Latest AI News

Apple plans to unveil a software development kit at WWDC 2025 that will allow third-party developers to build applications using its AI models. This significant shift in Apple's approach aims to energize its Apple Intelligence ecosystem and compete with rivals like Google and Samsung. The initiative will initially focus on smaller on-device models before potentially expanding to more powerful cloud-based versions.

At Google I/O 2025, the company introduced Veo 3, a groundbreaking AI model that generates videos with synchronized audio, including ambient sounds and character dialogue. Google also launched Imagen 4, which renders intricate details with remarkable clarity and supports up to 2K resolution. These advanced AI models power Flow, Google's new filmmaking tool designed to help creators bring cinematic visions to life.

Google has expanded Project Astra, its universal AI assistant initiative, with enhanced agentic capabilities that can perform tasks autonomously on users' behalf. At Google I/O 2025, the company demonstrated how the system can recognize real-world objects through a smartphone camera and take action, such as automatically adding events to calendars from written invitations. Google also showcased new XR glasses with augmented reality features including live language translation, while expanding AI mode in Search to access email data and enable virtual clothing try-ons.

German chipmaker Infineon Technologies has partnered with Nvidia to develop an industry-first 800V high-voltage direct current (HVDC) power delivery architecture for AI data centers. Announced on May 20, 2025, this collaboration aims to address the surging power demands of advanced AI infrastructure, which is expected to require over one megawatt per server rack by decade's end. The new system significantly increases energy efficiency by allowing power conversion directly at the GPU.

Microsoft announced the integration of AI models from xAI, Meta, Mistral, and Black Forest Labs into its Azure AI Foundry platform, expanding its ecosystem beyond its OpenAI partnership. This strategic move, revealed at Microsoft's Build conference on May 19-20, 2025, gives developers access to over 1,900 AI models hosted directly in Microsoft's data centers. The multi-model approach positions Azure as a comprehensive AI platform for businesses building custom AI agents and applications.

At Computex 2025 in Taiwan, Nvidia CEO Jensen Huang unveiled NVLink Fusion, a groundbreaking program enabling customers to integrate non-Nvidia CPUs and GPUs with Nvidia's hardware. This strategic move allows companies like MediaTek, Marvell, and Qualcomm to connect their custom processors with Nvidia GPUs in AI data centers while benefiting from Nvidia's ecosystem. The technology represents Nvidia's effort to maintain its central position in AI development by embracing interoperability as competitors develop custom silicon solutions.

At Google I/O 2025, Google announced Agent Mode for its Gemini AI, enabling autonomous task completion through web browsing capabilities powered by Project Mariner. This new feature will allow Gemini to perform complex tasks like apartment hunting on sites like Zillow, from finding listings to scheduling tours. Google also revealed compatibility between its Gemini API/SDK and Anthropic's Model Context Protocol, signaling growing industry collaboration on agent standards.

Microsoft announced the integration of Anthropic's AI coding agent into GitHub, expanding its autonomous programming capabilities. This addition comes alongside Microsoft's own coding agent and OpenAI's offering, highlighting the company's shift toward a more neutral stance in the AI race. The GitHub Copilot coding agent functions as an autonomous peer programmer that can handle tasks like fixing bugs, adding features, and improving documentation in well-tested codebases.

Google has launched its most premium AI subscription tier yet, priced at $249.99 monthly, targeting power users seeking cutting-edge AI capabilities. The new AI Ultra plan, announced at Google I/O 2025, offers access to Google's most advanced AI models and experimental features, alongside substantial cloud storage and ad-free YouTube. This move positions Google competitively against similar high-end offerings from OpenAI and Anthropic as tech giants explore ways to monetize costly AI development.

Microsoft and GitHub have officially joined the steering committee for Anthropic's Model Context Protocol (MCP), a move announced at Microsoft Build 2025. This strategic alignment comes as MCP gains significant momentum in the AI industry, with both OpenAI and Google having pledged support earlier this year. The protocol enables secure, standardized connections between AI models and various data sources, positioning it as a foundational technology for the emerging agentic AI ecosystem.

Researchers from Duke University have developed WildFusion, an innovative AI framework that combines vision, vibration, and touch to help robots navigate complex outdoor environments. Announced on May 19, 2025, at the IEEE International Conference on Robotics and Automation in Atlanta, this breakthrough technology enables robots to sense and adapt to unpredictable terrains much like humans do. The same day, POSTECH researchers unveiled a complementary haptic device designed to enhance safety in industrial settings through improved human-robot interaction.

Microsoft has committed to powering all its AI data centers with 100% renewable energy by 2026, as artificial intelligence energy consumption is projected to increase tenfold from 2023 levels. This ambitious initiative comes amid growing environmental concerns about the tech industry's carbon footprint. The company is making significant investments in wind, solar, and other clean energy sources to support its goal of achieving carbon negativity by 2030.

MIT researchers have discovered that vision-language models, widely used in medical image analysis, cannot understand negation words like 'no' and 'not'. This critical limitation could lead to serious diagnostic errors when these AI systems are asked to retrieve medical images with specific criteria. The study, published on May 14, 2025, introduces NegBench, a new benchmark to evaluate and improve negation understanding in AI vision systems.

Scientists at the University of Bristol have developed a revolutionary soft robot that mimics an octopus's distributed nervous system, enabling autonomous decision-making through fluid dynamics. The robot can sense its environment, adapt its grip to different objects, and predict pulling forces without requiring electronic sensors or a central processor. This breakthrough represents a significant advancement in creating more intuitive and adaptable soft robotics for real-world applications.

University at Buffalo researchers have developed an artificial intelligence system that analyzes children's handwriting to detect early signs of dyslexia and dysgraphia. The technology, published in SN Computer Science, examines subtle patterns in handwriting samples to identify spelling issues, poor letter formation, and other indicators of these learning disabilities. This AI-powered approach could revolutionize early screening by making it more accessible, especially in underserved areas facing shortages of speech-language pathologists.

AMD has unveiled significant advancements in its AI strategy at Computex 2025, announcing ROCm platform support for its Radeon RX 9000 series and new Radeon AI PRO graphics products. The expanded software ecosystem enables developers to accelerate applications with popular AI frameworks on both Linux and Windows. The company's new Radeon AI PRO R9700 graphics card, featuring 32GB of memory and PCIe 5.0 support, is specifically designed for local AI inference, development, and memory-intensive workloads.

NVIDIA has unveiled Isaac GR00T-Dreams, a groundbreaking technology that generates vast amounts of synthetic training data for humanoid robots. This innovation dramatically accelerates robot development by creating months' worth of training data in just hours, addressing a critical bottleneck in physical AI advancement. CEO Jensen Huang has positioned physical AI as the next trillion-dollar industry, with applications initially targeting factories before eventual consumer adoption.

MSI unveiled the Claw A8 BZ2EM at Computex 2025, marking its first gaming handheld powered by AMD's Ryzen Z2 Extreme chip. The device features an 8-inch Full HD 120Hz display, up to 24GB of LPDDR5X RAM, and a substantial 80Wh battery. Available in white and lime green finishes, this Windows 11 handheld aims to compete directly with established players like Steam Deck and ROG Ally.

Amazon has introduced a new AI-powered audio feature called 'Hear the highlights' that transforms product research into short-form listening experiences on select product pages. The feature uses large language models to analyze product details, customer reviews, and web information, presenting it in a conversational format to help shoppers make informed decisions quickly. Initially available to a subset of U.S. customers, Amazon plans to expand the feature to more products and users in the coming months.

Snowflake Inc. raised its fiscal 2026 product revenue forecast to $4.325 billion on Wednesday, exceeding previous estimates as enterprises prioritize AI spending. The company reported first-quarter revenue of $1.04 billion, with product revenue growing 26% year-over-year to $996.8 million. Snowflake's strategic AI partnerships with OpenAI and Anthropic are enabling customers to build and deploy advanced AI models within its secure data cloud environment.