Broadcom Inc. announced on April 20, 2026, the formal extension of its strategic partnerships with Meta Platforms Inc. and Alphabet Inc. to co-develop custom artificial intelligence (AI) semiconductors through the end of 2029. These multi-year agreements solidify Broadcom’s position as the dominant provider of custom silicon solutions for hyperscale data center operators. The extensions ensure a continuous pipeline of high-performance computing hardware as both Meta and Alphabet scale their generative AI capabilities and internal infrastructure.
Under the renewed agreement with Meta, Broadcom will remain the lead design partner for the Meta Training and Inference Accelerator (MTIA) program. The collaboration is set to focus on the development of next-generation ASICs (Application-Specific Integrated Circuits) utilizing 3-nanometer and 2-nanometer manufacturing nodes. These custom chips are engineered to handle Meta’s specific AI workloads, including content ranking, recommendation engines, and large-scale generative models. Meta’s strategy involves increasing its reliance on internal chip designs to optimize power efficiency and reduce the long-term costs associated with third-party hardware procurement.
Simultaneously, Broadcom confirmed the extension of its decade-long collaboration with Alphabet’s Google regarding the Tensor Processing Unit (TPU) series. The new contract covers the co-design and manufacturing of future TPU generations through 2029. These accelerators are central to Google’s AI infrastructure, powering services such as Google Search, YouTube, and the Gemini family of models. The partnership will integrate Broadcom’s proprietary high-speed interconnect technologies and advanced high-bandwidth memory (HBM) interfaces to support the massive data throughput required for AI training clusters.
Broadcom CEO Hock Tan stated that the extensions are a testament to the company’s ability to provide specialized engineering expertise to the world’s largest technology firms. Tan emphasized that Broadcom’s custom silicon business model allows customers to define specific architectural requirements while Broadcom provides the underlying intellectual property and physical design integration. Broadcom’s AI-related revenue has grown substantially in recent years; for fiscal year 2024, the company reported AI revenue of $11 billion, and the 2026 agreements are expected to support continued growth in this segment.
The custom silicon market has become a vital component of Broadcom’s portfolio as hyperscalers seek alternatives to general-purpose GPUs. By utilizing Broadcom’s design services, Meta and Alphabet can integrate specialized networking and switching capabilities, such as the Tomahawk and Jericho chipsets, directly into their compute environments. While the specific financial terms of the extensions were not disclosed, Broadcom indicated that the volume of chip shipments would scale in tandem with the partners’ multi-billion-dollar data center expansion plans.