A View from the Watchtower

ARM just made its move. See my blog from last week on this profound step:  A View from the Watchtower: ARM - From Architecture to Platform

ARM stepped up the stack — from architecture to platform — signaling that the future of AI infrastructure will be defined at the system level, not just at the chip level. But at the same time, something equally important is happening beneath it. The industry is beginning to ask a different question: What if no one should own the blueprint at all?

The new reality: multiple architectures, one economic driver

We are entering a phase of the semiconductor industry where multiple architectural models are not just coexisting — they are all accelerating at once. ARM is moving up the stack into platform ownership. RISC-V is expanding as a sovereignty-driven alternative. x86 (Intel and AMD) continues to anchor enterprise and cloud infrastructure. And NVIDIA has effectively defined the modern AI system architecture — where GPUs, interconnect, and software form the center of gravity. But this is not fragmentation. This is convergence under a single economic force: AI.

AI is not just another workload. It is the defining economic driver of the next decade:

• It determines infrastructure investment
• It reshapes data center architecture
• It drives power, performance, and system-level constraints
• It dictates how hardware and software co-evolve

And in that world, no single architecture is sufficient. Instead, we are seeing a new equilibrium emerge:

• ARM as the efficiency-driven platform orchestrator
• x86 as the legacy and enterprise backbone
• NVIDIA as the AI system layer
• RISC-V as the sovereignty and customization layer

Different roles. Same system.

The system we are actually building

Imagine the system now being deployed in a modern AI data center. It is no longer a single architecture. It is a composition:

• NVIDIA GPUs running the AI training and inference stack
• x86 CPUs managing legacy workloads, virtualization, and enterprise software
• ARM-based processors orchestrating efficiency, scheduling, and system services
• RISC-V components emerging for custom control, sovereignty, or domain-specific acceleration
• Chiplets from multiple vendors, integrated through advanced packaging and high-speed interconnect

Each of these elements comes with its own:

• Instruction set
• Software stack
• Toolchain
• Validation methodology
• Support model

Individually, they work. But together, they form a system that no single vendor fully owns — and no traditional design flow was built to manage. So the question is no longer:

Can we design each component correctly?

The question becomes:

How do we coordinate the integration, validate the full system behavior, and support it over time — across architectures, vendors, and geopolitical boundaries?

That is a system problem.

The original model: trust the blueprint

For decades, the semiconductor industry operated on a simple premise: You could trust the foundation. ARM became the most successful embodiment of that idea:

• A shared instruction set
• A licensed architecture
• A global ecosystem built on consistency

It wasn’t just technology. It was institutional trust. Design teams didn’t need to question the ISA. They didn’t need to validate the foundation. They could build. And that model scaled to hundreds of billions of devices.

Does RISC-V change the premise?

RISC-V doesn’t just introduce a new architecture. It removes the idea that architecture must be owned.

• Open ISA
• No licensing dependency
• Custom extensions
• Local control

At first, this looked like a cost play. It isn’t. It’s a control play. Viewing recent announcements from the Watchtower makes this very clear.

Sovereignty is now a design requirement

China, RISC-V, and the geopolitical accelerant

If sovereignty is becoming a design requirement, China is the clearest example of why. Export controls and access restrictions have forced a fundamental shift in strategy: Dependency on foreign-controlled architectures is no longer acceptable. RISC-V provides a path forward:

• No licensing dependency
• Full control over extensions
• Alignment with national semiconductor initiatives
• A foundation for building domestic AI infrastructure

This is not a theoretical shift. It is already happening: Chinese hyperscalers and semiconductor companies are actively investing in RISC-V-based designs — including AI-oriented processors — as part of a broader push toward technological independence. And this has ripple effects across the entire industry:

• ARM faces pressure as a licensed architecture in geopolitically sensitive markets
• Intel and AMD remain constrained by export controls and policy decisions
• NVIDIA sits at the center of AI demand, but is directly exposed to geopolitical regulation
• RISC-V becomes the only architecture that can be freely adopted, modified, and deployed without external approval

This doesn’t mean RISC-V replaces ARM or x86. It means the world is no longer operating under a single trusted blueprint. RISC-V is gaining traction not because it’s “better” in a traditional sense. It’s gaining traction because the world has changed.

• Export controls are real
• Supply chains are political
• AI infrastructure is strategic
• National and corporate autonomy matter

In that environment, dependency becomes risk. And RISC-V offers something fundamentally different: The ability to build without permission.

But openness introduces a new problem

If ARM’s model was built on centralized trust, then RISC-V’s model introduces distributed responsibility. Anyone can implement it. Anyone can extend it. Anyone can modify it. That is certainly a powerful concept. But it creates a new question: Who validates the system?

Provenance becomes the new fault line

In a world of AI systems and chiplet-based architectures, this question gets sharper:

• Where did this IP come from?
• Who verified it?
• What assumptions were made?
• How does it behave under real workloads?
• Can I trust it in production?

This is not just about design. This is about provenance. The lineage of the system. The traceability of decisions. The integrity of the lifecycle. And as systems become more distributed, provenance becomes harder to establish — and more important to prove.

This is not a future scenario. This is the system being built today.

The paradox

We are now entering a fascinating world with two competing models:

ARM

• Centralized
• Trusted
• Controlled
• Integrated

RISC-V

• Open
• Flexible
• Sovereign
• Fragmented

Each solves one problem. Neither solves the whole system. Because the real challenge has shifted.

The problem is no longer the architecture

For decades, the industry focused on:

What architecture should we build on?

Now the question is different:

How do we coordinate and validate systems built across multiple architectures, vendors, and organizations?

Because modern systems are:

• Multi-die
• Multi-vendor
• Multi-layer
• Continuously evolving

And increasingly:

• Geopolitically constrained

This is where EDA 3.0 becomes unavoidable

This is not an ISA problem. This is not a traditional EDA problem. This is a coordination and traceability challenge at the system scale. EDA 1.0 enabled design. EDA 2.0 optimized the flow. EDA 3.0 begins when:

The system — not the chip — becomes the unit of design
And the ecosystem — not the company — becomes the unit of execution

In that world, what matters is not just:

• What was designed
• But how it was defined
• How it was verified
• How it was integrated
• And how it behaves in reality

The next layer is not about control — it’s about trust

If ARM represents control through ownership, and RISC-V represents freedom through openness, then the next layer must solve something different: Trust across independence. Because in a heterogeneous system:

• You don’t control every component
• You don’t own every interface
• You don’t build every layer

But you are still responsible for the outcome.

The next blueprint isn’t an ISA

The industry’s first great blueprint was architectural. It allowed the world to scale compute. The next blueprint will be different. It won’t define instructions. It will define ecosystem coordination.

Final thought

ARM’s move up the stack signals a world where platforms matter more. RISC-V’s rise signals a world where control is being redistributed. Together, they point to the same conclusion: The semiconductor industry is no longer just designing technology. It is managing trust at scale. And in a world of open architectures, distributed systems, and geopolitical pressure:

Trust no longer comes from ownership.
It comes from traceability.

That is both the shift and the gap. And that is where the next era of semiconductor design — EDA 3.0 — will be defined.