AI news, 23 July: Alphabet hikes AI spending to as much as $205 billion

Alphabet raises 2026 capital spending forecast on AI demand. Google’s parent company reported strong second-quarter results driven by an 82% jump in cloud revenue to $24.8 billion, yet its shares fell after it lifted full-year capital expenditure guidance to $195–205 billion, up from the prior $180–190 billion range. Most of the quarter’s $44.9 billion in capex went to AI infrastructure, with finance chief Anat Ashkenazi citing persistent supply constraints and robust demand. CEO Sundar Pichai noted the Gemini app has reached 950 million monthly active users and that work is advancing on Gemini 4. Investors remain nervous about the scale of Big Tech’s AI buildout even as cloud growth accelerates, according to CNBC and Reuters coverage of the earnings.

White House adviser accuses China’s Moonshot AI of stealing from Anthropic. U.S. science and technology adviser Michael Kratsios alleged that Moonshot AI conducted large-scale “distillation” — using outputs from a stronger model to train a weaker one — of Anthropic’s Claude Fable 5 to help develop its newly released Kimi K3 model. He also pointed to acquisition of restricted Nvidia chips accessed via Thailand. The claims, reported by BBC and others, intensify U.S. scrutiny of Chinese open-weight models that have challenged American leads on some benchmarks. Distillation for research is common; the accusation centres on alleged industrial-scale, covert copying of proprietary capabilities.

AMD strikes multibillion-dollar chip and investment deal with Anthropic. Advanced Micro Devices will supply Anthropic with up to 2 gigawatts of next-generation Instinct MI450 AI chips (deployments from first half of 2027) in a servers deal worth tens of billions of dollars, while investing up to $5 billion in the Claude maker, according to Reuters and company statements. The partnership helps Anthropic secure scarce computing capacity and gives AMD a major win against Nvidia’s dominance in AI accelerators.

Rubio directs diplomats to counter “kill switch” narrative on U.S. tech. Secretary of State Marco Rubio instructed U.S. diplomats in a July cable, reviewed by Reuters, to push back against claims that American AI and technology contain a government “kill switch.” The guidance follows a short-lived White House restriction that temporarily blocked foreign access to Anthropic’s most advanced models, which alarmed European partners and fuelled calls for digital sovereignty. Talking points frame testing pauses or narrow controls as ordinary policy, not remote disablement, and urge promotion of U.S. AI systems over rival “sovereign” builds.

OpenAI-linked Australian megacenter drops recycled-water cooling plan. Developers of a planned 612-megawatt data centre in Sydney that OpenAI intends to help build and use abandoned recycled-water cooling after pipeline infrastructure proved unavailable, switching to a liquid-coolant system that experts say is more energy-intensive, Reuters reported exclusively. The shift underscores tensions between AI infrastructure races and local water/energy constraints; Australian authorities are moving toward usage limits while communities push back on resource demands.