# The One Decision That Determines Every Chip Architecture

An episode of [The Hardware Podcast with Fexingo: Chips, Devices, and Electronics Engineering Conversations](https://invisiblenetwork.ai/podcast/1896799982), 5 Jun 2026, 11 min.

Audio: https://audio.fexingo.com/business/the-hardware-podcast/episode-0032.mp3

Every chip design starts with a single high-stakes choice: which instruction set architecture to build on. This episode of The Hardware Podcast with Fexingo walks through why that decision shapes everything from performance to software ecosystem, with a close look at ARM's rise against x86 in data centers and how RISC-V is shaking up the landscape. Lucas and Luna break down the trade-offs between fixed ISA and custom extensions, the role of backward compatibility, and why Apple's M-series chips were a pivotal moment for ARM. They also explore the hidden costs of switching architectures and why startups are betting on RISC-V for specialized workloads. If you're curious about what gives a chip its DNA, this one's for you. #InstructionSetArchitecture #ARMvsx86 #RISCV #ChipDesign #AppleSilicon #DataCenterChips #HardwareEngineering #Technology #FexingoBusiness #BusinessPodcast #TechPodcast #Semiconductors #CPUArchitecture #SoftwareEcosystem #BackwardCompatibility #CustomExtensions #M1Chip #AmpereComputing Keep every episode free: buymeacoffee.com/fexingo

## Transcript

Lucas: So you're designing a new chip. You've got the budget, the team, a clear performance target. You've maybe even sketched out the pipeline. But before any of that matters, you've got to answer one question that locks you into a decade of consequences: which instruction set architecture do you build on?
Luna: Right — the ISA. It's the contract between the hardware and the software. Every program that runs on the chip speaks that language.
Lucas: Exactly. And once you commit, you're not just choosing a technical spec. You're choosing an ecosystem, a licensing model, a talent pool, and a compatibility burden that shapes every follow-on chip you'll ever make. It's arguably the single most consequential decision in chip design.
Luna: And it's not just a theoretical choice. Look at the data center over the last five years. ARM went from basically zero server share to powering some of the biggest cloud instances from AWS, with their Graviton chips, and now Ampere and others.
Lucas: The shift is real. arm based servers are projected to account for over 20 percent of the data center market by 2028, according to Omdia. That's up from virtually nothing in 2018. And it's not just AWS — Google and Microsoft are designing their own ARM chips too.
Luna: Before we dive deeper — we should mention that this show is ad-free by design. If you get value from episodes like this, and you want to support the work that goes into them, there is a simple way. It's buy me a coffee dot com slash fexingo. No pressure, just a way to keep the conversation going without commercial breaks.
Lucas: Absolutely. It's listeners like you who make this possible. So back to ISA decisions — what's the core trade-off between ARM and x86?
Lucas: The textbook answer is complexity versus efficiency. x86 is a complex instruction set — each instruction can do more work, but decoding is heavier, power consumption is higher. ARM is a reduced instruction set — simpler instructions, easier to pipeline, lower power. But that's not the whole story anymore.
Luna: Because modern ARM chips aren't exactly RISC in the pure sense. They've added extensions, SIMD, cryptographic instructions. And x86 has gotten micro-op translation and aggressive power management. The lines have blurred.
Lucas: Right. The real differentiator today is the ecosystem and the licensing model. x86 is essentially a duopoly — Intel and AMD. They own the ISA, and they're the only ones who can build high-performance x86 cores. ARM, on the other hand, licenses its architecture to anyone. That opened the door for Apple, Amazon, Qualcomm, and a wave of startups.
Luna: Apple's M1 was the watershed moment. It proved that an arm based chip could outperform x86 on both performance and power efficiency in a laptop. That changed the perception overnight.
Lucas: Absolutely. Before the M1, ARM was seen as a mobile architecture — great for phones, not serious for high-performance computing. Apple showed that with enough design investment, ARM could beat Intel at its own game. And that sent a signal to the entire industry.
Luna: But the decision isn't just about performance per watt. It's about software compatibility. If you choose a new ISA, you're asking every software vendor to recompile, potentially rewrite, or emulate. That's a huge friction.
Lucas: That's the central tension. The ISA determines what software can run natively. x86 has a massive installed base — decades of binaries, libraries, operating systems. Switching to ARM means running through emulation layers like Rosetta, which works but adds overhead. For a startup designing a chip for a specific workload, like AI inference, you might decide compatibility doesn't matter because you control the software stack.
Luna: And that's exactly where risc v comes in. It's an open-source ISA — no licensing fees, no single vendor control. Anyone can design a risc v core and extend it with custom instructions for their specific use case.
Lucas: risc v is fascinating because it flips the model. Instead of choosing between two proprietary ISAs, you get a base that's free and extensible. But that flexibility comes with fragmentation risk. If every chip has different custom extensions, you lose the portability that made ARM and x86 valuable.
Luna: So it's a classic trade-off: control and optimization versus ecosystem and compatibility. For a specialized chip — say, a neural network accelerator for a self-driving car — risc v makes sense. For a general-purpose processor that needs to run Windows or Linux applications, you probably stick with x86 or ARM.
Lucas: Exactly. And there's another layer: the instruction set itself influences the microarchitecture decisions. For example, x86's variable-length instructions make decoding complex, which eats power and area. ARM's fixed-length instructions simplify the decoder, freeing up transistors for more execution units. That's part of why ARM cores can be smaller and more power-efficient.
Luna: But fixed-length also means you might need more instructions to do the same work. So it's not a free lunch. You trade instruction density for decoding simplicity.
Lucas: Right. And the density matters in memory-bound workloads. If your code is fetching more instructions, you need more cache bandwidth, more memory bandwidth. So the ISA choice ripples into the entire memory hierarchy design.
Luna: I want to talk about backward compatibility. x86 is famous for supporting instructions that are decades old. That's a blessing and a curse. It means old software runs, but it also means the architecture carries legacy baggage that bloats the decoder and complicates the pipeline.
Lucas: That's a huge point. Intel and AMD have to maintain compatibility with every x86 instruction ever released. That's thousands of instructions, many of which are rarely used. ARM has a cleaner break — they've deprecated instructions over time. And risc v starts from a clean slate, with a minimal base set that's only about 50 instructions.
Luna: So if you're designing a chip today, which ISA would you choose? And does it depend on whether you're building for a phone, a laptop, a server, or an embedded device?
Lucas: Great question. For embedded and IoT, ARM and risc v are the dominant players. ARM has the ecosystem, risc v offers zero licensing cost. For smartphones, ARM is essentially the only choice — the entire Android ecosystem is built on it. For laptops, ARM is proving itself with Apple, but x86 still dominates the Windows world. For servers, x86 is still the incumbent, but ARM is growing fast, especially in cloud-native workloads where power efficiency matters.
Luna: And for a startup with a radical new idea? Like a chip designed specifically for graph analytics or for a new kind of memory architecture?
Lucas: For a truly novel workload, risc v is tempting because you can customize the instruction set to your exact needs. But you have to build the software ecosystem from scratch. Some startups are hybrid — they use risc v for the core and add a custom accelerator. Others, like Tenstorrent, are building risc v based AI chips with their own extensions.
Luna: There's also the question of talent. If you design an x86 chip, you need engineers who understand that complex microarchitecture. ARM and risc v have a wider pool of engineers from the mobile and embedded worlds. But finding people who can design a high-performance out-of-order risc v core is still rare.
Lucas: That's a practical constraint that often gets overlooked. The team's experience with a particular ISA can make or break the project timeline. If everyone on the team has worked on ARM cores for a decade, switching to a new ISA might add months of learning curve.
Luna: So the decision comes down to a set of interconnected factors: performance targets, power budget, software requirements, licensing costs, talent availability, and timeline. No single ISA wins on all fronts.
Lucas: Exactly. And that's why we see so many different architectures coexisting. The market is segmenting. Hyperscalers like Amazon and Google are building their own ARM chips because they control the software stack and can optimize for their specific workloads. Meanwhile, Intel is fighting back with efficiency cores and new packaging technologies.
Luna: And then there's the x86 emulation of ARM — like Microsoft building x86 emulation for Windows on ARM. That blurs the lines even further. If emulation becomes good enough, the ISA might matter less for software compatibility.
Lucas: That's the long-term trend. As emulation and binary translation improve, the ISA becomes more of an implementation detail. What will matter more is the microarchitecture — the actual pipeline design, the cache hierarchy, the memory subsystem. The ISA is just the language; the real art is how you execute it.
Luna: So the takeaway for our listeners: if you're evaluating a new chip — whether you're buying it or building it — don't just look at the ISA. Look at the whole package. The ISA tells you what software runs; the microarchitecture tells you how fast it runs.
Lucas: And the decision of which ISA to build on is a bet on the future of that ecosystem. It's not just a technical choice — it's a strategic one. And as we've seen with ARM's rise in the data center, those bets can pay off in ways nobody predicted a decade ago.
Luna: So what comes next? Could risc v disrupt both ARM and x86 in the next ten years, or will it remain a niche for specialized workloads?
Lucas: I think it's going to grow significantly, especially in areas where customization matters: AI accelerators, edge devices, maybe even in some server roles. But to unseat ARM or x86 in general-purpose computing, it needs a software ecosystem that matches Windows, macOS, and Linux. That's a multi billion dollar investment that no single company is making. So for now, the ISA landscape is a three-horse race, and each horse has its own strengths.
Luna: Well, we'll be watching. Thanks for joining us on The Hardware Podcast.
Lucas: Thanks, Luna. And thanks to our listeners. If you found this useful, you know where to find that coffee link. Until next time.

Shows are listed from Apple Podcasts' charts and directory, and Podcast Index where it is used. Every show and episode belongs to its publisher and plays from the publisher's own feed; this site hosts none of them.