Newsroom
19 November, 2026 / News / AI / Tags: openai, amd, chip, jalape, models

OpenAI’s AI models have delivered substantial efficiency gains in advanced semiconductor designs for AMD and its own projects, compressing what would typically take years into under five months
OpenAI’s artificial intelligence models have advanced AMD’s flagship GPU intellectual property with power, performance, and area improvements of up to 14 percent. These gains exceed one year of conventional design progress and were achieved by rewriting SystemVerilog code that defines chip circuitry. Verification tools confirmed the changes preserved the original behavior of each block. Some modifications have already been incorporated into a design now in manufacturing, while others remain in production files for later revisions. AMD separately verified single-digit percentage improvements in key blocks of its next-generation flagship CPU using similar AI-generated designs.
The same methodology was applied to Jalapeño, OpenAI’s custom inference chip developed with Broadcom. The models reduced the standard-cell area by about 13 percent in the vector unit and 10 percent in the matrix unit. These results were verified on silicon, demonstrating practical benefits in a specialized high-performance computing environment. The models directed GPT-o3 mini at the design during early 2025, highlighting the system’s ability to handle complex electronic design automation tasks directly.
OpenAI is developing a specialized model in partnership with Synopsys to operate electronic design automation tools end-to-end. This development builds on the proven success of AI rewriting code and running verification steps, positioning advanced large language models at the center of future semiconductor innovation. The approach compresses what has historically taken years into a fraction of the time while maintaining rigorous validation standards.
OpenAI’s Preparedness Framework addresses the risks associated with AI self-improvement, treating it as a critical area of concern. The company has committed to halting further development if a model reaches the Critical self-improvement threshold until appropriate safeguards are fully implemented. This stance acknowledges that faster capability gains could complicate human oversight of AI systems, even as they accelerate progress in specialized technical domains such as chip design.
The collaboration between OpenAI and AMD illustrates how generative AI is reshaping workflows in the semiconductor industry. By combining model-driven code generation with traditional verification, the partnership has achieved efficiency levels previously unattainable through manual processes alone. Similar principles are now being extended to OpenAI’s internal projects, including the Jalapeño inference chip, signaling a broader shift toward AI-native design methodologies across the technology sector.
These results come at a time when chip designers face increasing pressure to deliver higher performance with lower power consumption and smaller footprints. The AI models’ ability to propose and validate changes in SystemVerilog represents a significant leap forward in automated design exploration. While OpenAI has not disclosed the specific AMD products involved or a detailed breakdown of gains across the three metrics, the verified outcomes underscore the potential for large language models to serve as collaborative partners in complex hardware development.









