🧠 AWS and NVIDIA Commit 2 Million Additional GPUs to Next-Generation Physical AI Infrastructure

• ●AWS and NVIDIA announced a deal to deploy 2 million additional GPUs, targeting the infrastructure demands of agentic and physical AI workloads at scale.
• ●The partnership pairs NVIDIA’s next-generation compute with AWS’s cloud backbone — a combination designed to handle the high-throughput, low-latency requirements that physical AI inference and training impose.
• ●Watch for downstream effects on robotics simulation pipelines and embodied-AI training costs as this capacity comes online.
• ●🔎 Read More →
• What matters: A 2-million-GPU commitment from AWS and NVIDIA signals that physical AI infrastructure is now a hyperscaler-level priority, not a niche research bet.

🧪 Technology, Research & Innovation

🧠 WALL-SS Scales Robot World Models to Long Horizons Using Next-Scale Autoregression

• ●WALL-SS introduces a next-scale autoregressive architecture for generative world models, enabling robots to predict environment evolution across long, flexible time horizons — not just short clips.
• ●The unified formulation explicitly links actions to consequences and supports continuous interaction, addressing a core limitation of prior clip-level prediction models used in robot planning and policy evaluation.
• ●If the approach holds up under broader benchmarking, it could meaningfully reduce the data burden for robot learning by substituting high-quality simulated rollouts for real-world demonstrations.
• ●🔎 Read More →
• What matters: Long-horizon world models that couple actions to outcomes are the missing link between robot simulation and deployable planning systems.

🤖 Bimanual Robot Learns Juggling Patterns on Physical Hardware in Minutes, Bridging Sim-to-Real Gap

• ●Researchers demonstrated an online learning framework that lets a bimanual robot acquire diverse juggling patterns directly on physical hardware within minutes — even when the simulation model diverges significantly from reality.
• ●The key finding: an imperfect model still provides enough structure to guide rapid on-robot adaptation, challenging the assumption that sim-to-real fidelity must be high before hardware learning is viable.
• ●This philosophy — use approximate models as scaffolding, not ground truth — could accelerate deployment of dynamic manipulation skills in warehouse and assembly settings where precise simulation is impractical.
• ●🔎 Read More →
• What matters: Robots that learn dynamic skills on hardware in minutes — not days — reframe sim-to-real fidelity from a prerequisite into an optional accelerant.

🚀 Product, Hardware & Model Launches

🧠 NVIDIA Launches New Jetson Edge Platform Targeting Physical AI Deployment

• ●NVIDIA unveiled a new Jetson edge platform purpose-built for physical AI applications, extending its embedded compute lineup to meet the inference demands of robots, drones, and autonomous systems at the edge.
• ●Edge-native physical AI requires sustained low-latency inference without cloud round-trips — the new Jetson platform is positioned to handle that constraint where prior generations hit throughput ceilings.
• ●Combined with the AWS GPU infrastructure announcement, NVIDIA is now staking out both ends of the physical AI compute stack: cloud training and edge deployment.
• ●🔎 Read More →
• What matters: NVIDIA is building a closed-loop physical AI stack — train in the cloud, deploy at the edge — and the new Jetson platform is the deployment half of that bet.

🚗 AMD EPYC Processors Power Kodiak AI’s New Autonomous Driving Platform

• ●AMD announced that its EPYC processors are accelerating Kodiak AI’s new autonomous driving platform, marking a direct challenge to NVIDIA’s dominance in AV compute hardware.
• ●EPYC’s high core count and memory bandwidth make it a credible fit for the sensor fusion and perception pipelines that AV stacks run continuously — workloads that are CPU-bound as much as GPU-bound.
• ●Kodiak AI’s platform choice signals that AV compute is becoming a competitive market, with AMD now a named alternative for production-grade autonomous driving deployments.
• ●🔎 Read More →
• What matters: AMD landing a named AV platform win with EPYC confirms that autonomous driving compute is no longer a single-vendor market.

💰 Business, Startups & Investment

🤖 Gatik Raises $200M to Scale High-Frequency Autonomous Trucking on Regional Routes

• ●Gatik closed a $200M funding round to expand its autonomous trucking operations, with capital earmarked for scaling its model of high-frequency regional routes between distribution centers and retail stores.
• ●The short, repeatable route model reduces operational domain complexity — a deliberate constraint that lets Gatik achieve commercial reliability faster than long-haul AV trucking competitors.
• ●At $200M, this round gives Gatik the runway to densify existing corridors and enter new regional markets before the long-haul AV players reach commercial scale.
• ●🔎 Read More →
• What matters: Gatik’s $200M bet on constrained regional routes is a direct argument that operational simplicity — not technical ambition — is the fastest path to AV trucking revenue.

🧠 NVIDIA Jetson Orin Nano 2 Brings Physical AI Inference to Drones and Robots

• ●NVIDIA’s Jetson Orin Nano 2 is now targeting drones and robots as primary deployment platforms, bringing physical AI inference capability to form factors where power and size constraints previously limited compute.
• ●The Orin Nano 2’s positioning in the Jetson lineup fills the gap between low-power microcontrollers and the full Orin NX — giving developers a mid-tier option for edge AI without sacrificing model complexity.
• ●For robotics startups, accessible edge inference hardware at this price-performance point lowers the barrier to shipping physical AI products without custom silicon investment.
• ●🔎 Read More →
• What matters: Jetson Orin Nano 2 makes physical AI inference accessible to drone and robot developers who previously had to choose between capability and power budget.

⚡Infrastructure Scale::The AWS–NVIDIA 2-million-GPU commitment marks the moment physical AI moved from research priority to hyperscaler infrastructure category.

⚡World Models Maturing::WALL-SS’s long-horizon autoregressive approach closes a critical gap between robot simulation and real-world planning — watch for benchmark results in the next 60 days.

⚡On-Robot Learning::Minutes-to-competency juggling results challenge the field’s sim-fidelity orthodoxy and point toward faster hardware-in-the-loop training pipelines.

⚡Compute Competition::AMD’s EPYC win at Kodiak AI and NVIDIA’s dual Jetson launches confirm that AV and robotics compute is now a multi-vendor market with real architectural choices.

⚡Capital Discipline::Gatik’s $200M raise on a constrained regional-route model raises the question every AV investor should be asking: is operational simplicity now worth more than technical ambition at the funding table?