have powered infotainment or an instrument cluster. New centralized architectures create opportunities across digital cockpits, advanced driver assistance, gateways, audio, and central compute.
BlackBerry and Vector’s new Alloy Kore platform combines QNX’s operating system and virtualization with Vector’s automotive middleware. BlackBerry says it could increase the average selling price per vehicle by “multiples.” That is management’s expectation, not a result yet, but it identifies the strategy clearly: move from selling individual foundational components to selling more of a ready-made vehicle platform.
2. Higher-performance centralized computing
As automakers replace many small controllers with fewer powerful processors, the software foundation must manage more cores and mixed-criticality workloads. QNX SDP 8.0 was built for this transition, scaling up to 64 CPU cores.
Consolidation can reduce the number of individual operating-system instances, but it can also make the remaining platform much more valuable. The question is whether QNX captures that added value in its pricing.
3. Physical AI outside automotive
Robots and autonomous machines are beginning to inherit the problems cars have dealt with for years: real-time responsiveness, functional safety, long product lives, security, and over-the-air updates.
If physical AI moves from controlled demos into factories, warehouses, hospitals, farms, and public environments, manufacturers will need a certified software foundation. QNX has relevant technology and credibility. BlackBerry is introducing a General Embedded Development Platform to make QNX easier to adopt in these markets.
But this opportunity is not automatic. Robotics has a strong Linux and ROS ecosystem, and many manufacturers prioritize developer familiarity, flexibility, and cost before certification. QNX must make itself easier for developers to use, not merely safer for executives to approve.
Its QNX Everywhere initiative—free non-commercial access, online learning, VS Code tooling, and a growing collection of open-source ports—is an attempt to fix that ecosystem gap.
4. A long-lived royalty model
Once a safety-critical operating system is selected and certified, replacing it can be expensive and risky. That gives QNX a durable position within programs it wins. If development licences convert into successful high-volume products, royalties can compound across overlapping product generations.
The catch is that automotive royalty revenue depends on the customers’ production volumes. Winning the operating system for a vehicle that sells poorly is still a win, just a very polite one.
Who competes with QNX?
There is no single competitor across every use case.
|Competitor or alternative |Where it is strong |How it threatens QNX
|**Wind River VxWorks** |A commercial, deterministic RTOS with safety-certification support across aerospace, industrial, automotive, and medical systems|The closest broad commercial competitor for safety-critical embedded platforms |
|**Green Hills INTEGRITY** |Security, isolation, and ISO 26262 ASIL D-certified automotive systems |A direct alternative in highly secure and safety-critical designs |
|**Linux and Automotive Grade Linux** |Open source, huge developer ecosystem, flexibility, rapid innovation, and lower apparent licensing cost |OEMs may use Linux directly or build internal platforms, particularly for non-safety workloads |
|**Android Automotive OS** |Rich application ecosystem, infotainment, Google services, and an expanding software-defined vehicle platform |Google is moving deeper into vehicle functions, but Android can also run above a QNX hypervisor |
|**AUTOSAR Adaptive and commercial implementations**|Standardized architecture and middleware for high-performance automotive computers |Often complements an operating system rather than replacing it, but it shifts value into standardized middleware and reduces differentiation|
|**OEM and chip-vendor platforms** |Control, vertical integration, and optimization for proprietary hardware |Large automakers, suppliers, or semiconductor companies may build more of the stack internally or bundle an alternative with the chip |
The most important competitive threat may be neither VxWorks nor Green Hills. It may be the combination of open-source Linux, internal development, and large platform companies subsidizing software to sell something else.
BlackBerry acknowledges this directly. Its annual filing says some OEM and Tier 1 customers are developing embedded solutions internally and exploring open-source software perceived as free. It also warns that larger competitors can bundle products or sell at extremely low margins.
QNX’s answer is that open-source code is not the same as a certified, supported, deterministic platform with decades of lifecycle maintenance. For a music app, free software may be compelling. For a braking controller, “we found it on GitHub” has less boardroom charm.
Still, Linux is improving, safety-oriented projects are expanding, and Google is pushing Android further into the vehicle architecture. QNX cannot rely on history alone.
What is the moat?
QNX has several meaningful advantages:
● Trust and operating history: more than 275 million vehicles and decades in mission-critical systems.
● Safety certifications: these reduce the effort customers face in certifying complete systems.
● Deterministic microkernel design: fault isolation and predictable response are structural strengths.
● Automotive relationships: QNX works across automakers, Tier 1 suppliers, and major semiconductor platforms.
● Switching costs: replacing foundational software after development and certification begins is painful.
● Hardware neutrality: QNX can work across major processor architectures and allow Android or Linux to run alongside it.
But its weaker points are just as important:
● a smaller developer ecosystem than Linux or Android;
● dependence on long and unpredictable customer production cycles;
● historically modest revenue relative to its enormous installed base;
● pricing pressure from open source and bundled alternatives;
● concentration in automotive, which still produces about 80% of QNX revenue; and
● execution risk in turning new products such as Alloy Kore into meaningful sales.
What I would watch from here
For anyone evaluating BlackBerry through the QNX lens, I would track these signals rather than headlines about the old phone company:
Development-licence growth. It is an early indicator that manufacturers are beginning new QNX-based programs.
Royalty-backlog growth and quality. The backlog is useful, but its conversion depends on actual unit volumes. Management’s disclosure about timing and customer mix matters.
Royalty revenue. This confirms that old design wins are finally reaching production.
Revenue per vehicle or program. Adoption of the hypervisor, safety products, audio, middleware, and Alloy Kore could expand QNX’s content per device.
General Embedded Market wins. The physical-AI thesis becomes more credible if non-automotive revenue grows materially beyond its current 20% share.
Developer adoption. Downloads, training, ports, silicon support, and third-party applications will show whether QNX Everywhere is building an ecosystem rather than simply a nicer landing page.
Competition from Google, Linux, and in-house platforms. In particular, watch which layer each platform controls. An Android announcement is not automatically a QNX loss—but Android’s expansion into non-safety vehicle functions does increase the competitive overlap.
Actual Alloy Kore customers. Trials are encouraging. Production commitments, revenue, and higher content per vehicle would be much stronger evidence.
My conclusion
QNX is best understood as the quiet control layer beneath intelligent machines.
AI may decide what a car or robot should do. QNX helps ensure that the surrounding system performs the action reliably, isolates failures, and meets the deadlines required in the physical world.
That makes it relevant to AI—but in a second-order way. QNX does not become valuable simply because someone adds the words “physical AI” to an investor presentation. Its opportunity grows if three things happen together:
1. vehicles and machines consolidate more software onto powerful central computers;
2. those systems require certified, deterministic foundations; and
3. BlackBerry captures more revenue per platform while expanding beyond automotive.
The evidence so far is encouraging: accelerating QNX revenue, high gross margins, a growing royalty backlog, broad automotive adoption, new development-licence sales, and credible partnerships with automakers and NVIDIA.
The unanswered question is scale. QNX is technically important, but BlackBerry must still prove that technical importance can become a much larger business. The 275-million-vehicle footprint tells us QNX has earned trust. The next phase is about monetizing more of it—and carrying that trust from cars into robots and other autonomous machines.
That is a more interesting BlackBerry story than keyboards. Admittedly, the bar was not impossibly high.
Sources
● BlackBerry fiscal 2026 annual report
● BlackBerry Q1 fiscal 2027 quarterly report
● BlackBerry Q1 fiscal 2027 results
● QNX Software Development Platform 8.0
● QNX Hypervisor and Hypervisor for Safety
● QNX Everywhere developer initiative
● Wind River VxWorks Cert Edition
This is a technology and business analysis, not investment advice.