Artificial intelligence is beginning to reshape one of the most tightly controlled resources in modern technology: the ability to design and optimize advanced computer chips. For years, companies such as Nvidia have dominated this space, not only because of powerful hardware but also due to sophisticated software ecosystems that make their chips easier to use, tells Wired.com.
A new wave of startups is challenging that advantage by using AI itself to handle one of the most specialized and expensive tasks in computing: writing low-level code that allows software to run efficiently on specific chips. One such company, Wafer, is training AI systems using reinforcement learning to generate optimized “kernel” code, which directly interfaces with hardware. This approach builds on existing large language models but adds layers of automation that can adapt code to different types of silicon.
The implications are significant. Many competing chips, including those from AMD, Amazon, and Google, already offer comparable raw computational power. What has held them back is the difficulty of programming them efficiently. By automating this process, AI could reduce reliance on Nvidia’s proprietary tools and make it easier for companies to switch between hardware platforms.
At the same time, other startups are applying AI to chip design itself. Companies such as Ricursive Intelligence are working toward systems that allow engineers to design custom silicon using natural language, dramatically lowering the barrier to entry. This could enable smaller firms to create specialized chips tailored to their needs, rather than relying on standardized hardware.
Taken together, these developments point to a broader shift: AI is not just consuming computing power but helping to distribute access to it. By automating expertise that was once scarce and expensive, it could open up chip development and optimization to a much wider range of organizations, potentially reshaping the competitive landscape of the tech industry.