In a move set to redefine the future of artificial intelligence infrastructure, OpenAI has announced a major partnership with Broadcom to build 10 gigawatts of in-house AI accelerators. The collaboration aims to expand OpenAI’s compute capacity and reduce dependency on third-party chip providers, marking a significant step in the company’s long-term strategy to control the full stack of AI development—from model design to hardware execution.
The partnership, unveiled earlier this week, is one of the largest AI hardware collaborations to date. The initiative will see Broadcom design and manufacture custom AI chips optimized for OpenAI’s large-scale model training and inference workloads. The chips are expected to power OpenAI’s next-generation models, including future iterations of its widely used GPT and Sora systems.
Industry experts believe this move positions OpenAI among a select group of technology giants building their own dedicated compute infrastructure. The scale of the 10-gigawatt buildout also underscores the growing energy demands of large-scale AI systems and reflects a shift toward vertical integration in the AI industry.
According to the company, these new accelerators will be deployed in OpenAI’s global data centers over the coming years. They will support both model training and real-time inference while improving power efficiency and latency across AI workloads. Broadcom’s role will involve co-developing custom silicon using advanced packaging technologies that enhance performance without compromising scalability.
OpenAI stated that the partnership will help ensure greater control over its compute pipeline, which has become a critical bottleneck in AI development. Demand for AI chips has skyrocketed worldwide, with shortages and rising costs affecting many companies relying on Nvidia’s GPUs. By building proprietary accelerators, OpenAI hopes to secure a reliable supply chain and achieve significant cost efficiencies over time.
In a statement, Sam Altman, CEO of OpenAI, emphasized that compute availability is central to the company’s long-term mission of advancing safe and scalable AI systems. “To build the next generation of AI, we must scale both our models and our infrastructure. This collaboration with Broadcom allows us to design the compute foundation we need for the future of intelligence,” he said.
Broadcom, known for its leadership in semiconductor innovation, will contribute decades of experience in chip architecture and manufacturing. The company’s CEO Hock Tan called the partnership a milestone in AI hardware development. “Our collaboration with OpenAI reflects the convergence of software intelligence and hardware performance. Together, we are setting new benchmarks for efficiency, power, and scalability in AI compute systems,” Tan stated.
The 10-gigawatt project is expected to unfold over several phases, with initial deployments planned for major AI research hubs in the United States, followed by international expansion. The chips will be manufactured using cutting-edge 3-nanometer and 2-nanometer process technologies, ensuring high performance and energy efficiency.
Industry analysts note that OpenAI’s move mirrors the strategic paths taken by other major players such as Google (with its TPU processors), Amazon (with Inferentia and Trainium chips), and Meta (with its MTIA accelerators). However, the scale of OpenAI’s investment—coupled with its exclusive focus on AI model development—positions it uniquely in the semiconductor race.
The 10-gigawatt figure refers to the total computing power capacity OpenAI plans to deploy globally, which could make it one of the largest single AI compute infrastructures ever built. This capacity will be essential as AI models become exponentially larger, requiring unprecedented levels of parallel processing and energy.
According to estimates from energy research firms, the combined power demand from global AI data centers could exceed 50 gigawatts by 2030, with OpenAI’s expansion alone expected to represent a significant fraction of that. The company has also indicated its intention to source renewable energy for its data centers to minimize environmental impact.
The move is also seen as part of OpenAI’s broader strategy to reduce its dependence on Nvidia’s GPU ecosystem, which currently dominates the market. By developing its own accelerators, OpenAI gains more autonomy over performance tuning, software integration, and deployment schedules—key factors in maintaining competitive advantage in the fast-evolving AI landscape.
A report from Livemint highlighted that the partnership could lower training costs for large-scale models by as much as 40% over time, while also improving inference speeds. Analysts suggest that these efficiencies could allow OpenAI to accelerate its product roadmap and deliver faster, more capable iterations of AI tools like ChatGPT, Codex, and DALL·E.
Beyond cost and performance, the collaboration also underscores the increasing geopolitical importance of semiconductor independence. The U.S. government has been encouraging domestic AI leaders to secure chip partnerships with American manufacturers as part of broader efforts to strengthen national AI and semiconductor supply chains.
Broadcom’s facilities in the U.S. and Asia will play a central role in chip fabrication, while OpenAI will oversee design optimization and deployment. Sources indicate that the companies are also exploring joint research on AI-driven chip design, where machine learning tools are used to automate and improve semiconductor engineering.
OpenAI’s expansion into hardware follows a series of ambitious steps taken in 2025 to broaden its technological capabilities. Earlier this year, the company launched Sora, its generative video model, and expanded GPT-5 training operations with increased compute requirements. The partnership with Broadcom, experts say, is the logical next step to support these advancements.
The collaboration is expected to have ripple effects across the AI ecosystem, influencing startups, enterprises, and governments that rely on OpenAI’s technologies. For developers and enterprises, improved compute efficiency could translate into faster access to new models, lower API costs, and more sustainable AI deployment options.
While the exact financial details of the partnership were not disclosed, sources familiar with the deal described it as “multi-billion dollar in scope” and among the largest semiconductor collaborations ever initiated by an AI firm.
As the AI arms race intensifies, OpenAI’s partnership with Broadcom signals a decisive move toward long-term scalability and infrastructure independence. By controlling more of its compute stack, OpenAI is not only securing its future capacity but also shaping the direction of the global AI hardware industry.
The deployment of the 10-gigawatt infrastructure is expected to begin in 2026, with phased rollout continuing through the end of the decade. Once operational, it could serve as the backbone for OpenAI’s next generation of multimodal AI systems—setting new standards for performance, efficiency, and integration across the AI landscape.