AI that responds in under 100ms doesn't happen by accident. @togethercompute VP of Kernels @realDanFu shares how they use NVIDIA CUDA, TensorRT-LLM, Dynamo and Together ATLAS on NVIDIA Blackwell to power ultra-low-latency inference and long context code generation for @cursor_ai. From building megakernels that fuse an entire model's forward pass into a single launch to enabling real-time voice agents, Together AI is optimizing latency at every layer of the stack.

🏭 @TRACTIAN is automating industrial maintenance with physical AI agents, powered by NVIDIA's full-stack platform. An #NVIDIAInception member, Tractian's platform processes terabytes of sensor data across 200,000 heavy industry machines, with 500 million inference requests served through 50 specialized agents. Tractian built its physical AI models on NVIDIA GB300 NVL72 systems with CUDA-X libraries to deliver: ✅ 35% reduction in model training time ✅ 50% reduction in inference latency ✅ Machine failure prevented roughly every 15 minutes Read the full case study ➡️ nvda.ws/4p0bG61

🧬 @PrimaMente is scaling faster in the race to decode Alzheimer's and Parkinson's. Powered by @nebiusai and running on NVIDIA accelerated computing, Prima Mente boosted pre-training throughput by 12%, now processing 1.19M tokens/s across 16 nodes as it scales models from 7B to 100B parameters. Learn more below 👇

America is a nation of builders. For 250 years, America has built railroads, power grids, factories, semiconductors, and the internet. Now, America is building again.

💡 Query engines are often bottlenecked by memory and I/O bandwidth. NVIDIA hardware advances in high bandwidth memory, NVLink-C2C, and decompression engines in NVIDIA Grace Blackwell remove bottlenecks by increasing storage capacity, accelerating data movement between CPUs and GPUs, and speeding up data access. NVIDIA GQE (GPU Query Engine) shows how GPU-native design, compression, partition pruning, and accelerated data movement can help developers build faster data platforms on modern NVIDIA hardware. ➡️ nvda.ws/4wkpieI

AI agents can now run for hours across your enterprise systems, which makes governing them a non-negotiable. 🔒 The NVIDIA Secure Agent Workspace Reference Design provides enterprises with a blueprint for securely deploying always-on agents. This is achieved by moving execution off the user's device into a managed workspace with controlled identity, network access, and policy enforcement. 📖 Learn how to govern autonomous agents in your enterprise AI factory: nvda.ws/4f0bYFA

💡 Federal agencies aren't just piloting AI anymore — they're deploying it at mission scale. At AWS Summit Washington D.C., join leaders from NVIDIA, @awscloud, and @OpenAI to learn how leading models enable the federal community to deploy powerful, production-ready AI in secure environments and support inference at scale. 📅 Wednesday, July 1 | 9:30 a.m. ET 📍 Walter E. Washington Convention Center, Room 145AB 🔗 Learn more: nvda.ws/4wjBwV5

NVIDIA inference software keeps driving down token costs, long after AI infrastructure is deployed. ⚡ In just one month on NVIDIA Blackwell, software optimizations improved DeepSeek V4 performance by up to 5×, reducing token costs to roughly one-fifth of previous levels. NVIDIA's integrated inference software stack compounds improvements across runtimes, kernels, networking, and hardware, delivering up to 20× higher throughput on the same GPU. Co-designed with NVIDIA GPUs, CPUs, networking, and systems, and powered by CUDA-native open source frameworks, NVIDIA's inference software stack ensures new model breakthroughs and optimizations run on NVIDIA from day zero, and keep improving throughput and lowering cost after deployment. See how @Baseten, @Cognition, @DeepInfra, @togethercompute, and @Cursor_ai are turning continuous software innovation into lower cost per token: nvda.ws/4eRT43m

👏 Congratulations to @IREN_Ltd on achieving NVIDIA Exemplar Cloud status on NVIDIA HGX B300 for training workloads. This validation reflects close engineering collaboration and gives enterprises confidence that demanding AI training workloads can perform at scale on IREN’s AI Cloud. Learn more ⤵️

📣 @AnthropicAI Claude models are now generally available in @Microsoft Foundry, running on the NVIDIA GB300 NVL72 platform on Azure with NVIDIA Quantum-X800 InfiniBand networking. For enterprises building the next generation of agentic AI, this means more powerful autonomous and domain-specific agents, backed by the inference performance and lowest cost per token.

NeoSpace is helping private banks in Latin America expand credit access with AI foundation models built on NVIDIA. 🏦 An #NVIDIAInception member, NeoSpace developed NeoData, a large tabular foundation model platform trained on NVIDIA GB200 NVL72 systems via @OracleCloud. NeoData delivers a unified, production-grade predictive layer across trillions of enterprise financial records. A major Latin American bank with 60M customers used NeoData to achieve: ✅ 30–50% higher accuracy vs. traditional ML models ✅ 19% lift over the baseline fine-tuned model ✅ 10% increase in installment credit offers 📝 Read the full customer story: nvda.ws/4vIWNrs

📣 We are collaborating with the @NPS_Monterey to empower thousands of students to work and build applications with AI.   Hear from Todd Lyons, EVP at the @npsfoundation, on how #NVIDIADGX and NVIDIA Omniverse are helping to create synthetic data, train tailored foundational models, and deploy solutions at scale for the public sector.

🧪 @EliLillyandCo's new AI factory is accelerating drug discovery and reshaping how scientists work. Powered by #NVIDIADGX SuperPOD with DGX B300 systems, LillyPod is turning drug discovery into an always‑on AI engine—running longer, more complex workflows. ✅ Billions of research hypotheses evaluated in parallel ✅ Running open models with no token or budget limits ✅ Deploying Nemotron 3 Ultra on-prem, optimized for NVIDIA Blackwell Ultra GPUs at FP4 ✅ Training foundational models from chemistry and biology to clinical and manufacturing Read @RandDWorld's article on LillyPod ⤵️ rdworldonline.com/six-months…

Up to 10x better inference per watt. One-tenth the cost per million tokens. This is what NVIDIA Vera Rubin NVL72 delivers. Join us for the virtual event on June 30 as Dion Harris, Senior Director of HPC and AI Hyperscale Infrastructure Solutions at NVIDIA, and Harsh Banwait, Director of Product at @CoreWeave, explain what it means for the agentic era. #theCUBE 📆 Save the date: nvda.ws/4wgbHFt