Meta outlines its RSC supercomputer and its work on two chips: MTIA for accelerating AI training, set for release in 2025, and MSVP, for video processing needs
#riscv Mikael Strandlund : Meta's Engineering and Infrastructure teams are excited to host AI Infra @Scale, a one-day virtual event featuring a range of speakers from Meta … Roman Levenstein : I'm thrilled to finally share the amazing work my team, MTIA SW, has been diligently working on for a while! … Tweets: @risc_v : MTIA v1 is @Meta's first-generation #AI inference accelerator! The processor cores are based on #RISCV and are customized to perform necessary compute and control tasks. Learn more: https://ai.facebook.com/... #RISCVeverywhere Jay Hack / @mathemagic1an : Meta announces their own AI ASIC today Optimized specifically for inference in recommendation workloads Provides >2x more useful computation per watt than GPUs at a certain level of workload complexity https://ai.facebook.com/... https://twitter.com/... [image] Dan Nystedt / @dnystedt : Meta tapped TSMC to manufacture its new AI chips, the company said in a blog post. The new MTIA accelerator chips are manufactured on TSMC's 7nm process technology, the post says. $META $TSM https://ai.facebook.com/... Calista Redmond / @calista_redmond : RISC-V AI news from @Meta! Each chip include two processor cores (one with vector extension) based on the RISC-V ISA. @risc_v #RISCVeverywhere https://ai.facebook.com/... Rowan Cheung / @rowancheung : 1. Meta reveals their custom AI chip Meta revealed plans to create its custom chip, the Meta Training and Inference Accelerator (MTIA), specifically designed for running AI models. Mita promises more computing power and efficiency than standard CPUs. [video] Boz / @boztank : We want to deliver awesome AI powered experiences to billions of people. A big part of that is ensuring the entire end-to-end stack is up to the task, from the data centers and the silicon to the software layer and end user experience. Great progress so far, best yet to come! Boz / @boztank : Lots of big AI news from us today: our next gen data centers designed from the ground up for AI workloads, our first custom silicon for AI training & inference, and phase two of our Research Supercluster, one of the world's most powerful AI supercomputers https://ai.facebook.com/... Yann LeCun / @ylecun : A series of AI announcements by Meta: - MTIA v1: an AI chip for fast inference: https://ai.facebook.com/... - RSC: 5 exaflops, 16,000 GPU Research Super Cluster for AI research: https://ai.facebook.com/... - AI-focused data centers: https://ai.facebook.com/... [image] Yann LeCun / @ylecun : @aphysicist @ArturTanona Nope Aaron Slodov / @aphysicist : @ylecun @ArturTanona Did you get an invite to testify in front of congress yet? Clive Chan / @itsclivetime : MTIA seems cool (especially that it supports eager) but it seems to have missed the boat on large language models - where's the interconnect and HBM and fp8? Might be a great cost savings for their internal vision / recsys workloads but feels like yesterday's accelerator https://twitter.com/... @josephjacks_ : In the future, everyone serious about software will make their own hardware. This is already true today at Hyperscale. https://twitter.com/... Alex Barinka / @alexbarinka : Custom chips, AI coding tools, supercomputers: The quest for AI leadership is expensive, and Meta's spending at record levels. Here's what Meta's been working on behind the scenes and how it'll change its business today and in the future: https://www.bloomberg.com/... Tiernan Ray / @tiernanraytech : Meta unveils first custom artificial intelligence chip The circuits come with software optimized to run PyTorch, and emphasize the task of making recommendations. $META #AI #deeplearning #semiconductors https://www.zdnet.com/... Sarah Frier / @sarahfrier : In Meta's year of extreme cost cutting, there's one area that is getting record spending: AI. @alexbarinka gives the rundown of custom chips, a gen AI coding helper, and more: https://www.bloomberg.com/...
Context & Ripple Effects
Meta’s chip roadmap follows a period in which the company was described as reworking an earlier custom-chip effort while trying to close an AI infrastructure gap. This report turns that effort into a defined portfolio: RSC supplies large-scale GPU research capacity, while MTIA and MSVP target workloads that Meta can tune in-house.
The roadmap later gained execution evidence when Meta said MTIA v1 and its next-generation accelerator had entered production. That makes this announcement an early marker of Meta’s move from buying general-purpose AI systems to building workload-specific infrastructure.
First-order effects
- Meta gains a planned in-house path for AI training/inference and video processing alongside its RSC GPU fleet, with MTIA manufacturing assigned to TSMC’s 7nm process.
- Engineering teams must support distinct hardware and software paths: MTIA for AI workloads and MSVP for video processing, rather than relying on a single compute architecture.
Second-order effects
- A successful MTIA deployment could shift part of Meta’s workload demand away from general-purpose GPU systems, while increasing the importance of compiler, runtime, and model-optimization work for its custom silicon.
- TSMC becomes a direct manufacturing dependency in Meta’s accelerator roadmap; video and AI workloads can be allocated according to the economics and performance of each specialized chip.
Third-order effects
- The announcement points to a more heterogeneous AI infrastructure model, in which hyperscalers combine large GPU clusters with custom accelerators for high-volume internal workloads.
- If this approach proves repeatable, competitive advantage will increasingly depend on an integrated stack—silicon, systems, and software—not simply access to the largest GPU fleet.
The trend: Hyperscalers are segmenting AI infrastructure across GPUs and custom ASICs to tailor compute to the workloads they operate at scale.