Anthropic: Claude Now Writes More Than 80% of Its Own Code

According to Anthropic, the company's AI model Claude now writes more than 80% of its own merged code. The fact that AI has become deeply involved in its own development points to an accelerating self-improvement loop in software development.

Anthropic says this pace is "faster than expected," suggesting that AI-generated code is shifting from a mere assistive tool to a mainstay of the development process. That said, a human review and approval process for generated code remains in place.

Google Signs $920 Million-Per-Month Computing Deal with SpaceX

Google, under Alphabet, has signed a contract to receive computing power from Elon Musk's SpaceX for $920 million per month (approximately ¥140 billion). The deal runs from October 2025 through mid-2029 and is expected to reach a total of around $40 billion.

This marks Google's second partnership with an AI rival in a matter of weeks, underscoring once again that securing massive compute resources has become the single most critical challenge for AI companies.

Apollo Raises $35 Billion to Fund AI Chip Purchases for Anthropic

Investment firm Apollo Global Management has completed $35 billion (approximately ¥5.3 trillion) in debt financing to support Anthropic's AI chip purchases. One of the largest single investments in AI infrastructure to date, the deal signals that Anthropic plans a substantial expansion of its computing power.

Anthropic already receives funding and compute resources from Google and Amazon, but this arrangement with Apollo shows that the forms of investment flowing into AI companies are diversifying.

Reid Hoffman to Leave Microsoft's Board to Focus on New AI Company "Manas"

Reid Hoffman, LinkedIn co-founder and venture capitalist, announced that he will step down from Microsoft's board of directors later this year. The move is to focus on his new AI company, Manas, with Hoffman saying he "needs to get back into founder mode."

Hoffman, who has also served on OpenAI's board, is known as a key figure in the AI industry. Details about Manas remain unclear at this point, but his full commitment suggests the company's ambition.

University of Cambridge Succeeds in Clinical Trial of AI-Designed "Universal Vaccine"

The University of Cambridge in the UK announced that the first clinical trial of a "universal vaccine" using AI-designed antigens has succeeded. The research team applied machine learning to genome sequences of the sarbecovirus group to design a "super-antenna" shared across the group. It was administered to 39 healthy volunteers, confirming both safety and immune response.

The result is regarded as an important step toward next-generation vaccines capable of responding to unknown variants. It is also drawing attention as a case of AI being involved from the earliest stages of the drug development process.

Gemma 4 QAT Posts Strong Benchmark Results on AMD's 7900 XTX

The QAT (Quantization-Aware Training) version of Google's open language model Gemma 4 is reporting good benchmark results on AMD's Radeon 7900 XTX. According to tests, it delivers higher speed and reduced VRAM usage compared to the previous model, with no degradation in quality observed.

The maturing of local LLM runtimes on AMD GPUs is also meaningful from the perspective of reducing dependence on NVIDIA.

---