Newsletter Archive

OpenAI renegotiated its Microsoft deal to sell models across any cloud provider, capping Microsoft's revenue share and ending its de facto exclusivity. Amazon is already moving to offer GPT models on AWS Bedrock within weeks, intensifying competition among cloud giants for AI workload hosting.

Axios · 2 min read Industry

NVIDIA released Nemotron 3 Nano Omni, an open multimodal model that consolidates vision, audio, and language into one system for AI agents. The model delivers 9x higher throughput than competing open omnimodal models while maintaining leading accuracy across video, audio, document, and image understanding—eliminating latency and context loss from chaining separate models.

NVIDIA AI · 5 min read Industry

OpenAI and AWS announced Bedrock Managed Agents powered by OpenAI models, while simultaneously OpenAI and Microsoft amended their exclusive cloud deal, allowing OpenAI to serve products on AWS and other providers. Microsoft retains primary partnership status and IP license through 2032 (now non-exclusive) but drops revenue-share arrangement.

Stratechery · 43 min read Industry

US power-plant equipment spending is projected to triple to $65 billion by 2030, with AI datacenters driving 40% of the investment surge, signaling massive infrastructure buildout required to meet compute demand.

Bloomberg Tech · 1 min read Hardware

Box CEO Aaron Levie argues AI workforce disruption is concentrated in tech because engineering work has tight feedback loops and measurable productivity gains—conditions absent in legacy enterprise systems, where AI adoption will be slower and riskier than Silicon Valley headlines suggest.

Fortune · 6 min read Community

OpenAI released Privacy Filter, a 1.5B-parameter open-source NER model for detecting and redacting PII across eight categories (emails, phone numbers, credentials, etc.) that runs directly in browsers or on laptops with only 50M active parameters. Available on Hugging Face under Apache 2.0, it handles high-throughput log scrubbing and dataset cleaning without routing sensitive data to external APIs.

MarkTechPost · 5 min read Tools

The Trump White House is drafting an executive action to overturn its own ban on Anthropic, allowing federal agencies to deploy the company's models including its new Mythos system. The shift reverses a recent supply chain risk designation and follows meetings between White House officials and Anthropic CEO Dario Amodei.

Axios · 3 min read Policy

OpenAI and Anthropic held separate classified briefings with House Homeland Security Committee staff on their new cyber-capable AI models, including GPT-5.4-Cyber and Mythos Preview. Anthropic has withheld public release of Mythos due to its ability to rapidly exploit security flaws, while OpenAI is using tiered release for its cyber model—both working with federal agencies as lawmakers confront the cybersecurity risks these frontier models pose to critical infrastructure.

Axios · 3 min read Policy