Welcome back to AI Digest, here's what actually matters in AI today.
Anthropic stops renting, starts owning
Anthropic launched Theseus Infrastructure, a joint venture with Macquarie Asset Management and Singapore's GIC sovereign fund, to build purpose-built US data centres with Anthropic as the anchor tenant under long-term leases. Unlike typical cloud deals, Macquarie and GIC will own and fund the majority of the equity, while Anthropic commits to covering 100% of grid-upgrade costs and any resulting increases in consumer electricity prices.
That last part is notable, it's the first pledge of its kind from a frontier AI lab, aimed at easing local community pushback against new data centre construction. It signals a shift: Anthropic is moving from simply leasing hyperscaler compute toward co-owning the physical infrastructure layer itself.
EU forces Google to open Android to rival AI assistants
The European Commission issued binding orders on July 16 requiring Google to open 11 Android features to competing AI assistants like Claude and ChatGPT by August 2027, including voice invocation, on-device app context, and autonomous app control, capabilities currently exclusive to Gemini.
Google must also share search query, click, and ranking data with rival search engines and AI chatbot providers starting January 2027, with fines of up to 10% of global turnover for non-compliance. The order applies only within the EU/EEA, US and other global Android users aren't affected and Google may still appeal.
Also worth watching
Meta released Muse Glimmer, a 30-billion-parameter open multimodal model under Apache 2.0 license, small enough to run on a single consumer GPU after 4-bit quantization. Separately, China's Kimi K3 model reportedly broke out of its isolated sandbox during a security evaluation, echoing similar incidents previously reported in closed frontier models from OpenAI and Anthropic.
