The morning briefing.
While you slept: Google inks a massive AI compute deal with SpaceX, OpenAI negotiates a government stake, and Alibaba pushes multimodal agents.
AI Infrastructure. Google's massive deal with SpaceX for Nvidia chips highlights the intense demand for AI computing power. This $920 million per month agreement for 110,000 Nvidia chips underscores the scarcity of AI infrastructure, even for major cloud providers. Such large-scale investments are becoming crucial for companies to meet the growing demands of their AI platforms.
AI Governance. The Trump administration is reportedly negotiating a government stake in OpenAI, proposing a "Public Wealth Fund" to benefit American citizens. This move raises concerns about a "too big to fail" dynamic, similar to past financial crises. Separately, new research explores zero-knowledge verification for frontier AI training, a critical step for enforcing future international agreements on AI regulation.
Agentic AI. Advances in autonomous AI agents continue, with Alibaba's Qwen3.7-Plus demonstrating multimodal capabilities for developing applications. Researchers are also exploring the accountability of AI agents, as seen with OpenClaw's use of external code. Furthermore, a new open-source voice model from The Decoder offers continuous listening and speaking capabilities, pushing the boundaries of real-time AI interaction.
SpaceX signs $920 million per month deal with Google for 110,000 Nvidia AI chips ahead of IPO
SpaceX is leasing AI computing capacity to Google for $920 million per month, providing access to approximately 110,000 Nvidia chips for Google's Gemini Enterprise platform. This deal highlights the scarcity of AI infrastructure and the intertwined nature of big tech companies' businesses.
New open-source voice model listens nonstop and decides every 0.4 seconds whether to speak or stay silent
A new open-source voice model, Audio Interaction, operates continuously, translating, transcribing, and chatting without waiting for recordings to end. Code and model weights are available on GitHub under the Apache 2.0 license.
OpenClaw used Gavriel Cohen’s code and exposed the AI Agent accountability problem
The case of OpenClaw using Gavriel Cohen's code has brought to light significant issues regarding the accountability of AI agents. This incident underscores the challenges in tracking and assigning responsibility for actions taken by autonomous AI systems.
GrapheneOS user reported to authorities for using GrapheneOS
A GrapheneOS user was reportedly reported to authorities simply for using the privacy-focused operating system. This incident raises concerns about digital privacy and potential surveillance based on software choices.
DeepSeek V4 Powers Goedel-Architect: 500x Cost Advantage in Formal Theorem Proving
Princeton University's Language and Intelligence Lab has published a paper on Goedel-Architect, an agent framework powered by DeepSeek V4, which achieves state-of-the-art formal theorem proving results with a 500x cost advantage. This breakthrough significantly reduces the computational expense of complex mathematical proofs.
Tencent Launches WorkBuddy Enterprise Edition: From Super Individuals to Super Teams
Tencent Cloud has launched WorkBuddy Enterprise Edition and Agent Suite, an AI-native solution aimed at enhancing organizational evolution by fostering collaboration among AI-augmented teams. The platform seeks to bridge the gap between individual empowerment and collective intelligence.
PUDU Robotics: From HKUST Roots to Global Commercial Robot Leader
PUDU Robotics, founded in 2016, has become the world's leading commercial service robotics company with a 23% global market share and a valuation exceeding $1.4 billion. The company's success highlights the growing demand for autonomous service robots in various industries.
OpenAI and the Trump administration are negotiating a government stake in the AI startup
OpenAI and the Trump administration are reportedly negotiating a direct government stake in the AI startup, potentially through a "Public Wealth Fund" to benefit American citizens. Critics express concern that such an arrangement could create a "too big to fail" dynamic.
Qwen3.7-Plus is Alibaba's bid to turn multimodal AI into a full-blown autonomous agent
Alibaba's Qwen team released Qwen3.7-Plus, a proprietary multimodal agent model that integrates visual perception, GUI operation, and coding into a single agent loop. A demo showed an agent built on the model autonomously developing a vocabulary learning app with over 10,000 lines of code.
Hasbro is turning Optimus Prime and Mr. Potato Head into AI-powered voices
Hasbro is transforming iconic childhood characters like Optimus Prime and Mr. Potato Head into AI-powered voices. This initiative is a licensing strategy that could also dilute the unique magic of these beloved figures.
How Far Did They Go? The Persuasive Tactics of Covert LLM Agents in a Discontinued Field Experiment
A study analyzed a dataset from a discontinued field experiment on Reddit where undisclosed AI-generated accounts engaged users in live debate. The research examines the persuasive tactics employed by covert LLM agents in real-world social interactions.
SentinelBench: A Benchmark for Long-Running Monitoring Agents
SentinelBench is introduced as a new benchmark for evaluating AI agents on long-running monitoring tasks, addressing the need for sustained attention rather than continuous action in many real-world scenarios. This benchmark helps assess how agents notice external events to make progress.
Zero knowledge verification for frontier AI training is possible
Research suggests that zero-knowledge verification for frontier AI training is technically feasible, offering a primitive for enforcing governance frameworks. This capability could enable coordinated regulation of high-impact AI models without relying solely on self-reporting.
Insurance of Agentic AI
This paper examines the novel risk exposures introduced by agentic AI systems, which extend beyond information generation to autonomous planning and decision execution. These capabilities create challenges for traditional insurance categories.