Reports
AI-generated structured vendor updates
Anthropic Locks in Multi-Gigawatt Next-Gen TPU Capacity with Google and Broadcom
Anthropic has signed a new agreement with Google and Broadcom to secure multiple gigawatts of next-generation TPU capacity, expected online starting 2027. This expansion aims to power frontier Claude models and meet surging global customer demand. The partnership significantly expands Anthropic's $50 billion U.S. compute infrastructure commitment.
Microsoft Releases Copilot Studio Multi-Agent System, Advancing Connected Enterprise AI Architecture
Microsoft announced the general availability of multi-agent systems in Copilot Studio, enabling agent orchestration across tools and data sources via open protocols (A2A) and integrations with Fabric and the Microsoft 365 Agents SDK. This moves beyond isolated AI experiences to scalable, collaborative agent systems, with enhanced prompt building and governance controls.
Google Introduces Flex and Priority Inference Tiers for Gemini API
Google adds Flex and Priority service tiers to its Gemini API. Flex is a cost-optimized tier offering a 50% price reduction for latency-tolerant workloads via a synchronous interface. Priority is a high-reliability tier ensuring critical requests are not preempted during peak loads. This provides developers a unified way to balance cost and reliability based on AI task types, such as background agentic workflows versus interactive applications.
Google Launches Gemma 4 Open Models, Targeting Edge Inference and AI Agent Architecture
Google introduces the Gemma 4 open model family, with four sizes from 2B to 31B parameters, emphasizing breakthrough intelligence-per-parameter and native support for agentic workflows, multimodality, and long context. The small models are engineered for edge devices, aiming to bring frontier reasoning to mobile and IoT scenarios.
Google Introduces Flex and Priority Tiers for Gemini API
Google adds Flex and Priority service tiers to Gemini API, enabling developers to optimize cost and reliability through a single interface. Flex offers 50% cost savings for latency-tolerant workloads, while Priority ensures highest reliability for critical apps. This change simplifies management of synchronous/asynchronous tasks in AI agent architectures.
Google Launches Gemma 4 Open Model Family
Google introduces Gemma 4 open model family with four size variants, optimized for edge and mobile devices. The series supports multimodal processing, long context windows and 140+ languages under Apache 2.0 license.
AMD Announces Breakthrough MLPerf Inference 6.0 Results, Showcasing Multinode Scaling and Multimodal Capabilities
AMD's MLPerf Inference 6.0 submission, powered by Instinct MI355X GPUs, surpassed 1 million tokens per second for the first time on models like Llama 2 70B and GPT-OSS-120B. The results highlight efficient multinode scaling, rapid enablement of new workloads (e.g., text-to-video model Wan-2.2-t2v), and reproducible performance across a broad partner ecosystem.
Cisco Joins NIST GenAI Trust Program
Cisco participates in NIST's GenAI Trust Program, focusing on measurable AI trust evaluation frameworks including adversarial testing (Cat-and-Mouse) and code generation challenges to verify AI output reliability.
Cisco Discloses Memory Poisoning Attack Method in AI Coding Assistants
Cisco's security team discovered and validated a persistent memory poisoning attack method targeting AI coding assistants like Claude Code, demonstrating how tampering with MEMORY.md system files can persistently manipulate AI behavior. This vulnerability prompted Anthropic to remove user memory files' system prompt privileges in v2.1.50.
NVIDIA Demonstrates AI Factories as Flexible Grid Assets for Peak Demand Management
NVIDIA, in collaboration with EPRI, National Grid, and Emerald AI, demonstrated how AI factories powered by Blackwell GPU clusters can dynamically adjust power consumption in response to grid signals. This allows them to act as 'shock absorbers' during peak demand while maintaining performance for high-priority AI workloads.
Cisco Extends Zero Trust Security to AI Agent Ecosystem
At RSA 2026, Cisco introduced security innovations for AI agents, extending Zero Trust Access with agent discovery in Identity Intelligence, agentic IAM in Duo, and MCP enforcement in Secure Access SSE. It launched AI Defense: Explorer Edition for self-serve testing and DefenseClaw open source framework to automate security deployment.
SK Hynix Jumps to TSMC 3nm for HBM4E Logic Die to Counter Samsung's 4nm Lead
SK Hynix plans to use TSMC's 3nm process for the logic die in its 7th-gen HBM4E, a leap from the 12nm used in HBM4. This aims to reverse the performance gap with Samsung (which used 4nm logic in HBM4) and deliver higher bandwidth and power efficiency for next-gen AI chips like NVIDIA's Vera Rubin Ultra.
Project Rheo: NVIDIA Shifts Robot Training Control from Real Hospitals to Simulation
NVIDIA unveils Project Rheo, a blueprint combining Isaac Sim, GR00T VLA models, and synthetic data generation for hospital robotics. Developers train Physical AI policies in digital twins—loco-manipulation (surgical tray pick-and-place) and precision bimanual tasks (trocar assembly)—with Cosmos Transfer 2.5 for cross-scene generalization.
NVIDIA Releases Cosmos World Model Suite, Enhancing Synthetic Data and Reasoning for Physical AI
NVIDIA has released significant updates to its Cosmos World Foundation Models (WFM) suite, including Transfer 2.5, Predict 2.5, and Reason 2. These models are designed to accelerate the generation of high-fidelity, physics-aware synthetic data and support downstream fine-tuning and reasoning for physical AI systems like robotics and autonomous vehicles, addressing the bottleneck of real-world data scarcity.
NVIDIA Warp: Differentiable Physics Simulation for AI Training on GPU
NVIDIA Warp is a framework for GPU-accelerated, differentiable physics simulation. It enables writing high-performance kernels in Python, with automatic differentiation, and integrates with PyTorch/JAX. The 2D Navier-Stokes example demonstrates end-to-end optimization, reducing the cost of generating training data for physics AI.
Trend Micro Report Highlights AI Supply Chain Risks and Model Attack Surfaces
Trend Micro's 'Fault Lines in the AI Ecosystem' report systematically analyzes security risks in the AI supply chain, including training data poisoning, third-party plugin vulnerabilities, and model theft attacks. It indicates that enterprise AI security boundaries have expanded from traditional IT infrastructure to the model layer and data pipelines.
NVFP4 + TeaCache Drive 10x FLUX.2 Inference Speedup, Locking Blackwell Ecosystem
NVIDIA and BFL optimize FLUX.2 on DGX B200/B300 using NVFP4 4-bit quantization, TeaCache step skipping, CUDA Graphs, and torch.compile, achieving 6.3x (single GPU) to 10.2x (dual GPU) latency reduction vs H200, with 40% memory savings. The stack is tightly coupled to TensorRT-LLM visualgen and Blackwell hardware.
OpenAI provides video generation infrastructure to Higgsfield via GPT-4.1, GPT-5, and Sora 2 model stack
OpenAI showcased in its developer blog how the third-party app Higgsfield leverages its combined GPT-4.1, GPT-5, and Sora 2 models to transform simple inputs into high-quality social videos. This demonstrates OpenAI's strategy of positioning its multimodal models as core components of external AI inference infrastructure.
OpenAI Releases Chain-of-Thought Monitorability Framework
OpenAI introduces a new chain-of-thought monitoring evaluation suite with 13 metrics across 24 test environments. Research shows monitoring model's internal reasoning is more effective than output-only monitoring, offering new approach for scalable AI control.
NVIDIA Launches Interactive AI Agent for GPU-Accelerated Data Science with Nemotron Nano-9B
NVIDIA unveils an interactive AI agent powered by Nemotron Nano-9B-v2 and CUDA-X libraries, enabling natural language orchestration of ML workflows. It achieves 3x-43x GPU acceleration over CPU for data processing, model training, and hyperparameter optimization.