AI-Designed CPU Runs Doom, Marking Hardware Engineering Milestone

A groundbreaking demonstration has seen an AI computing enthusiast successfully run the classic video game Doom on a custom-designed CPU created entirely by GPT-5.6 Sol, with the game's viewport overlaid on a pulsing schematic of the CPU within the "Turing Complete" sandbox environment. This achievement, showcased by Angel (@Angaisb_) via a video on August 23, 2026, highlights the advanced capabilities of generative AI in hardware design, moving beyond theoretical concepts to tangible, functional computing systems. The AI-designed CPU, named 'Codex-R32,' was constructed from primitive logic components within the "Turing Complete" game, and ran a C-based PureDOOM port compiled into native RV32IM machine code.

This feat is more than a novelty; it signifies a pivotal moment in the convergence of artificial intelligence and hardware engineering, demonstrating AI's capacity to autonomously architect complex computational systems from the ground up. Historically, CPU design has been an intensely human-driven, iterative process, relying on specialized expertise in electrical engineering, computer architecture, and intricate electronic design automation (EDA) tools. The ability of a large language model like GPT-5.6 Sol to conceive and realize a functional CPU, even within a simulated environment, suggests a dramatic acceleration and democratization of hardware development. This could lead to a future where bespoke silicon, optimized for highly specific tasks, becomes far more accessible and rapidly deployable, potentially disrupting the semiconductor industry's traditional multi-year design cycles.

The "Turing Complete" sandbox, where this demonstration took place, is a game-like environment designed for learning computer science by building logic gates and circuits to create increasingly complex components and architectures. Its nature as a virtual, logic-gate-level construction platform makes the AI's success particularly impressive, as it implies a deep understanding of fundamental digital logic and system integration rather than merely high-level abstraction. This contrasts with current AI tools in hardware design, which primarily assist human engineers in tasks like generative design for enclosures, optimizing layouts, simulating physics, and automating PCB placement, rather than designing entire functional CPUs from scratch. While AI is increasingly used to accelerate specific stages of hardware prototyping, verification, and optimization, the Codex-R32 represents a significant leap towards autonomous architectural generation.

The choice of Doom as a benchmark is also highly symbolic. Since its release in 1993, Doom has become a ubiquitous test for the viability of new or minimal computing platforms, famously running on everything from calculators to toasters due to its optimized code and the release of its source code. Its successful, albeit unplayable, execution (initial performance was 0.7 FPS, later optimized to 15-20 FPS with various hardware and compiler tweaks) on the Codex-R32 validates the AI's design as a genuinely functional, programmable processor. This is a critical distinction from merely generating schematics, confirming the AI's ability to produce an architecture capable of executing a complex, real-world software application.

Looking ahead, this demonstration heralds a future where AI could fundamentally reshape the hardware landscape. The immediate impact lies in significantly shortening design cycles and reducing the immense costs associated with cutting-edge chip development, which can reach hundreds of millions of dollars for a 5 nm node chip. AI-driven design could enable rapid iteration and exploration of vast design spaces that are currently inaccessible to human engineers, potentially leading to novel architectures with extreme optimization for power, performance, and area (PPA). Experts anticipate that within the next five years, AI will heavily automate workload generation, debugging, verification, and physical design problems, accelerating design closure. Some AI-designed chip components are already emerging, with reported sizes 500 times smaller than human-envisioned designs.

However, challenges remain. The need for human oversight in AI-designed systems is crucial, as AI can still "hallucinate" non-functional elements or make faulty arrangements. Ensuring reliability, security, and explainability of AI-generated designs will be paramount, particularly for mission-critical applications. Furthermore, while GPT-5.6 Sol designed the CPU, the demonstration still required an AI coder to get Doom running, highlighting that a complete, fully autonomous hardware-to-software stack remains a future goal. The industry is already moving towards domain-specific architectures and agentic AI, which demands flexible and reconfigurable hardware. The Codex-R32 provides a glimpse into a future where AI not only assists in designing these specialized components but potentially architects entire systems, paving the way for a new era of highly customized and efficient computing tailored precisely to the demands of an increasingly AI-driven world.