NVIDIA & Oracle to Build DOE’s Largest AI Supercomputer with 100,000 Blackwell GPUs

NVIDIA and Oracle have announced a collaboration to build the U.S. Department of Energy’s largest AI supercomputer. The project will involve two new systems at Argonne National Laboratory, named Solstice and Equinox. The Solstice system will feature a record-breaking 100,000 NVIDIA Blackwell GPUs, while the Equinox system will include 10,000. Together, these systems are expected to deliver a combined 2,200 exaflops of AI performance. The supercomputers will be used to accelerate scientific discovery and research in areas such as healthcare and materials science. This initiative is part of a broader effort to advance the United States’ leadership in artificial intelligence.

Sources:

Uber, NVIDIA, and Stellantis Partner for 100,000-Vehicle Global Robotaxi Fleet

Uber, NVIDIA, and Stellantis have announced a major collaboration to scale a global Level 4 autonomous vehicle network. Uber plans to expand its autonomous fleet to 100,000 vehicles, with the rollout beginning in 2027. The partnership will utilize NVIDIA’s DRIVE AGX Hyperion 10 autonomous vehicle platform and its associated AI software. Automaker Stellantis will be one of the first to provide vehicles for the service, committing an initial 5,000 units for Uber’s robotaxi operations, with production targeted for 2028. The collaboration also involves Foxconn for hardware and systems integration, aiming to create a scalable and efficient solution for autonomous ride-hailing services worldwide.

Sources:

Fincantieri and NextGeo Collaborate on Autonomous Marine Drones for Energy Sector

Italian shipbuilding group Fincantieri, through its subsidiary Ingegneria dei Sistemi (IDS), has signed a memorandum of understanding with Next Geosolutions (NextGeo) to develop unmanned surface vehicles (USVs). The strategic collaboration will focus on creating USVs for civil applications within the oil & gas and renewable energy industries. The partnership will combine the companies’ expertise to define specifications for new USV platforms designed for open-sea operations such as environmental monitoring and offshore infrastructure inspection. A key focus will be the evolution of the Surface Advanced Naval Drone (SAND) project, developed by IDS, and the creation of a new USV model to meet market demands.

Sources:

Adobe Supercharges Creative Cloud with Third-Party and In-House AI Models

Adobe has rolled out a suite of new AI-powered features and integrations for its Creative Cloud platforms, including Photoshop, Premiere, and Lightroom. A key update is the introduction of conversational AI assistants in applications like Adobe Express and Photoshop, allowing users to perform tasks using natural language. The company is also integrating AI models from partners such as OpenAI, Google, Luma AI, and Runway, alongside its own Firefly models. New features in Photoshop include Generative Fill with model selection from partners, Generative Upscale leveraging Topaz Labs’ technology, and a ‘Harmonize’ feature for blending elements into new scenes. Adobe is also launching Firefly Custom Models, which enables enterprises to develop proprietary generative AI models.

Sources:

Agentic AI Browsers Face Growing Threat from Prompt Injection Attacks

The emergence of AI-powered browsers with agentic capabilities, such as OpenAI’s Atlas, introduces new security vulnerabilities, particularly from prompt injection attacks. Prompt injection occurs when an AI bot is manipulated by hidden commands within content it processes, causing it to perform unintended actions. Security experts warn that as AI becomes more agentic, allowing it to take actions on a user’s behalf, the risk of data exfiltration and other malicious activities increases. While some bots can detect direct injection attempts, indirect prompt injection remains a significant challenge that may not be entirely solvable. Recommended security measures include limiting the capabilities of AI systems, sandboxed code execution, and maintaining human oversight.

Sources:

Confluent Launches AI and Private Cloud Solutions for Real-Time Data Streaming

Confluent has unveiled Confluent Intelligence, a fully managed service designed to provide real-time, context-rich data to AI applications. This new offering includes a Real-Time Context Engine that streams structured data to AI agents and applications, aiming to solve the challenge of AI models working with stale or fragmented data. The service is built on Confluent Cloud and is designed to help organizations build and scale event-driven AI systems with reliable and secure data. In a related announcement at their Current 2025 conference, Confluent also launched Confluent Private Cloud. This offering aims to simplify the deployment, management, and governance of data streaming on private infrastructure, targeting industries with strict regulatory and security requirements by bringing cloud-native features to on-premises environments.

Sources:

dbt Labs Open-Sources MetricFlow to Standardize AI and Data Metrics

dbt Labs has announced that it is open sourcing MetricFlow, its SQL generation tool that powers the dbt Semantic Layer, under an Apache 2.0 license. This move is part of dbt Labs’ commitment to the Open Semantic Interchange (OSI) Initiative, which aims to create vendor-neutral standards for semantic data exchange across various analytics and AI platforms. By open-sourcing MetricFlow, dbt Labs intends to provide a transparent and extensible engine for the community to build more trustworthy AI applications with consistent and reliable metrics. The initiative also includes making MetricFlow’s JSON-based metadata layer available, which can serve as a universal schema for data understanding across different tools, even without adopting the entire MetricFlow engine.

Sources:

HashiCorp Previews Project Infragraph for Agentic Infrastructure Automation

At its HashiConf 2025, HashiCorp, now part of IBM, introduced Project Infragraph, a new strategic initiative aimed at creating a real-time infrastructure graph. This project is designed to be the foundation for “agentic infrastructure,” where systems can autonomously observe, reason, and act. Project Infragraph will be delivered as a capability within the HashiCorp Cloud Platform (HCP) and is intended to provide a unified control plane with relational visibility across hybrid and multi-cloud environments. The goal is to address challenges related to visibility, ownership, and data governance in complex infrastructures. A private beta for Project Infragraph is expected to open in December 2025.

Sources:

Google’s Angular Team Releases Open-Source Tool to Score AI-Generated Code

The Angular team at Google has released a new open-source tool called Web Codegen Scorer. This tool is designed to provide quantifiable metrics on the quality of AI-generated frontend code. It evaluates the generated code against frontend development best practices, including framework guidelines, accessibility, and security. The goal of the Web Codegen Scorer is to help developers trust and verify the production-readiness of code produced by large language models. The tool was developed to address the challenge of ensuring that the speed of AI-driven development does not compromise code quality and reliability.

Sources:

IBM Hits Quantum Error Correction Milestone Using Commercial AMD Chips

IBM has made a significant advancement in quantum computing by successfully running a complex error-correction algorithm on commercially available AMD chips. This is a crucial step towards building fault-tolerant quantum computers, as it demonstrates that expensive, custom-built hardware is not necessary to handle quantum errors. The IBM researchers reported that their quantum algorithm ran ten times faster than required to keep pace with a quantum computer. This development is considered a breakthrough in error correction and puts IBM a year ahead of schedule on its roadmap to build a large-scale quantum computer by 2029. The ability to use off-the-shelf hardware for quantum error correction could significantly reduce the cost and complexity of building practical quantum computers.

Sources:

NVIDIA has announced NVQLink, an open system architecture designed to tightly couple quantum processors with GPU supercomputers. This new interconnect will allow for high-speed communication between quantum and classical computers, which is essential for tasks like quantum error correction and hybrid quantum-classical applications. Several U.S. national laboratories and 17 quantum hardware builders are already backing the new system. NVQLink is integrated with NVIDIA’s CUDA-Q software platform, enabling researchers to build and test applications that utilize both quantum processors and GPUs. NVIDIA’s CEO, Jensen Huang, has referred to NVQLink as the “Rosetta Stone” that will connect quantum and classical supercomputers, heralding the beginning of the quantum-GPU computing era.

Sources:

Malware Campaign on npm Targets AWS, Google Cloud, and Azure Developer Credentials

A sophisticated malware campaign has been identified on the npm registry, targeting developers’ credentials for major cloud platforms. Security researchers discovered 10 malicious packages designed to steal authentication keys for AWS, Google Cloud Platform, and Azure. The attack utilizes a multi-stage process to harvest sensitive information from developers’ systems, including configuration files for Kubernetes and Docker. This allows attackers to gain access to enterprise cloud infrastructure, potentially leading to data breaches and service disruptions. The malware also attempts to bypass multi-factor authentication by stealing browser session cookies. The malicious packages, which were available for over four months and downloaded thousands of times, employed multiple layers of obfuscation to hide their payload.

Sources:

Python Software Foundation Rejects $1.5M US Grant Due to Restrictive DEI Clause

The Python Software Foundation (PSF) has withdrawn its application for a $1.5 million grant from the U.S. National Science Foundation (NSF) due to restrictive clauses regarding Diversity, Equity, and Inclusion (DEI) programs. The grant was intended to fund projects aimed at enhancing the security of Python and the Python Package Index (PyPI). However, the grant conditions required the PSF to affirm that it would not operate any programs that promote DEI, a restriction that would apply to all of the foundation’s activities. The terms also included a ‘claw back’ provision, which would have allowed the government to reclaim funds if the PSF was found to be in violation, creating a significant financial risk. The PSF board unanimously voted to withdraw the proposal, stating that accepting the conditions would be a betrayal of its mission to support a diverse and international community of Python programmers.

Sources: