{"company":{"slug":"meta-platforms","name":"Meta Platforms","legal_name":"Meta Platforms, Inc.","aliases":["Meta","Facebook","META"],"primary_layer":"L3","primary_topic_id":"infiniband-ethernet","secondary_topics":null,"hq_country":"US","hq_city":"Menlo Park","founded_year":2004,"is_public":true,"ticker":"META","description":"Largest single Ethernet-for-AI buyer; Ultra Ethernet Consortium driving member. AI Research SuperCluster + MTIA accelerator deployments use Ethernet fabric not InfiniBand."},"profile":{"snapshot":"Largest Ethernet-for-AI buyer; UEC 1.0 driver scaling multi-GW RoCE superclusters over InfiniBand","stack_layer":"L3","why_join_top_2":["Meta raised 2025 capex to $66-72B and guided 2026 dollar growth 'notably larger' (signal 66bf1b34), funding multi-GW Ethernet superclusters Prometheus (~1GW, 2026) and Hyperion (toward 5GW) — frontier-scale fabric work.","As UEC 1.0 founding steering member (June 11 2025, signal c3c6447a) hiring 2 RDMA/RoCE transport roles in Apr-May 2026, you help write the InfiniBand-alternative standard, not just consume it."],"peer_comparison":{"substitute":{"name":"Cornelis Networks","delta":"Cornelis pushes InfiniBand-class Omni-Path as the non-Ethernet path Meta's UEC 1.0 effort is explicitly designed to displace."},"direct_competitor":{"name":"HPE","delta":"HPE sells Slingshot Ethernet fabric as a product to exascale labs; Meta builds its own RoCE/DSF fabric in-house at multi-GW scale, not for sale."}},"self_quality_score":{"notes":"Risk is falsifiable by end-2026 against the Prometheus cluster's RoCE performance.","originality":7,"evidence_density":9,"anti_generic_pass":true,"reverse_hype_present":true},"strategic_position":{"top_risk":"If RoCE/DSF fabric underperforms on the ~1GW Prometheus cluster as it comes online in 2026, Meta could revert to InfiniBand for frontier training, undercutting its UEC thesis.","their_bet":"Betting open Ethernet (UEC 1.0 + Disaggregated Scheduled Fabric) replaces InfiniBand at million-endpoint scale, funded by $66-72B 2025 capex and the Prometheus/Hyperion buildout (signals c3c6447a, b2fb752d).","what_they_do_best":"Operates the world's largest in-house Ethernet/RoCE AI training fabric, including 24,576-GPU RoCE clusters for Llama-class models (signal 88ae9cf9)."},"top_business_signals_90d":[{"summary":"Meta is building multi-gigawatt Ethernet/RoCE superclusters Prometheus (~1GW, online 2026) and Hyperion (toward 5GW), explicitly not InfiniBand.","signal_id":"b2fb752d-9f4e-4041-9e07-0cb77130a2e5","evidence_quote":"the ~1GW 'Prometheus' cluster coming online in 2026 and 'Hyperion' (Louisiana) scaling toward 5GW, both built on Meta's Ethernet/RoCE-based AI fabric"},{"summary":"Meta signed a CoreWeave compute agreement worth up to $14.2B through 2031 to supplement its own Ethernet-fabric training capacity.","signal_id":"02f51ab8-f79c-4912-9a09-2e0ad4f5b736","evidence_quote":"Meta signed a cloud compute agreement with CoreWeave worth up to $14.2 billion through 2031"},{"summary":"At OCP 2025 Meta detailed Disaggregated Scheduled Fabric and RoCE roadmap on 24,576-GPU Ethernet clusters as an explicit InfiniBand alternative.","signal_id":"88ae9cf9-df49-4170-9d28-05c55cd8300c","evidence_quote":"24,576-GPU Ethernet/RoCE training clusters used for Llama-class models as an explicit alternative to InfiniBand"}],"top_hiring_specialties_60d":[{"job_count":2,"specialty":"RDMA/RoCE transport & collective communication"},{"job_count":1,"specialty":"NIC & Ethernet switching ASIC design"},{"job_count":2,"specialty":"AI fabric reliability & program management"}]},"avg_fit_score":86.50158756684493,"fit_scores":[{"overall_score":88.3529411764706,"growth_score":92,"comp_score":null,"learning_score":90,"stability_score":80,"culture_score":70,"jobseeker_profile_id":"ffe9a03e-5b9a-4f0d-b945-3bf8a263e6a4"},{"overall_score":86.1875,"growth_score":88,"comp_score":null,"learning_score":90,"stability_score":82,"culture_score":65,"jobseeker_profile_id":"6a34a657-002e-427c-99ec-e765ae9869ad"},{"overall_score":86.375,"growth_score":88,"comp_score":null,"learning_score":90,"stability_score":80,"culture_score":70,"jobseeker_profile_id":"8c8d2974-3f3c-4c8b-bdb8-5dfef850c0de"},{"overall_score":85.0909090909091,"growth_score":92,"comp_score":null,"learning_score":90,"stability_score":80,"culture_score":68,"jobseeker_profile_id":"5f251125-ca59-45f0-9dac-10d0af8fdff2"}],"recent_business_signals":[{"signal_type":"earnings","signal_date":"2025-10-29","signal_data":{"driver":"AI infrastructure spend","period":"Q3 2025","revenue_usd":51240000000,"capex_2026_guidance":"Total dollar growth notably larger than 2025","one_time_tax_charge_usd":15930000000},"source":"ai_synthesis_sdk"},{"signal_type":"other","signal_date":"2025-10-14","signal_data":{"event":"OCP Global Summit 2025","technology":["Disaggregated Scheduled Fabric (DSF)","RoCE/Ethernet AI fabric","FBOSS"],"fabric_choice":"Ethernet over InfiniBand for large-scale GenAI training","reference_cluster_gpus":24576},"source":"ai_synthesis_sdk"},{"signal_type":"customer_win","signal_date":"2025-09-30","signal_data":{"scope":"AI compute capacity contract","term_end":"2031","partner_name":"CoreWeave","contract_value_usd":14200000000},"source":"ai_synthesis_sdk"},{"signal_type":"earnings","signal_date":"2025-07-30","signal_data":{"eps":7.14,"driver":"AI infrastructure / data center buildout","period":"Q2 2025","revenue_usd":47520000000,"net_income_usd":18340000000,"capex_guidance_2025_usd":"66-72B (raised from 64-72B)"},"source":"ai_synthesis_sdk"},{"signal_type":"capacity_expansion","signal_date":"2025-07-14","signal_data":{"fabric":"Ethernet-based AI training fabric","clusters":["Prometheus","Hyperion"],"hyperion_capacity_gw":5,"prometheus_capacity_gw":1,"prometheus_online_year":2026},"source":"ai_synthesis_sdk"},{"signal_type":"partnership","signal_date":"2025-06-11","signal_data":{"role":"Steering/founding member","relevance":"Open Ethernet transport standard positioned as alternative to InfiniBand for AI/HPC training fabrics","partner_name":"Ultra Ethernet Consortium (UEC)","spec_version":"UEC 1.0"},"source":"ai_synthesis_sdk"}],"recent_talent_signals":[{"job_title":"Technical Program Manager, AI Network Fabric Buildout","job_level":"senior","job_specialty":["network infrastructure program management"],"posted_date":"2026-05-07"},{"job_title":"Network Reliability Engineer, AI Cluster Interconnect","job_level":"mid","job_specialty":["network reliability","fabric monitoring"],"posted_date":"2026-04-28"},{"job_title":"Software Engineer, RDMA Transport and Collective Communication","job_level":"staff","job_specialty":["RDMA transport","collective communication libraries"],"posted_date":"2026-04-09"},{"job_title":"Network Engineer, AI Datacenter Fabric","job_level":"senior","job_specialty":["RDMA over Converged Ethernet","datacenter fabric"],"posted_date":"2026-03-03"},{"job_title":"ASIC Design Engineer, Network Interface and Switching","job_level":"principal","job_specialty":["NIC ASIC design","Ethernet switching silicon"],"posted_date":"2026-02-20"}],"citation_url":"/companies/meta-platforms","last_updated":"2026-05-14T14:33:37.941141+00:00","generated_at":"2026-05-19T12:00:09.935Z"}