PyTorch Foundation Welcomes Helion as New Project for AI Kernel Development
The PyTorch Foundation announced that Helion, a Python‑embedded DSL for AI kernel development, is joining its portfolio alongside PyTorch, DeepSpeed, Ray and vLLM. Helion automates ahead‑of‑time autotuning and compiles to multiple backends such as Triton and TileIR, simplifying custom kernel authoring and boosting hardware performance portability. The move reflects the industry’s shift from training‑centric workloads to an inference boom where cross‑platform compatibility is critical. Alongside Helion, ExecuTorch is being integrated into PyTorch Core to extend edge‑and‑on‑device capabilities.
JEDEC Announces May Forums on Next-Gen Memory for AI, Server, Cloud, and Mobile Computing
JEDEC will host two back‑to‑back forums on May 12‑13, 2026 in San Jose, focusing on next‑generation memory for mobile, edge, server, cloud and AI workloads. The events feature keynotes from industry giants such as AMD, Intel, Google, Microsoft, Samsung and...
AI Alliance Announces ‘Project Tapestry’ and Appoints Yann LeCun as Chief Science Advisor
The AI Alliance, a nonprofit coalition of more than 200 organizations, unveiled Project Tapestry, an open‑source platform designed for distributed, federated training of frontier AI models. The initiative aims to let institutions, industries, and nations collaborate on a shared base...
Kimia Raises $7M To Launch Its AI Knowledge Platform for the Chemical Industry
Kimia, a Sydney‑based startup, closed a $7 million seed round to launch an AI‑driven knowledge platform tailored for the chemical industry. The system pulls data from scientific literature, internal documents and supplier catalogs, structuring it for real‑time queries by commercial and...
MLCommons Releases New MLPerf Inference v6.0 Benchmark Results
MLCommons unveiled the MLPerf Inference v6.0 benchmark suite, adding six new or updated datacenter tests and a fresh edge object‑detection benchmark. The release introduces an open‑weight GPT‑OSS 120B large‑language‑model test, an advanced‑reasoning DeepSeek‑R1 scenario, DLRMv3 sequential recommender, text‑to‑video generation, and...
ORNL Introduces ‘Photon’ Framework for Accelerating AI Vulnerability Discovery on Frontier
Oak Ridge National Laboratory’s CAISER team unveiled Photon, a new framework that uses the Frontier exascale supercomputer to accelerate AI vulnerability discovery. By repurposing the DeepHyper training system, Photon runs thousands of jailbreak prompts in parallel, achieving over 95% GPU...
Optibrium Introduces Graphical Interface for QuanSA to Enhance Ligand-Based Affinity Predictions
Optibrium has released a new PyMOL plugin that adds a graphical user interface to its QuanSA ligand‑based affinity prediction tool. QuanSA uses physically‑motivated machine learning to deliver free‑energy perturbation‑level accuracy without needing a protein structure, dramatically lowering computational cost. The...
SandboxAQ Launches New AQtive Guard Capabilities
SandboxAQ unveiled major upgrades to its AQtive Guard platform, expanding AI Security Posture Management (AI‑SPM) ahead of RSA Conference 2026. The new suite adds real‑time guardrails for inbound prompts and outbound responses, autonomous risk analysis for Model Context Protocol (MCP)...
NIST Awards Over $3M to Small Businesses Advancing AI, Biotech, Semiconductors, Quantum and More
The National Institute of Standards and Technology (NIST) has awarded $3.19 million in Phase II SBIR grants to eight small businesses across seven states. The funding supports research in artificial intelligence, biotechnology, semiconductor, quantum and other advanced technologies, with each award ranging...
New MIT Framework Uses Search to Handle LLM Errors in AI Agents
MIT CSAIL and Asari AI introduced EnCompass, a Python framework that adds systematic search and backtracking to AI agents using large language models. Developers annotate "branchpoints" where LLM outputs may vary, then the runtime explores execution paths with strategies such...
Cerebras Systems Raises $1 Billion Series H
Cerebras Systems closed a $1 billion Series H financing round, valuing the company at roughly $23 billion post‑money. The round was led by Tiger Global and included investors such as Benchmark, Fidelity, AMD and Coatue. Proceeds will accelerate production of the Wafer Scale...
TUM Unveils EU’s 1st 7nm AI Chip with Local Processing and RISC-V Architecture
Technical University of Munich unveiled the EU’s first 7‑nanometer AI chip, a neuromorphic processor built on an open‑source RISC‑V architecture. Designed by Prof. Hussam Amrouch, the chip processes data locally, promising higher privacy and security than cloud‑centric solutions. Production will shift...
