NVIDIA AI Infrastructure

@MLCommons 🎉 Great to see a new benchmark for end-to-end agentic workloads. We are excited to share our results on MLPerf Endpoints.

MLPerf Inference now measures multi-turn agents. 990 trajectories, Kimi K2.6 + Qwen3.6-35B-A3B, Pareto-curve performance, three-level accuracy. Built on MLPerf Endpoints. mlcommons.org/2026/07/agenti… MLPerf #AgenticAI #LLM #Inference #Kimi #Qwen #MLCommons

The data center has a new job: generate tokens. @NVIDIA SVP Kevin Deierling at #COMPUTEX2026 on building the AI factory through extreme co-design: ✅ 40% more GPUs per gigawatt ✅ 10x more tokens at 1/10th the cost ✅ 800G → 1600G networking Agentic AI isn't coming. It's here. ▶️ Watch the full keynote: nvda.ws/4eXzPa8

This foundation in computational modeling for #research and #education continues today. The DGX GB300 will enable a substantial increase in computational capacity while operating within NPS' existing power infrastructure. @nvidia @NPSfoundation More: linkedin.com/feed/update/ur…

📣 An #NVIDIADGX GB300 system will become operational at @NPS_Monterey this month. To enrich AI skills for officers and students in the public sector, DGX GB300 will support the development of AI applications and address operational challenges in fields ranging from weather modeling, oceanic and operations research, to disaster resilience and response planning. 💡 Read more on how we’re training the next generation of AI leaders with NPS: nvda.ws/4wA5C6W 🔗 Learn more about the NPS and NVIDIA collaboration at Converge @ NPS: nvda.ws/3RsPPb1

The network is what determines if the AI factory mission succeeds or fails. NVIDIA's Gilad Shainer and @650Group's @AlanWeckel break down the four networking architectures behind modern AI factories — including how NVIDIA Spectrum-X Ethernet is purpose-built to eliminate jitter. 📺 Watch the full conversation: nvda.ws/4aGaVJt

Excited and honoured to have @nvidia as part of RAISE Hackathon 2026. Let the games begin. NVIDIA didn't set out to build the backbone of AI, CUDA launched in 2006 as a way to make gaming GPUs programmable for general computing. Nobody was thinking about transformers yet. Nearly two decades later, that same parallel-processing bet is why almost every model trained this year ran on their chips. Sometimes the platform outlives the plan. #RAISEHackathon

Congrats to @SpaceXAI on Grok 4.5 — trained on NVIDIA GB300 NVL72 systems and purpose-built for coding, agentic tasks, and knowledge work. This is what happens when world-class AI infrastructure meets world-class AI research.

NVIDIA Vera: The Max Single-Threaded CPU at Scale. Agentic AI systems run sequentially, each reasoning step, tool call, and code execution happens one at a time on the CPU. When that CPU slows under load, the entire agentic loop slows and valuable GPUs go underutilized. A max single-threaded CPU at scale keeps every agent step fast, even when the system is fully loaded. That's what NVIDIA Vera is built to do, and why Perplexity is already building on it. Read More nvda.ws/3T1kfl1

🔒 What does it actually cost to run AI inference inside a trusted execution environment? The team behind NVIDIA Confidential Computing shares benchmark results for CC on NVIDIA HGX B300 — testing throughput, TTFT, and TPOT — across batch sizes from 4 to 256, showing under 8% overhead across all configurations. 🔗 See the results: nvda.ws/4f74Veg

As AI-native models grow to handle real-world network complexity, legacy hardware constraints can limit what networks can achieve. NVIDIA AI Aerial's GPU-accelerated, AI-native architecture enables a new class of Layer 1 and Layer 2 algorithms that help close the massive MIMO performance gap in real deployments, with up to 1.62x throughput gains in AI-based beamforming. The shift to AI-native RAN is the path to enhancing performance, efficiency, and programmability for radio networks. Learn more ➡️ nvda.ws/4yabvcz

Powered by NVIDIA Blackwell GPUs, #NVIDIADGX B200 is the foundation for your AI factory — delivering 3x the training performance and 15x the inference performance of previous-generation systems. Specs include: ✅ 8x Blackwell GPUs ✅ 1,440 GB of GPU memory ✅ NVLink bandwidth: 14.4 TB/s ✅ 2 TB of system memory Dive into the features and learn how DGX B200 enables enterprises to accelerate any AI workload ➡️ nvda.ws/3TnRy1M

How NVIDIA Runs Its Own AI Factory | AI Factory Insider Ep. 2 x.com/i/broadcasts/1…

What does it take to scale an AI factory? Scale-up and scale-out networking are both essential: NVLink pools GPU memory for larger models, and NVIDIA Spectrum-X Ethernet scales AI factories out across racks and data centers to hundreds of thousands of GPUs. 🔗 Learn more: nvda.ws/44d0REd

💡 Between every model step — tool calls, code execution, data processing, orchestration — the CPU is on the critical path. Slow it down, and your entire AI factory feels it. NVIDIA Vera tackles this head-on, built to keep agentic workflows moving at full speed. Dive into how the Vera CPU design maximizes fleet performance ➡️ nvda.ws/4f0t3Pt

👏 Congratulations to @GlobalAiDotCom on achieving NVIDIA Exemplar Cloud status on NVIDIA GB300 NVL72 systems for training workloads. This validation reflects close engineering collaboration and gives enterprises confidence that demanding AI training workloads can perform at scale on Global AI's AI Cloud. Learn more ⬇️

Measuring AI performance requires new standards. Join NVIDIA, Nscale, and VAST Data at #RAISESummit for a fireside chat on Time to First Token and explore how accelerated computing, high-performance data platforms, and AI infrastructure work together to maximize performance, efficiency, and ROI. 📅 July 8 | Paris 🔗 nvda.ws/4y0EPlG

📣 Europe is entering a new era of AI-powered industry. At #RAISESummit, join NVIDIA's Paolo Guglielmini to explore the AI 5-Layer Cake — a framework for building AI at scale, from infrastructure to intelligence, and enabling the next generation of AI factories. 📅 July 8 | 8:40 a.m. CEST 🔗 Learn more: nvda.ws/4eMAHOH

📺 Watch the full recap video. nvda.ws/44dCgz2

💡 Purpose-built infrastructure for the era of agentic AI is here. Our #NVIDIAGTC Taipei recap traces AI's move into the agentic era — agents that plan, use tools, and act on nanosecond timescales — and the infrastructure behind it: ✅ NVIDIA Vera Rubin - 10x the throughput of Blackwell ✅ Vera CPU, the CPU for agents - already adopted by @AnthropicAI, @CoreWeave, @OpenAI, and @Oracle ✅ DSX platform - helping to build AI factories optimized for lowest token cost Thank you to Taiwan and all our ecosystem partners including @ASUS, @HonHai_Foxconn, GIGABYTE, Pegatron, Quanta, Wistron, and Wiwynn.

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.