Industry Briefing
A single destination for timely, editor-curated robotics news from around the world.
Startup Wants to Run AI Inference From Space
The rapid growth of large language models is driving a global surge in energy demand for data centers, prompting operators to seek alternative power sources. Among them is Orbital Inc., a Los Angeles-based startup that recently emerged from stealth mode to announce plans for space-based data centers. Backed by venture capital firm Andreessen Horowitz, Orbital aims to utilize solar energy from a constellation of small satellites in low Earth orbit to power AI inference workloads, such as chatbots. Orbital's founder and CEO, Euwyn Poon, emphasizes the limitations of terrestrial energy sources, stating, “There simply isn’t enough capacity here [on Earth], and the only way is up.” The company envisions a network of up to 10,000 satellites, each equipped with GPU server racks powered by solar panels. The first test of this concept is scheduled for 2027, with a prototype satellite launch aboard a SpaceX Falcon 9 rocket. While Orbital's approach aims to reduce launch costs and improve efficiency, it faces significant engineering challenges, including radiation effects on GPUs, thermal management in space, and maintenance difficulties. Experts like Dr. Amit Verma from Texas A&M University caution that the operational feasibility of such systems will depend on the specific applications they support. Despite these hurdles, Orbital plans to finalize its satellite designs by 2026 and establish a manufacturing facility by 2028, with the goal of tapping into major AI firms as customers. Poon remains optimistic about overcoming technical challenges, asserting that their engineering efforts will pave the way for the future of space-based data processing.
Data-center Space Ai Inferencing
Advantech Releases Edge AI Inference System
Nvidia and Fort Robotics have announced a collaboration aimed at enhancing the performance of robotic and autonomous mobile robot applications. This partnership focuses on ensuring that these technologies operate with deterministic performance and meet Safety Integrity Level (SIL) certification standards. The initiative underscores the growing importance of safety and reliability in the rapidly evolving field of robotics, as industries increasingly integrate autonomous systems into their operations. By leveraging Nvidia's advanced computing capabilities alongside Fort Robotics' expertise in safety protocols, the two companies aim to set new benchmarks for performance and safety in robotic applications.
Factory / Control
ARBOR Technology Unveils ARES-2100 Series for Next-Generation Edge AI with Intel Core Series 3 Processors
A new technology aimed at enhancing industrial automation has been introduced, focusing on machine vision and lightweight edge AI inference workloads. This innovation is set to facilitate the deployment of smart factory applications, allowing for more efficient operations in manufacturing environments. The development comes in response to the growing demand for advanced automation solutions that can improve productivity and reduce operational costs. By leveraging cutting-edge AI capabilities, the technology enables real-time data processing and decision-making at the edge, minimizing latency and enhancing overall system performance. This advancement is expected to significantly impact the manufacturing sector, providing businesses with the tools necessary to adapt to the rapidly evolving landscape of industrial automation.
Rethinking AI TCO: Why Cost per Token Is the Only Metric That Matters
Traditional data centers have undergone a significant transformation, evolving from mere storage and processing facilities into advanced AI token factories. This shift has been driven by the increasing reliance on AI inference as the primary workload, marking a new era in data management. As of October 2023, these centers are now focused on generating intelligence rather than just handling data. This evolution reflects the growing demand for sophisticated AI capabilities, highlighting the critical role that modern data centers play in supporting the advancements in generative and agentic AI technologies. The transition is reshaping how businesses and organizations utilize data, emphasizing the importance of intelligence generation in today's digital landscape.
Hot French startup ZML releases free product to speed inference across lots of AI chips
ZML, a hot French AI startup endorsed by Turing Award winner Yann LeCun, has now released ZML/LLMD, software that could make running AI less costly.
AI Fundraising Startups AI inference Exclusive ZML
NVIDIA CEO Jensen Huang at Dell Technologies World: ‘Demand Is Going Parabolic, Utterly Parabolic’
NVIDIA has announced the launch of its Vera Rubin NVL72, a new AI inference technology that significantly reduces costs, operating at one-tenth the cost per token compared to previous models. This advancement allows agent sandboxes to run 50% faster than traditional CPUs, while enterprise data queries can achieve speeds up to three times faster with the Vera CPU. The introduction of this technology is set to benefit over 5,000 enterprises, enhancing their operational efficiency and data processing capabilities. This announcement comes as part of NVIDIA's ongoing efforts to innovate in the AI sector, aiming to provide businesses with more cost-effective and powerful tools for data management and analysis.
Microsoft unveils AI-focused mini PC "Surface RTX Spark Dev Box" featuring NVIDIA SoC.
At the recent Build 2026 conference, Microsoft unveiled its new AI-focused desktop PC, the Surface RTX Spark Dev Box. This powerful machine is equipped with NVIDIA's RTX Spark technology, delivering an impressive computational performance of up to 1 petaflop and featuring 128GB of memory. These specifications enable local inference and training of models with over 120 billion parameters. Additionally, the device comes pre-installed with a variety of development tools, catering to the needs of developers working in artificial intelligence and machine learning.
NVIDIA Enters Production With Dynamo, the Broadly Adopted Inference Operating System for AI Factories
NVIDIA has unveiled NVIDIA Dynamo 1.0, an open-source software designed for generative and agentic inference at scale, marking a significant advancement in artificial intelligence technology. The announcement was made today, highlighting the software's potential for widespread global adoption across various industries. This initiative aims to enhance the capabilities of AI systems, enabling them to generate and infer data more efficiently and effectively. By making Dynamo 1.0 open-source, NVIDIA seeks to foster collaboration and innovation within the tech community, encouraging developers and researchers to contribute to and expand upon the software's functionalities. This move aligns with NVIDIA's commitment to advancing AI technology and supporting its integration into diverse applications worldwide.
Leading Inference Providers Cut AI Costs by up to 10x With Open Source Models on NVIDIA Blackwell
In a groundbreaking development in artificial intelligence, various sectors are harnessing the power of tokens to enhance user interactions. From providing diagnostic insights in healthcare to facilitating character dialogues in interactive gaming and enabling autonomous resolutions in customer service, these AI-driven applications are revolutionizing how industries engage with users. This shift is occurring as organizations seek to improve efficiency and user experience by leveraging advanced AI technologies. The integration of tokens allows for seamless communication and problem-solving, showcasing the versatility of AI across different fields. As of October 2023, this trend is expected to continue evolving, further embedding AI into everyday interactions and services.
NVIDIA, Telecom Leaders Build AI Grids to Optimize Inference on Distributed Networks
At the NVIDIA GTC 2026 conference, major telecommunications operators from the United States and Asia highlighted the evolving role of telecommunications networks in the distribution of artificial intelligence as AI-native applications continue to expand their user base. This shift is driven by the increasing demand for efficient and scalable AI solutions that can support a growing number of agents and devices. The conference showcased innovative strategies and technologies that these operators are implementing to leverage their networks for enhanced AI capabilities, marking a significant step in the integration of AI into telecommunications infrastructure.
Your Next AI Query May Travel Where the Power Is
Nvidia, in collaboration with InfraPartners, Prologis, and the Electric Power Research Institute (EPRI), is set to launch a pilot project later this year to construct approximately 25 micro data centers near utility substations across five U.S. states. This initiative aims to address the growing energy demands of the AI industry, which is projected to consume 9 to 17 percent of the country’s electricity generation by 2030. By strategically locating these small data centers, each with a capacity of 5 to 20 megawatts, the project seeks to enhance flexibility in power consumption and optimize the use of available electricity. The approach involves shifting computational workloads to different substations based on real-time power availability, thereby alleviating pressure on overloaded substations and maximizing overall energy efficiency. With U.S. grid operators typically utilizing only 53 percent of their generation capacity, this strategy could significantly increase the effective power supply for data centers. As AI workloads evolve, particularly in inference tasks that require less intensive computational resources compared to training, the micro data centers can dynamically route workloads to where power is most accessible. The project, termed “distributed inference,” is expected to begin construction by the end of 2026, with the goal of demonstrating a new model for data center operations that aligns with the increasing demand for energy-efficient solutions in the tech industry.
Ai-data-centers Nvidia Epri Power-generation
NVIDIA Unlocks AI Compute at Scale, Inviting Capital Partners to Power the AI Infrastructure Buildout
As artificial intelligence transitions from model development to production inference, the demand for computing power is rapidly increasing. This surge is driving the establishment of continuously operating AI factories that are capable of generating tokens at scale. The shift towards these large-scale, multi-tenant accelerated computing environments is essential to meet the growing needs of AI applications. This transformation is occurring as organizations recognize the necessity for robust infrastructure to support the evolving landscape of AI technology. The move towards these advanced facilities is expected to enhance efficiency and scalability, enabling businesses to leverage AI more effectively in various sectors.
Why everyone from OpenAI to SpaceX is building their own chips (and turning up the heat on Nvidia)
Nvidia, a leader in the AI chip market, may soon face increased competition as OpenAI announces its development of a new custom inference chip named Jalapeño, in collaboration with Broadcom. This strategic move comes as OpenAI joins a growing list of tech giants, including Google, Apple, and SpaceX, who are seeking to reduce their reliance on a single supplier for critical technology. The initiative reflects a broader industry trend aimed at diversifying chip sources to mitigate risks associated with dependence on Nvidia. OpenAI's plans signal a significant shift in the competitive landscape of AI hardware, potentially reshaping the dynamics of the market.
AI a24 agility AI chips AI loops Anthropic
DEEPX and Sixfab Launch 'DEEPX AI HAT' to Drive Edge Physical AI on Raspberry Pi
A new partnership has emerged to enhance the Raspberry Pi ecosystem by providing high-performance, low-power AI acceleration. This initiative aims to eliminate the reliance on cloud services for real-time inference, particularly benefiting applications in robotics and smart automation. The collaboration has introduced a production-ready Starter Kit and software development kits (SDKs) designed to lower entry barriers for developers worldwide. By making these tools accessible, the partnership seeks to empower a broader range of innovators to integrate advanced AI capabilities into their projects, fostering growth and development within the global tech community.
Liquid AI's smallest model yet LFM2.5-230M beats models 4X its size at data extraction, can run 'anywhere'
Liquid AI, a company founded by former MIT computer scientists, has unveiled its latest AI language model, LFM2.5-230M, which is designed for efficient data extraction and local deployment on devices such as smartphones and laptops. Released today, this 230-million-parameter model is noted for its ability to run on various hardware platforms, outperforming larger models like Alibaba's Qwen3.5 and Google's Gemma 3 in specific benchmarks. Targeting developers and engineers, LFM2.5-230M operates under a dual-use commercial license, allowing free access for individuals and companies with annual revenues below $10 million, while larger enterprises must secure a paid agreement. The model distinguishes itself by utilizing the LFM2 architecture, enabling high inference speeds with a minimal memory footprint, making it suitable for edge computing. Liquid AI's launch reflects a broader industry shift towards architectural efficiency rather than sheer parameter counts, as major AI firms focus on models with hundreds of billions of parameters. The LFM2.5-230M is specifically tailored for lightweight data extraction tasks, allowing businesses to automate processes without relying on costly cloud services. In practical applications, the model has been successfully deployed in a humanoid robot, demonstrating its capability to process complex commands efficiently. Available immediately on platforms like Hugging Face, LFM2.5-230M aims to revolutionize how enterprises manage data extraction, moving away from traditional, rigid systems to more adaptable AI-driven solutions.
Technology
NVIDIA and AWS Collaborate to Bring AI to Production at Scale
NVIDIA has announced a collaboration with Amazon Web Services (AWS) aimed at enhancing the development of artificial intelligence systems. This partnership focuses on addressing the challenges of building AI at scale, which necessitates low-latency inference, rapid vector search capabilities, and effective GPU price-performance. The initiative is designed to provide infrastructure solutions that can expand without increasing operational complexity. This announcement comes as the demand for advanced AI technologies continues to rise, with organizations seeking efficient and scalable systems to leverage AI's potential. By combining NVIDIA's expertise in GPU technology with AWS's cloud services, the two companies aim to streamline the AI development process, making it more accessible for businesses looking to implement AI solutions.
New Server Hopes to Break Through AI’s “Memory Wall”
Majestic Labs, an AI hardware startup, is addressing the memory limitations of large language models (LLMs) with its upcoming server, Prometheus, set to launch in 2027. This innovative server will feature up to 128 terabytes of memory, significantly surpassing the capabilities of Nvidia’s current offerings. Co-founder Sha Rabii emphasizes that this substantial memory increase will enhance performance and efficiency, particularly as models grow larger. Prometheus employs a unique DRAM-centric architecture, utilizing LPDDR6 memory and a proprietary memory interface with miniature copper cables that allow for greater memory placement flexibility. This design aims to overcome the “memory wall” that hampers LLM performance, providing a memory bandwidth of up to 25.6 terabytes per second. To complement its memory capabilities, Prometheus will incorporate the Ignite AI processing unit, which combines ARM application cores with RISC-V vector and tensor cores on a single chip. This integration allows for seamless handling of LLM inference tasks without the need for processor handoffs. Majestic Labs is also focused on ensuring compatibility with existing AI frameworks like PyTorch and OpenAI’s Triton, allowing customers to run their models without modifications. The server, designed in compliance with the Open Compute Project, will be modular, enabling future memory upgrades. Despite the advanced technology, Majestic Labs aims to offer competitive pricing by leveraging DRAM instead of more expensive high-bandwidth memory. Rabii claims that this approach could reduce customer capital expenditures and power consumption significantly, potentially by 10 to 50 times, depending on the workload.
Memory Server Ai-accelerators Performance
New Data Shows NVIDIA Blackwell Ultra Delivers up to 50x Better Performance and 35x Lower Costs for Agentic AI
The NVIDIA Blackwell platform has gained significant traction among top inference providers, including Baseten, DeepInfra, Fireworks AI, and Together AI, achieving reductions in cost per token by as much as 10 times. Building on this success, NVIDIA has introduced the Blackwell Ultra platform, which aims to enhance these cost efficiencies further. This development reflects NVIDIA's commitment to advancing AI technology and providing more affordable solutions for inference tasks, thereby supporting the growing demand for cost-effective AI applications in various industries.
NVIDIA BlueField-4 Powers New Class of AI-Native Storage Infrastructure for the Next Frontier of AI
NVIDIA has unveiled its latest innovation, the BlueField®-4 data processor, which is integral to the comprehensive NVIDIA BlueField platform. This advanced technology supports the newly launched NVIDIA Inference Context Memory Storage Platform, marking a significant advancement in AI-native storage infrastructure. The announcement was made today, highlighting NVIDIA's commitment to enhancing data processing capabilities for the evolving demands of artificial intelligence applications. The BlueField-4 processor is designed to optimize performance and efficiency, enabling organizations to harness the full potential of AI-driven data management and storage solutions.
Silicon Valley stunned as China’s DeepSeek-R1 surpasses US AI giants, undercuts costs
Tensions escalated in Silicon Valley this week following the launch of DeepSeek's R1 model, a Chinese AI technology that has surpassed leading U.S. companies such as OpenAI, Meta, and Anthropic in third-party performance benchmarks. The unease began earlier with the release of DeepSeek v3, which already outperformed Meta’s Llama 4. The R1 model, unveiled on January 20, has been noted for its significant improvements in model inference capabilities, raising concerns among American tech firms about competition and innovation in the rapidly evolving AI landscape.
News Feed
How NVIDIA’s Inference Software Stack Powers the Lowest Token Cost
As organizations transition from experimental AI pilots to fully operational AI factories, there is a significant shift in infrastructure decision-making. This change, observed in late 2023, emphasizes the importance of cost efficiency, focusing on the cost per token rather than just peak chip specifications. Companies are now prioritizing how many useful tokens can be generated per dollar spent, per watt of energy consumed, and within specific latency requirements. This strategic pivot aims to enhance the overall performance and affordability of AI systems, ensuring they can meet the growing demands of the market while maintaining efficiency.
Bentley prices drop to 270,000; second-hand luxury fuel car prices plummet; Cambrian loses 70 billion in market value in a
OpenAI has reportedly achieved a significant technological breakthrough, successfully reducing the operational costs of its AI model inference by over 50% through a series of system optimizations. This development was disclosed by engineers within the company on June 30. Meanwhile, in the automotive sector, the price of second-hand luxury fuel vehicles has seen a dramatic decline, with a Bentley listed at approximately 26,800 yuan and a Porsche Macan at around 15,000 yuan as of July 1 in Qingdao. This drop is attributed to a notable increase in depreciation rates, which reached 30% in May alone, a stark contrast to the previous annual average of 30%. In the realm of artificial intelligence, Anthropic announced that its Claude model is now fully available on Microsoft Foundry, allowing enterprise users to deploy the model within the Azure environment. Additionally, Amazon Web Services revealed plans to invest $1 billion to establish a new AI department aimed at assisting clients in building AI systems. In financial news, Warren Buffett has paused his annual donations to the Bill and Melinda Gates Foundation, pending the outcome of an investigation related to the late Jeffrey Epstein. Furthermore, the U.S. Federal Reserve Chairman emphasized the ongoing challenge of high inflation levels in the country, indicating a commitment to maintaining price stability. As for the tech industry, Meta is reportedly planning to sell excess AI computing resources to external clients, intensifying competition with major cloud service providers.
Anthropic's Claude model officially launched on Microsoft Foundry.
On June 29, Anthropic announced the official availability of its Claude model on Microsoft Foundry, specifically within the Azure AI Foundry platform. This development allows enterprise users to deploy and utilize the Claude model directly in the Azure environment, leveraging existing Azure authentication, billing, and governance systems. Currently, the platform supports Claude Opus 4.8 and Claude Haiku 4.5, which include features such as prompt caching and extended thinking capabilities. These functionalities are designed for applications in programming, intelligent agents, and complex reasoning scenarios. Anthropic also noted that users can choose to run their inference either on Azure or in an Anthropic-hosted environment. Looking ahead, both companies plan to gradually align their features and models for enhanced compatibility.
NVIDIA Confidential Computing to Help Expand Apple’s Private Cloud Compute
NVIDIA has announced that its GPUs equipped with Confidential Computing technology are now being utilized for confidential inference in Apple’s Private Cloud Compute (PCC). This development marks a significant expansion of Apple’s cloud capabilities, extending beyond its own data centers to include Google Cloud. The announcement was made during Apple’s annual Worldwide Developers Conference (WWDC), where the company showcased its latest advancements and innovations aimed at enhancing data security and privacy for users. This collaboration with NVIDIA is expected to bolster Apple’s commitment to maintaining user confidentiality while leveraging cloud resources effectively.
Summary of Microsoft's "Build 2026" announcement: expanding development platforms, safety measures, and hardware.
At the recent Build 2026 conference, Microsoft announced significant enhancements to its development platform aimed at supporting the creation of autonomous AI agents and ensuring safety controls. The tech giant introduced a new inference model named MAI, alongside high-performance local development terminals and a roadmap for quantum computing initiatives. These advancements reflect Microsoft's commitment to advancing AI technology and improving developer tools, positioning the company at the forefront of innovation in the tech industry.
Horizon Robotics Open-Sources HoloMotion-1: 4B Robot Cerebellum Model Running 300FPS on Device
Horizon Robotics has unveiled its latest innovation, HoloMotion-1, a sophisticated robot cerebellum model featuring 4 billion parameters. This advanced technology enables real-time inference at an impressive rate of 300 frames per second on edge devices, significantly enhancing the control capabilities of whole-body humanoid robots. The launch, which took place recently, represents a substantial advancement in robotics, aiming to improve the functionality and responsiveness of humanoid robots in various applications. By leveraging this cutting-edge model, Horizon Robotics seeks to push the boundaries of robotic intelligence and performance, paving the way for more versatile and efficient robotic systems in the future.
AI
Predicting the World Without Future Frames: BeingBeyond Launches Next-Generation Embodied Intelligence Model Being-H0.7
BeingBeyond has introduced its latest innovation, the Being-H0.7 embodied intelligence model, which revolutionizes the approach to video generation. This new model distinguishes itself by eliminating the need for traditional rendering of future frames, allowing for accurate physical reasoning and dynamic predictions. As a result, Being-H0.7 not only cuts down on training expenses but also ensures a high inference speed. This advancement marks a significant step forward in the field of artificial intelligence, promising enhanced efficiency and effectiveness in various applications.
Embodied Intelligence Robotics AI Machine LearningRobotics needs a service framework.
RSF defines a common language for robot service capability, lifecycle operations, certification pathways, and service-provider networks.
