Google Signs $920 Million Monthly Contract with SpaceX, Florida Sues OpenAI — AI News Roundup for June 5

Google Signs $920 Million Monthly Computing Deal with SpaceX

Google announced a contract to provide SpaceX with computing resources for $920 million per month (roughly $11 billion on an annualized basis). The announcement came just one week before SpaceX's historic IPO. As demand for AI infrastructure expands rapidly, the tie-up between the space venture and big tech is expected to grow even deeper.

Florida Sues OpenAI, Casting ChatGPT as a "Defective Product"

Florida has filed a landmark lawsuit against OpenAI and its CEO, Sam Altman. The 83-page complaint treats ChatGPT as a "defective product" subject to product liability, citing risks to minors, the lack of age verification, and insufficient investment in safety. The suit also asserts a public nuisance claim, and damages could run into the billions of dollars. It is the first state-level lawsuit against an AI company in the United States, and it is drawing attention as a potential legal precedent going forward.

Anthropic Proposes a Global Slowdown of AI Development

Anthropic has proposed a global slowdown of AI development. It is an unusual move for a major AI company to call for restraining its own development pace. The proposal appears to be driven by safety concerns, though how it balances against competitors and the open-source community will be the key focal point.

NVIDIA Releases New 550B-Parameter Model "Nemotron-3-Ultra"

NVIDIA has released "NVIDIA-Nemotron-3-Ultra-550B-A55B-NVFP4" on Hugging Face. The model uses a LatentMoE architecture with 550B parameters (55B active), adopting a hybrid configuration of Mamba-2, MoE, and Attention. It supports a context length of up to 1 million tokens and 10 languages including Japanese, and is optimized for complex reasoning, agentic workflows, and tool use. Thanks to NVFP4 quantization, it runs on 4x GB200/B200 GPUs or 8x H100 GPUs.

Sakana AI Establishes a "Recursive Self-Improvement (RSI) Lab"

Tokyo-based Sakana AI has established a research lab dedicated to Recursive Self-Improvement (RSI), where AI improves itself. AI improving AI could lead to a sudden leap in capabilities — an intelligence explosion — making it an important research area from a safety perspective as well. The news sparked discussion on Hacker News, garnering 12 points and 7 comments.

Meta Is Building Tents for AI Servers Across the United States

Meta is building tent-structured facilities across the United States to house AI servers. These facilities, with a Mad Max-esque appearance, take about three months to construct and reportedly use jet engines as their power source. With data center demand vastly outstripping supply, it is a bold approach to scaling up infrastructure without waiting for conventional facilities to be built.