How flexible, deterministic silicon frees OEMs from the hardware cycle in the age of AI
For most of the history of embedded systems, there has been a quiet bargain between hardware and software. Hardware defined what a product could ever become. Software kept it running, patched its defects, and – within narrow limits – tuned its behaviour. The real innovation happened in the next silicon cycle, three to five years away, gated by board redesigns, certification, and supply chains. If you wanted your product to do something genuinely new, you waited for the next one.
That bargain is collapsing, and AI is the force pulling it apart. Electric vehicles gain range and features years after they roll off the line. Smart speakers learn new skills in the customer’s home. Industrial controllers pick up new protocols and on-device intelligence without a truck roll. These are not curiosities. They are the leading edge of a structural shift: deployed hardware is becoming a living product, a base on which new capability, new customer value, and new revenue can be delivered continuously, for the life of the device.
AI is rewriting the mindset of product design
The pace has changed, and it has changed permanently. The multi-year proof-of-concept, followed by the multi-year development programme, followed by the cautious launch, is no longer a competitive posture — it is an accelerating route to obsolescence. AI models that were state-of-the-art at the start of a project are routinely superseded before that project ships. OEMs moving at the old cadence now face the real prospect of their product being obsolete at launch.
The winners are moving differently. They are compressing development cycles, shipping earlier, and treating the first release as the start of the roadmap rather than the end of it. They understand that in a market where capability is defined by the latest model, the latest algorithm, the latest protocol, the product that keeps pace is the product that keeps selling.
This reshapes where differentiation lives. Base hardware is no longer the enduring source of advantage, it is the foundation on which advantage is built and rebuilt throughout the lifecycle, through software, configurable hardware behaviour, and adaptable SKUs with rapid time-to-market. Product life cycles will shorten. The cadence of meaningful change, in the customer’s hands, will accelerate.
From product cycles to living products
When a product keeps growing in the customer’s hands, the commercial model around it changes. Revenue is no longer bounded by the moment of sale. Features can be unlocked as subscriptions, added as tiers, or released as free upgrades that deepen loyalty and justify a premium at the next purchase. Differentiation compounds rather than decays, a product shipped eighteen months ago is more capable today than it was at launch, not less.
Equally important, the internal tempo of the business changes. Time-to-market for new features decouples from the silicon roadmap. A new AI model, a new codec, a new safety function can reach customers in weeks rather than waiting for the next SoC generation. Roadmaps start to look less like a series of monolithic launches and more like the rolling release cadence that cloud software teams have enjoyed for the last decade. Innovation velocity becomes a function of the engineering team, not the constraints of deployed product.
What makes a living product possible
Turning deployed hardware into a platform for continuous innovation requires silicon with three properties that have historically been in tension:
- Flexibility — the ability to host new functionality on hardware already in the field. New AI workloads, new interfaces, new algorithms. No board redesign, no waiting for the next generation.
- Determinism — the guarantee that anything deployed to the field behaves identically on every unit, meets its timing requirements, and leaves existing behaviour untouched.
- Cost-effective reuse — the ability to serve a range of products and SKUs from the same underlying platform, so that investment in flexibility compounds across the portfolio rather than being rebuilt for every new design.
The era of squeezing the last penny out of a bespoke BOM for a single SKU is ending. In a market defined by AI-driven feature velocity, that strategy optimises for the wrong variable. The OEMs who win will be the ones running cost-effective, flexible architectures that can quickly absorb new capabilities and serve a multitude of products from a common foundation. Reuse beats bespoke when the ground keeps moving.
Flexibility without determinism is a dead end. it produces a platform that can accept upgrades but cannot ship them with confidence, and the organisation eventually learns not to touch production firmware. Determinism without flexibility is a different dead end. a stable system frozen on the day it leaves the factory. The combination, delivered on a cost-effective platform that spans multiple products, is what makes a living product commercially viable at scale.
The XCORE® approach
XCORE was designed from first principles around this combination. Three properties matter most for OEMs building living products in an AI-driven market:
- True hardware parallelism. Independent processing resources execute concurrently on dedicated hardware. New AI models and new functionality can be added without competing with existing tasks for a shared execution unit.
- Static resource allocation. Memory, I/O, and compute are assigned at compile time. New modules arrive with guaranteed resources; existing modules keep guaranteed isolation. Nothing contends for anything at runtime.
- Deterministic execution. Timing is a specifiable, verifiable property of the design. What you measure in simulation is what runs on the device, in the customer’s home or factory, every time.
The practical consequence is straightforward. An OEM can extend a deployed product’s capabilities, a new on-device AI model, a richer audio algorithm, a new sensor fusion routine, an additional communication protocol and know, with engineering confidence rather than statistical hope, that the upgrade will land cleanly on every unit in the field. And because the same architecture scales across multiple SKUs and product lines, that investment pays back across the portfolio rather than a single product.
What this unlocks
When deployed hardware is treated as a living product, the strategic options available to an OEM widen significantly. Products can launch earlier, with a deliberate plan to grow into their roadmap in the field, rather than waiting for “complete” hardware that is already a generation behind. Teams can respond to new AI capabilities, new standards, and shifting customer needs in software time rather than silicon time. Recurring revenue becomes viable on products that were previously a one-time sale.
Differentiation is no longer just the specification at launch, it is the pace of improvement afterwards, which is far harder for competitors to match.
A different question at silicon selection
For most of the history of embedded systems, the question at silicon selection has been: what can this part do on day one, at the lowest possible unit cost? It is the right question for a product that will never change, in a market where the pace of innovation is slow. It is the wrong question for a living product in a market reshaped by AI.
The more valuable question now is: what can this part become, in the hands of our customers, across our portfolio, over the next ten years? Flexible, deterministic, parallel silicon is what makes that question answerable. OEMs who design around it stop shipping products and start shipping platforms, and the innovation cycle stops being set by the hardware, and starts being set by the ambition of the team behind it.
