NerdNews

OpenAI has developed its first in-house AI chip, called Jalapeño, in collaboration with Broadcom. This move marks a significant step towards achieving technological independence and reducing dependence on Nvidia. The new chip is an application-specific integrated circuit (ASIC) designed for specialized tasks, allowing OpenAI to deliver cheaper products and minimize wait times. The development of Jalapeño took just nine months, making it one of the fastest ASIC development cycles in high-performance advanced semiconductors.

The UK's Competition and Markets Authority (CMA) is investigating Microsoft's business software ecosystem, with many respondents arguing that the company's integration of productivity software, operating systems, and AI creates a form of lock-in that the market alone cannot discipline. Microsoft argues that the investigation misunderstands a vigorously competitive market and that AI is making competition more intense, not less.

OpenAI and Broadcom have unveiled Jalapeño, their first jointly-made AI chip. Designed as an 'Intelligence Processor', it's an accelerator architected for OpenAI's large language models, offering better performance per watt than current state-of-the-art chips. The chip is a result of a 9-month co-development process and is expected to increase efficiency and accessibility of ChatGPT tools.

Top AI researchers Jonas Adler and Alexander Pritzel are leaving Google for Anthropic, following other notable departures to OpenAI and Anthropic as these companies prepare to go public.

Google introduces OpenRL, an open-source project for post-training and fine-tuning Large Language Models (LLMs) on standard Kubernetes clusters, allowing for more efficient GPU utilization and improved user experience.

Google's Andi Gutmans talks about the transition from human-scale to agent-scale data management, using AI agents to organize data and manage metadata. He discusses technical hurdles, quality verification, and the tension between open source and proprietary AI models.

The US government is reportedly urging Meta to submit its AI models for evaluation amid growing safety and security concerns. Meta is the only major AI developer that hasn't voluntarily shared its models for review, while companies like OpenAI, Anthropic, Google, xAI, and Microsoft have already agreed to provide early access to their new models.

Facebook has introduced a new AI companion app for creators, designed to help them grow their audiences on the platform. The app includes features such as an AI-powered comment tool and a daily feed of priorities, and is currently being tested with select creators. The move is part of Meta's efforts to compete with rivals like TikTok and YouTube for creator attention.

Figma's new update adds code layers, support for animations and 3D transforms, and more AI features to improve collaboration and design workflows. The update also includes custom plug-in creation and improved AI assistant capabilities.

Large tech companies like Uber, DoorDash, and Cloudflare are using AI in earlier stages of software development, including product requirement validation and system design inputs, to improve efficiency and reduce errors.

A study by SNS Telecom and IT finds that private 5G networks are increasingly replacing LTE across many verticals, with a market potential far exceeding that of previous technology generations. The report projects that annual investments in private 5G networks for vertical industries will grow at a CAGR of approximately 34% between 2026 and 2029. Physical AI, automation, and workforce connectivity are driving the growth of private 5G networks, with many industrial organizations already seeing practical and tangible benefits from their deployments.

Telefónica Deutschland and Blue Planet have unveiled a proof of concept demonstrating how AI agents can support the design and fulfillment of 5G network slicing services, reducing complex service design tasks from weeks to minutes.

Pegasystems CEO Alan Trefler criticizes big software suppliers for promoting the use of thousands of unpredictable AI agents, citing potential risks and unpredictability. He argues that AI agents are too unpredictable to be used at scale to run mission-critical applications in enterprises. Instead, Trefler suggests using deterministic workflows and AI agents in a controlled manner, as seen in Pegasystems' Blueprint technology.

Companies are trying to reduce their AI spending as they realize the high costs of using AI for basic tasks. Accenture is one such company that has been attempting to stop its employees from depleting its token reserves by using AI to do simple tasks, after previously encouraging employees to use AI and even creating leaderboards to track usage.

OpenAI unveiled its first custom-built inference processor, Jalapeño, designed and manufactured in collaboration with Broadcom. The chip is specifically designed for inference and shows significantly better performance-per-watt than current state-of-the-art alternatives.

Identiverse 2026 discussed the challenges of solving identity for AI agents at scale, including ownership, delegation, least privilege, and auditability. Experts emphasized the need for a named human owner for each agent, scoped and short-lived access, and preserving intent throughout the delegation chain. Inventory and discovery are crucial steps in governance, and teams should prioritize making the secure path the default and building accountability into production.

Researchers found 62 live PyPI tokens leaking on public sources, affecting 125 packages with 25,000 monthly downloads. The tokens were decoded and checked for validity, and the affected packages were notified and the tokens revoked.

CIOs must shift from vague targets to operational metrics and responsible hardware lifecycle management to achieve sustainability in AI infrastructure. This involves setting achievable targets, moving towards location-based reporting, and rethinking the AI refresh cycle to reduce environmental impact.

A new paper argues Microsoft exaggerated its quantum computing claims, saying the company did not conclusively demonstrate a working topological qubit. Microsoft disagrees, standing by its results and roadmap.

Google is changing its Play Store billing policy, allowing developers to use alternative payment systems and reducing fees for some transactions. The new policy replaces the flat 30% billing fee with lower, decoupled fees that vary depending on the developer's earnings and whether they use Google's billing system.