Anthropic reported nearly $17 billion in quarterly revenue, about twelve times its level a year earlier, but lost $23.3 billion. Spending on computing capacity and employee donation matching contributed to the loss.
Lumentum CEO Michael Hurlston said Marvell’s planned purchase of the company could help meet rising demand for connections between AI chips. The stock-based deal gives Lumentum an implied value of $115 billion and is expected to close in early 2028.
Oracle filed plans to raise $20 billion by selling $10 billion in shares and $10 billion in bonds to help fund its AI infrastructure. Its stock fell 8.8% Monday as investors weighed the plan against concerns about the company’s existing debt.
OpenAI signed agreements in Abu Dhabi to use its AI models in energy, investment analysis and government services. The deals come as the company seeks backing from investors in the United Arab Emirates for a new $20 billion funding round.
SpaceX appears to have reduced Grok Bot usage limits weeks after Elon Musk called it his fastest-growing product. Some users reported sharply lower quotas, and Musk said the service costs too much to run compared with what it earns, though he did not confirm a cut.
MIT researchers used GPT-6 Astra to change a small Qwen model’s internal settings directly, raising its score on a math test from 7.9% to 48.1%. The edit mainly stopped the model from repeating itself, rather than adding new math skills.
OpenAI published an addendum describing safety tests and access rules for GPT-Rosalind-6, a biology research model offered to approved partners. The model answers more sensitive biology questions than earlier versions but is designed to refuse requests about biological weapons.
Anthropic says future AI models may soon meet its internal threshold for rapid progress driven by AI helping develop AI. It says no current model has reached that threshold and warns that faster development could outpace its safety measures.
The New York Times and three other newspapers sued Brave, alleging the browser company copied their articles and supplied the content to AI companies without permission. The lawsuits also challenge a Brave feature that bypasses paywalls and the company’s ad practices.
Trajectory raised $45 million to expand a platform that helps people without technical training improve AI models for specialized work. It lets experts give feedback on an open-source model’s performance, then uses that feedback to retrain it.