Enterprise AI deals surge; open models multiply

Anthropic / Claude ecosystem

Anthropic launches in-country Claude inference in India via AWS

Anthropic expands into India with local AI processing for Claude, enabling in-country inference through Amazon Bedrock. This helps Indian enterprises meet data residency and compliance requirements while potentially reducing latency.

Frontier model providers

Meta doubles efficiency of LLM-scale ads foundation model GEM

Meta detailed how it doubled training efficiency and improved MFU for its Generative Ads Model (GEM), an LLM-scale foundation model. The technical blog highlights architectural and infrastructure optimizations.

DeepSeek V4-Flash official version open-sourced

DeepSeek released the official open-source version of its V4-Flash model with post-training optimizations and GGUF quantization for affordable local deployment. The move intensifies open-source AI competition.

AI developer tooling & infrastructure

No significant new developments.

Cloud & platform providers

Cloudflare details optimizations for running Kimi and GLM at scale

Cloudflare published a blog post on techniques to run Kimi and GLM models at scale, including FP8/KV cache optimizations and integrity checks, improving throughput and latency with minimal overhead.

Cloudflare adds native gRPC support to Workers for edge RPC services

Cloudflare introduced native gRPC support in Workers, enabling lower-latency, higher-throughput RPC services at the edge, beneficial for AI/ML inference and data-intensive services.

Cloudflare introduces Computer, a new hardware-aware compute product

Cloudflare launched Cloudflare Computer, a hardware-aware compute product designed to run AI agents more efficiently than traditional containers, optimizing for latency and resource usage.

AI policy, regulation & governance

White House invites Meta and Anthropic for AI safety testing talks

The White House invited Meta and Anthropic to discuss voluntary cybersecurity testing of the most advanced American AI models, focusing on hacking capabilities. The meeting aims to coordinate safety measures among leading AI developers.

OAIC updates guidance on facial recognition technology in retail spaces

The Australian Information Commissioner updated guidance on the use of facial recognition in retail, emphasizing privacy-by-design, consent, and compliance with the Australian Privacy Principles.

Australian judge warns against AI-generated evidence in court

A Federal Circuit and Family Court of Australia decision dismissed a case due to reliance on AI-generated evidence, with the judge cautioning against the dangers of AI chatbots posing as legal advisors.

Industry & market moves

Augment Code raises $227M Series C at $2B+ valuation

AI coding assistant startup Augment Code secured $227 million in Series C funding at a valuation exceeding $2 billion, signaling strong investor confidence in AI-driven developer tools.

Superblocks and AWS partner for secure enterprise AI app development on Bedrock

Superblocks and AWS announced a strategic collaboration to enable secure enterprise AI application development on Amazon Bedrock, focusing on streamlined, governed AI development.

Amazon completes $50 billion investment in OpenAI

Amazon finalized its $50 billion investment in OpenAI, securing exclusive third-party cloud provider status for OpenAI's Frontier cloud. The deal deepens AWS-OpenAI ties and impacts the AI cloud landscape.

H2O.ai and CAN.B Group partner for Australian sovereign AI

H2O.ai and CAN.B Group formed a partnership to deliver sovereign AI solutions in Australia, integrating H2O.ai's AUSOVRN platform for regulated deployment in government and critical infrastructure.

HUMAIN invests in Saudi AI firm MOZN for enterprise AI

HUMAIN made its first investment in a Saudi company, MOZN, and partnered to co-build enterprise AI solutions locally and globally, focusing on BFSI and public sectors.

Yellow.ai to go public via $550M SPAC merger

Enterprise AI agent leader Yellow.ai announced a merger with Bluerock Acquisition Corp., a special-purpose acquisition company, valuing the combined entity at $550 million and targeting a Nasdaq listing as YAI.

EXL completes acquisition of iMerit to accelerate enterprise AI

EXL finalized its acquisition of iMerit, combining capabilities to create an end-to-end enterprise AI platform with expanded data annotation and AI services.

d-Matrix acquires Wallaroo for inference orchestration across chips

AI chip startup d-Matrix acquired Wallaroo.ai to enhance its inference orchestration capabilities across various chips, improving deployment efficiency for AI models.

Deel acquires AI security startup Clarity

Global hiring platform Deel acquired Clarity, a continuous AI security startup, to embed security throughout the hiring and workforce lifecycle, enhancing its AI-driven compliance features.

Coursera invests $100M in AI-native learning company LearnVector

Coursera made a strategic $100 million equity investment in LearnVector, an AI-native learning company founded by Andrew Ng, to accelerate agentic AI-based personalized education.

AI product & feature launches

Varonis introduces agent intent-based access control for AI agents

Varonis launched an Agent Intent-Based Access Control (IBAC) feature in its Atlas platform to securely connect AI agents to enterprise data, preventing dangerous or out-of-policy actions.

Stripe builds Kai knowledge AI platform on Deep Agents in one week

Stripe developed Kai, a company-wide AI productivity agent, using LangChain's Deep Agents framework in just one week. The agent connects to internal data and tools for employee queries.

Teradata launches enterprise-grade Data Analyst Agent on AWS Marketplace

Teradata launched an AI-assisted, conversational analytics agent available on AWS Marketplace, allowing enterprises to deploy the agent without data movement for quick, natural-language data queries.

Research with immediate practical relevance

Research finds security concerns in nearly half of MCP servers

A study by Island found that nearly 50% of MCP server builds have security issues in the tools exposed to AI agents, raising concerns about a second software supply chain risk.

Particle models achieve self-supervised 3D scene decomposition

Researchers introduced 3D-DLP, the first self-supervised approach to decompose colored 3D scenes into object-centric particles without annotations, advancing how AI perceives physical worlds.

MiniMax H3 now open source

MiniMax officially open-sourced its H3 model, a multimodal generation model, following an earlier open-source pledge. The release provides developers and enterprises with free access to advanced AI capabilities.

KAIST's AI builds workable schedules without optimization software

KAIST researchers developed RL-SPH, an end-to-end learned scheduling approach that achieves 100% feasibility on benchmarks and outperforms prior methods in speed and scalability, solving complex ILP problems.