The CPU war that never really ended: RISC, x86 and the future of computing

In the 1990s, the personal computer seemed to have a permanent centre of gravity: x86, Windows and the Intel-compatible upgrade cycle. Yet inside engineering labs, Unix workstations and server rooms, RISC processors were moving fast. MIPS, SPARC, PowerPC, PA-RISC and DEC Alpha showed that serious performance did not require a swollen instruction set. In many cases, the opposite seemed true. Three decades later, the old argument has returned. RISC is rising again through Arm laptops, Apple Silicon Macs, cloud CPUs, AI infrastructure and RISC-V designs. The x86 ecosystem is still powerful and far from finished. But the ground under it is moving. What frightened x86 in the 1990s was performance. What pressures it today is efficiency, customisation and control.

The simple idea behind RISC

RISC stands for Reduced Instruction Set Computer. The principle sounds modest: use a smaller, simpler set of instructions, execute them quickly, and let compilers do more of the organisational work. Instead of giving the processor a large menu of complex instructions, RISC designs favour regularity: load data, operate on it, store it back and keep the pipeline moving.

That was a sharp contrast with x86, which came from a different world. The x86 instruction set had grown through years of compatibility as the PC market expanded. It was commercially brilliant because old software kept working, but awkward because the processor had to carry history around with it. RISC designers looked at that baggage and saw an opportunity.

Why RISC made x86 nervous in the 1990s

The 1990s were dangerous for x86 because the PC had not yet absorbed the whole computing world. Engineering, publishing, graphics, databases and scientific workloads often lived on expensive Unix workstations and proprietary servers. Many were based on RISC.

Sun’s SPARC systems were common in technical environments. MIPS powered workstations and game consoles. IBM, Motorola and Apple pushed PowerPC. Hewlett-Packard had PA-RISC. Digital Equipment Corporation’s Alpha was famous for its 64-bit ambition. These processors did not merely compete with x86. In many high-end contexts, they looked superior.

The fear was easy to understand. If personal computing moved upward into professional workstations, and if servers became the main profit centre, then the architecture with the best performance story could gain influence. RISC vendors had strong operating systems and loyal professional customers. Their machines served people who measured waiting time in money.

How x86 survived the attack

The surprise is not that RISC scared x86. The surprise is that x86 won the mainstream anyway. Compatibility was the first shield. Businesses had invested heavily in DOS and Windows software, consumers wanted familiar applications and developers followed the installed base. A faster processor was attractive, but not if it forced a company to abandon its software stack or pay a premium for a specialised workstation.

Manufacturing was the second shield. Intel and later AMD benefited from the scale of the PC market. Huge volumes funded better fabs and aggressive roadmaps. RISC vendors often had excellent engineering, but many were tied to narrower markets. When the PC became good enough for more professional tasks, the economics shifted brutally.

The third shield was the cleverest: x86 learned to behave more like RISC inside. Modern x86 processors began translating complex instructions into simpler internal micro-operations. The outside world still saw compatibility. Inside the chip, the execution engine borrowed ideas associated with RISC, including pipelining, superscalar execution, out-of-order scheduling and register renaming.

The RISC machines faded, but the idea did not

By the early 2000s, many classic RISC families had lost momentum. Workstations were squeezed by cheaper x86 PCs, while proprietary Unix servers faced pressure from Linux on commodity hardware. Alpha disappeared into corporate history. MIPS retreated from the desktop conversation. SPARC and PA-RISC became specialised. PowerPC remained important in certain niches but lost the Mac when Apple moved to Intel in 2006.

It would be easy to call that the end of RISC. It was not. RISC lost a particular market battle, not the architectural argument. The idea continued to thrive where power, size and efficiency mattered. That place was mobile.

Arm became the quiet winner. It was spreading into embedded systems, handheld devices and eventually smartphones. When mobile computing exploded, performance per watt mattered as much as absolute speed. Arm’s efficient RISC heritage was perfectly timed for a world running on batteries.

Why RISC is rising again now

The current RISC revival is not a simple replay of the 1990s. It is broader and more practical. The question is no longer whether RISC can beat x86 in a workstation benchmark. The question is whether one general-purpose architecture should dominate every device, cloud server and AI system.

Apple’s move from Intel to Apple Silicon made the argument visible to ordinary buyers. The Mac did not switch to Arm as a compromise. It switched because Apple could design processors around its own priorities: battery life, media engines, neural processing, memory integration and quiet performance. That mattered. It showed that Arm-based PCs could be desirable, not merely efficient.

Windows on Arm has also become more credible. Newer Arm-based Windows laptops are closer to normal computers, with better performance, stronger battery life and improving app compatibility. x86 still has advantages in gaming, legacy software and broad peripheral support, but the psychological barrier has weakened.

The cloud has changed the CPU debate

The data centre is another reason RISC has returned. Cloud providers buy processors at enormous scale and care intensely about power, density and cost. If a custom Arm server chip can deliver better efficiency for a given workload, the savings affect the economics of the entire cloud.

That is why major cloud companies have built or adopted Arm-based server CPUs. The decision is operational. Owning more of the silicon stack lets cloud providers tune performance, reduce dependency on external roadmaps and differentiate their services. In the 1990s, RISC vendors sold high-end machines to customers. Today, some of the biggest customers are designing the machines themselves.

AI has pushed this further. Modern AI systems depend on GPUs, accelerators and specialised interconnects, but they still need host CPUs and control processors. In that environment, the CPU is part of a larger system rather than the only star. RISC architectures, especially Arm and increasingly RISC-V, fit a chiplet-heavy, accelerator-rich world where flexibility is valuable.

RISC-V adds a new twist

RISC-V is the most important new element in the story. Unlike Arm, which is licensed from a commercial company, RISC-V is an open instruction set architecture. Companies, universities and research groups can build compatible processors without paying for the basic ISA. Chip design is still hard, but the strategic conversation changes.

For embedded devices, microcontrollers, automotive systems and AI accelerators, RISC-V offers freedom to customise. Designers can add extensions, strip away unnecessary parts and build processors around specific tasks. For countries and companies worried about supply chain dependence, it also offers a route to architectural independence.

RISC-V is still young compared with x86 and Arm. Its software ecosystem, high-performance cores and commercial support are uneven. But it has momentum because it matches the mood of the industry. The future of computing is not one processor for everything. It is many processors and specialised accelerators working together.

Why x86 is not finished

It would be a mistake to write x86 off. x86 remains deeply entrenched in desktops, gaming PCs, enterprise software, servers and developer workflows. Intel and AMD still produce capable processors, and the x86 software ecosystem is enormous. In many situations, the safest and fastest buying decision is still x86.

Modern x86 chips are not museum pieces. They include advanced power management, hybrid core designs, AI acceleration features and sophisticated packaging. The better question is not whether RISC will kill x86. It is whether x86 will continue to own the default. In the 1990s, x86 survived because it became good enough, cheap enough and compatible enough. Today, it must prove that it is efficient enough, customisable enough and strategically attractive enough in a market that no longer revolves around the classic PC.

The old fear has a new shape

RISC scared the x86 world in the 1990s because it exposed a weakness: compatibility was powerful, but it came with complexity. x86 answered by turning that complexity into a hidden engineering problem and using the PC ecosystem as a fortress. That worked brilliantly.

Now the pressure comes from another direction. Computing has spread into phones, tablets, wearables, cars, cloud platforms and AI systems. The best chip is no longer simply the fastest general-purpose CPU. It is the chip that fits the device, workload, battery, thermal envelope and business model.

That is why RISC is rising again. Not because the old workstation wars are back, but because the industry has moved towards the conditions RISC always liked: efficiency and specialisation. x86 remains a giant, but it is no longer the only centre of the computing universe. The fear of the 1990s has returned as a quieter, broader pressure. This time, it is asking who gets to design the future of computing.