Modern edge platforms are absorbing workloads that used to live on separate boxes. The savings are real, but so is the bill, and most teams underestimate it.
Walk into any program review for an aerospace, defense, or automotive platform being designed today and you will hear some version of the same plan: take three or four boxes that used to live on separate hardware and consolidate them onto one. Flight controls and synthetic vision. ADAS and infotainment. Mission computing and AI inference. Real-time control and a Linux guest hosting the operator interface.
The business case is obvious. Less hardware. Less wiring. Less weight. Less power. Fewer suppliers to manage across the life of the program. It is the right direction, and it is happening regardless of whether the foundational software is ready for it.
That last part is the problem.
]
What "Consolidation" Costs
Putting a safety-critical workload and a general-purpose workload on the same processor is straightforward to draw on a slide. Doing it without compromising the safety case, the security posture, or the determinism guarantees that the program is built on, that is engineering that very few stacks were designed for.
We have started calling this the convergence tax: the hidden cost of forcing workloads to share hardware on a software foundation that was not built for mixed criticality.
It shows up as:
Consolidation without a foundation built for it is not consolidation. It is a deferred bill.
The instinct, when a platform misbehaves, is to add software on top: another monitor, another wrapper, another scheduler. That works for a release or two. It does not survive certification, and it does not survive the third or fourth generation of the platform.
The systems that age well are the ones whose isolation, determinism, and security properties are enforced at the layer beneath the operating systems, not bolted on above them. That is a separation kernel hypervisor doing real work: partitioning hardware against itself, scheduling workloads against a real-time clock, and presenting each guest with a view of the world it cannot escape from.
It is also a graphics stack that lets you use modern GPU hardware without throwing away the OpenGL code and certification artifacts that the program has been investing in for fifteen years. And it is a deterministic execution model for GPUs and accelerators, so AI and perception workloads can move to the edge without breaking the timing guarantees the rest of the system depends on.
The Argument We Are Making
Our patent portfolio is not a marketing artifact. It is a description, in formal language, of how we have spent years solving exactly these problems: separation kernel architecture, secure domain isolation, hypervisor-level security, OpenGL-on-Vulkan, shader emulation for legacy pipelines, and time-bounded execution for heterogeneous compute.
Programs being designed today will fly, drive, or operate for the next twenty to thirty years. The foundation underneath them needs to be that durable. We wrote a white paper laying out exactly how our IP supports that durability — and what it means for the systems being built on top of it.
Advancing Mission-Critical Edge Systems
A deeper look at the IP, architecture, and engineering decisions behind the next generation of secure, certifiable edge platforms.