Nvidia researchers have developed a technique that uses linear math to transfer key-value caches between different AI models without recomputing conversation history. The method enables enterprises to switch between small and large models mid-session while reducing compute costs and latency by 2.7 to 25 times compared to traditional recomputation, retaining up to 98% accuracy on compatible model pairs.

Enterprise AI deployments are failing at scale not because models lack capability, but because unconstrained autonomy creates accountability gaps that legal and compliance teams cannot accept. Gartner forecasts 40% of agentic AI projects will not survive to 2028, while McKinsey data shows responsible-AI maturity lags far behind deployment velocity. The competitive advantage is shifting from raw agent autonomy to governance, control, and the ability to get agents approved and kept approved in production.

Nvidia is raising prices for its Grace Blackwell and Vera Rubin flagship AI chip systems by approximately 17%, according to server makers. The increase adds to mounting cost pressures facing cloud providers and data center developers, who are already contending with power infrastructure delays and other unexpected expenses. The price hike reflects Nvidia's market position as demand for advanced AI chips remains strong despite the growing cost burden on customers.

Nvidia is in discussions to invest in AI startup Perplexity as part of an equity round valuing the company at over $30 billion, representing a more than 50% increase from its previous funding round a year ago. The investment reflects strengthening business ties between the two companies and comes as Perplexity's annualized revenue has grown to over $750 million from less than $250 million at the start of the year, driven partly by Perplexity Computer, an AI agent for automating professional tasks.

OpenAI has integrated ChatGPT into Apple Messages on Mac, enabling the AI to read, search, and send messages. Mac users can now ask ChatGPT to analyze their messages, including identifying their most frequent contacts.

Inherent, a British AI lab founded by DeepMind alumni, has released Faraday, an AI agent designed to replicate scientific papers. The company claims Faraday outperformed systems from Anthropic and OpenAI at this task. The capability could have implications for accelerating scientific research and innovation.