Meta Platforms has acquired Rivos, a RISC-V-focused semiconductor startup, in a move that underscores its long-term strategy to develop custom, in-house compute engines for its vast social media infrastructure. The acquisition, confirmed by Walden Catalyst — an investment arm of Walden International — follows years of Meta’s efforts to reduce dependency on third-party CPUs and GPUs, particularly as AI workloads strain datacenter economics. Meta, which operates at scale for 3.5 billion users across Facebook, Instagram, WhatsApp, and Threads, faces unique challenges compared to traditional cloud providers. Unlike hyperscalers that must support diverse enterprise workloads, Meta can tailor its hardware directly to its own applications, especially AI-driven recommendation systems. These systems rely on massive embeddings — sometimes reaching terabytes — that overwhelm even Nvidia’s Grace-Hopper and Grace-Blackwell hybrid chips, which are designed to handle such data with high memory bandwidth. Meta’s own AI accelerators, the MTIA v1 and v2, have been built around RISC-V cores but have so far only supported inference, not training. In June 2025, Meta released a paper at the International Symposium on Computer Architecture showing that its upgraded MTIA 2i chip reduced total cost of ownership (TCO) by 44% over Nvidia GPUs for specific AI inference tasks. This performance advantage, combined with the rising cost of GPU-based infrastructure, makes custom silicon a strategic imperative. Rivos, founded in 2021 by a team with deep roots in Apple, PA Semi, and Silicon Valley chip legacy, had been a key partner in designing Meta’s MTIA chips. The company was developing both a RISC-V CPU and a data-parallel accelerator — effectively a GPU — optimized for large language models and analytics. Its architecture featured unified memory access across DDR and HBM, similar to Nvidia’s “superchip” approach, and reportedly included a CUDA-compatible software stack, raising questions about legal and technical compliance. The acquisition, valued at over $850 million, likely reflects Meta’s desire to fast-track its silicon ambitions. Rivos had raised $370 million in funding and was preparing a $500 million Series B round that could have pushed its valuation above $2 billion. With the backing of Walden International and access to TSMC’s advanced nodes, Rivos had already taped out a 3.1 GHz processor and built a software ecosystem aimed at compatibility with existing AI frameworks. While Rivos’ CUDA compatibility is legally ambiguous — especially given Nvidia’s recent licensing restrictions on binary translation layers — Meta, with its vast resources and legal firepower, may be able to navigate this gray area. If successful, Meta could deploy a RISC-V-based, Grace-Hopper-like chip that matches performance but costs significantly less, potentially disrupting the AI hardware market. The acquisition also echoes Amazon’s purchase of Annapurna Labs, which laid the foundation for Graviton and Trainium chips. With Rivos’ expertise in CPU-GPU integration, memory coherence, and RISC-V innovation, Meta now has the talent and technology to build a full-stack compute engine tailored to its needs — and possibly even enter the hardware systems business. The next phase will be how quickly Meta can integrate Rivos’ team and deliver a true training-capable XPU and general-purpose CPU.