PluralisHQ has, for the first time, split and trained a large language model by distributing model weights at scale across consumer devices connected over the internet, with no loss in speed or performance. This was previously thought to be unsolvable. The method enables training of billion-parameter AI models over low-bandwidth connections, such as 80kbps, using consumer GPUs. This development allows individuals to pool globally distributed computing resources to train models larger than would be possible individually. PluralisHQ is recognized among the leading organizations in decentralized AI training, alongside PrimeIntellect and NousResearch. The decentralized AI sector has seen several new partnerships and product integrations. OpenLedger announced a collaboration with Gaianet_AI to integrate decentralized networks into its AI infrastructure, and separately, a partnership with UnifAI_Network to support autonomous agents with specialized AI and data infrastructure. NexusLabs revealed a partnership with QuillAI_Network to build modular AI security infrastructure for Web3, while PublicAI and PhalaNetwork are working together to create verifiable, private, and tamper-proof AI datasets and models using TEE-secured computing. Other developments include OpenPad AI's partnership with DeepSafe_AI to provide decentralized verification for cross-chain AI agent coordination, CertiK's integration of Chain_GPT for instant AI-generated token summaries and DeFi risk analysis, and Nosana_AI's collaboration with AskJimmy_AI to supply decentralized GPU compute for AI trading agents. Chainlink Labs has joined the Global Synchronizer Foundation to build enterprise-grade blockchain applications, and OKZOO has partnered with CARV to integrate environmental sensor data into AI agents for AIoT data sovereignty. AuditOne_DAO has launched AI agent v2 on SonicLabs, offering token assessment and security analysis features.