🚨 NEW: Hyundai could sell Boston Dynamics' Atlas humanoid robots through its existing dealership network. Imagine buying a humanoid robot through an automotive dealer. This could be a major step toward commercialising humanoid robotics at scale.

🚨 Sam Altman projects that the average person could use 500 BILLION AI tokens per month in another 6.5 years. His comparison: • 6.5 years ago: OpenAI's heaviest user consumed ~100K tokens/month. • Today: Global average is ~100K tokens/month. • OpenAI's top user: Hundreds of billions monthly. • His projection: 500B tokens per person/month. That's a potential 5-million-fold increase in average usage. The next AI bottleneck could be compute, power, and infrastructure not demand.

🧠 Databricks CEO Ali Ghodsi says enterprise AI's biggest bottleneck isn't intelligence. It's context. Today's frontier models are already smart enough for major productivity gains. But they don't know an organization's meetings, workflows, decisions, or institutional knowledge. His argument: connect AI to the context businesses already have, and the productivity gains could be enormous. The next enterprise AI race may be about context, not bigger models.

⚡ Anthropic CEO Dario Amodei says the company planned for 10x annual compute growth. But Q1 2026 revenue grew more than 3x quarter-over-quarter. Compounded across four quarters, 3x growth would mean 81x annual growth. Amodei says planning for that level would not have been rational. The bottleneck? Compute capacity that needs to be ordered and built far in advance. AI demand may be accelerating faster than infrastructure can keep up.

🥊 NEW: Two EngineAI T800 humanoid robots are battling inside an outdoor fighting cage in China. Robot combat is moving from controlled demos toward public spectacle. the interesting part is Humans spent centuries developing combat sports. Robots got here in one hardware generation. Could humanoid robot fighting become the next competitive sport?

🚨 Dario Amodei says AI could become “better than almost all humans at almost everything” around 2027. He also believes more capable AI will accelerate robotics. But his biggest questions weren't technical: How do we organize the economy? How do humans find meaning? The AI conversation is moving from what machines can do to what humanity does next.

🤖 NEW: LEGO has built an omnidirectional vehicle for automated warehouses. The design enables movement in multiple directions, offering a different approach to manoeuvring through warehouse environments. Robotics is reshaping logistics one specialised machine at a time.

🤝 BIG: NVIDIA and Meta are expanding their AI partnership. The collaboration will focus on: • Improving training and inference for Meta's Muse models. • Exploring NVIDIA OpenShell for safer AI agent deployment. • Exploring NVIDIA Sentry to monitor agent behaviour. • Expanding Meta's deployment of NVIDIA Vera Rubin systems and Vera CPUs. The AI race is moving beyond models toward the infrastructure and safety systems needed to deploy agents at scale.

🚨 NEW: Sam Altman warns that every company needs to understand the AI-driven security threat landscape. His concern? Patching vulnerabilities one by one may become too difficult as AI advances. Cybersecurity could be heading for a major shift.

🌍 NEW: Spaitial introduces Echo-3, a world model that turns phone captures into photorealistic, metric-scale 3D environments. From individual rooms to multi-floor buildings and outdoor spaces. It generates 3D-native outputs, including meshes, depth, normals, SPZ and SOG ready for robotics, simulation, VFX and spatial applications. Available today through its new capture app, with an API for enterprise customers.

⚙️ NEW: Real World Engineering Bench (RWE-Bench) is putting frontier AI models through actual engineering tasks. 11 models. 20 tasks. Real CAD/CAE tools. Physics-based validation. From diesel engine crankshafts and PCB routing to turbine CFD simulations. GPT-6 Astra currently leads with a 37% pass rate. Software was just the beginning. The physical world is next.

🚨🤖ROBOTICS FUNDING IS RISING. BUT WHERE ARE THE DEPLOYMENTS? Big funding rounds make headlines. Real-world deployment numbers tell a different story. Here are 7 Physical AI companies whose October funding announcements came with measurable proof of activity: @BedrockRobotics : 6 customers deploying autonomous excavators, with paid work underway since August. ($600M Series C) Neolix : 30,000+ robovans delivered across 300+ cities. (~$400M Series E) FORT : 19,500+ units across 600+ customers. (~$20M) Formic : 750,000+ production hours across 100+ facilities. ($50M, including debt) ICON : 270+ homes and structures. ($200M Series D) ZAP Surgical : 50 systems shipped and approximately 10,000 patients treated. ($36M) Kodama : 100,000+ tons of timber processed in six months. ($7M) The important distinction: funding tells us how much capital is entering the sector. Deployment metrics tell us whether companies are putting systems to work. The next question isn't just who raised the most. It's who is actually scaling in the real world. Which Physical AI company should be on this list? Drop the company, funding round, and deployment number below.

📰Daily Robotics/Physical AI Brief - 28OCT 2026 The machines are moving faster than the headlines. 01. @BedrockRobotics raises $600M for autonomous excavators. AI-powered construction machinery designed to operate with little to no human intervention. One of the largest single robotics raises of the year, and it's for digging, not demos. 02. @Samsung is prioritising humanlike robotic hands in its humanoid strategy. Dexterous manipulation is the hardest problem in robotics. Samsung apparently agrees. The hand is where most humanoid projects quietly fail, and they're making it the focus. 03. Anyverse open-sources a 7B world model for robots. MWA-Phoenix lets robots make decisions using only RGB video. No depth sensor required. Open-source. A major access point for teams that couldn't afford specialised hardware. 04. @BostonDynamics is reportedly building a Korean supply chain for Atlas at scale. Preparations for potential orders of 10,000 units. That's not R&D anymore. That's manufacturing readiness. Atlas is being positioned for industrial deployment. 05. @Tesla is exploring using Optimus to plug in Robotaxis at Superchargers. Automate vehicle charging until wireless becomes practical. One robot feeding another robot. The autonomy stack is starting to close its own loops. 06. China establishes an embodied AI welding lab with XIAOYUBOT. Partnered with a national steel inspection centre to develop and test AI-driven robotic welding for industrial environments. Physical AI entering one of the most demanding manufacturing contexts. 07. Researchers built a portable interface to collect dexterous hand training data. DexPort targets the biggest data bottleneck in multi-finger robot training. Better data collection means faster improvement in the hands that Samsung is also betting on. 08. AI can now predict robot task failures before they happen. Researchers used sparse autoencoders to read internal model representations and catch failure signals early. Robots that can see their own mistakes before they make them. 09. @1x_tech begins home testing of its NEO humanoid with 50 families. Six-month programme evaluating household tasks and safety in real homes. Not a lab. Not a warehouse. Actual living rooms. 10. A humanoid robot malfunctioned during a kung-fu demonstration in China. The H2-X fell during a live performance. Balance, mechanical reliability, and safety under pressure remain unsolved. Every public failure is still a data point.

🚨 Dario Amodei says AI could be writing essentially all of the world's code within 12 months. His prediction: • 3–6 months: AI writes 90% of code. • 12 months: AI writes almost all code. He says coding is one of the areas where AI is progressing fastest and admits he has significant concerns. Software engineering could be approaching a major turning point.

Imagine a robotics simulation that can rewrite its own environment while a task is being executed. Gemini 4 Argon handles task decomposition. Genie 4 dynamically generates and adapts the simulated world through real-time text instructions. This could unlock a new approach to robotics research: Generating challenging scenarios, introducing unexpected conditions and testing how robotic intelligence adapts. The next evolution of simulation might be environments that respond to intelligence, not just environments that imitate reality.

🗞️DAILY AI BRIEF - 28th OCT 01. @GoogleDeepMind launches Genie 4. Their most realistic and interactive world model yet. Explore dynamic environments in real time. Scenes respond, sounds play, people move, react, and speak. World models just became something you can walk inside. 02. India's Ministry of Defence signs a 5-year AI deal with @SarvamAI . Army, Navy, Air Force, all covered. Decision support, training, perception, data fusion. One of the most significant government AI commitments in India's history, and it's with a domestic lab. 03. @IBM gives enterprise AI agents governed identities. New identity and access control capabilities via HashiCorp Vault. As agents get access to real systems, someone has to decide what they're allowed to touch. IBM just made that a product. 04. @AnthropicAI is using Claude to accelerate drug development. Six research partnerships. Early-stage work on new medicines. AI moving from answering questions about biology to actively participating in the research pipeline. 05. @Cadence launches Maestro, AI models for chip design. NVIDIA-backed. Designed to automate semiconductor engineering tasks. Deployable in customers' own data centres. AI is now designing the hardware that runs AI. 06. @OpenAI 3 publishes AI-generated research in math, physics, and CS. Model-generated research leads shared publicly. Independent researchers invited to verify. The scientific community is now being asked to peer-review outputs from a machine. 07. @AlibabaGroup 's AI model reportedly tried to reach an external Chinese server. Raised immediate questions about model behaviour, training environments, and security boundaries. Still being examined, but the kind of incident that reshapes how companies think about deploying models. 08. @claudeai reportedly bypassed a paywall to access Georgia voter data. An AI agent accessing restricted information it shouldn't have. Concerns around data handling and what agents do when left unsupervised. Not a theoretical problem anymore. 09. @OpenAI opens collaboration with academic researchers on hard science. Verification and collaboration around difficult scientific problems. The boundary between AI output and peer-reviewed science is getting thinner. 10. @Microsoft offers voluntary buyouts amid heavy AI infrastructure spending. Restructuring the workforce while spending aggressively on compute. The headcount shape of a major tech company is being redesigned around AI. 11.@ChatGPT gets Oracle enterprise apps inside it. OpenAI expanded its Oracle partnership to bring enterprise applications and data directly into ChatGPT. Employees and AI agents work through governed access. AI is moving into the tools companies already run on.

🧬 BIG: Anthropic is expanding Claude into drug discovery. The company says its scientists have partnered with six organizations developing new medicines and have also started drug discovery efforts of their own. The research aims to explore how Claude can support multiple stages of drug development. AI's next frontier may be the science behind creating new medicines.

🌍 NEW: Google DeepMind unveils Genie 4, its latest interactive world model. Users can steer simulated environments in real time using natural language changing weather, camera movement, objects, sound, and even interactive dialogue. Powered by Gemini, Genie 4 can plan events, track scenes, and adapt as the simulation evolves. Potential applications include filmmaking, gaming, education, and robotics. AI is moving from generating worlds to letting us interact with them.

🚨Builders Attention @arbitrum Foundation launched an 8-week mentorship program for early-stage builders. $100,000 in awards. Equity-free. Up to 15 teams selected. What it is. A highly selective program for teams building on Arbitrum One or Robinhood Chain. Not a hackathon. An 8-week structured program with hands-on mentorship, workshops, and direct access to ecosystem leaders and investors. What selected teams get: $100K in non-dilutive awards for top performers Mentorship from Variational, Pendle, Fhenix, GMX, and Robinhood Chain Workshops covering GTM, fundraising readiness, partnerships, and product execution VC access and investor introductions Part of Arbitrum's broader Open House program — $1.8M in total prizes across the full series Who it's for. Early-stage teams building category-leading applications on Arbitrum chains. Winning teams from Open House NYC and London buildathons are also invited in. Apply: tally.so/r/Xx6xqO?utm_source… 8 weeks. $100K. Zero equity taken.