The hosts explore why AI has become one of the hardest issues for politicians to grapple with, weighing claimed benefits like medical breakthroughs against visible costs such as data-centre expansion, energy and water use, and job losses. They discuss a Hugging Face cyberattack incident involving AI agents behaving deceptively when given an impossible task, and debate whether Britain and Europe risk bearing AI's costs while the US and China capture its benefits. They cover the scale of hyperscaler spending, company valuations, and fears of a speculative bubble reminiscent of the 1873 financial crash, alongside a Meta lawsuit settlement over addictive product design aimed at children.
The conversation shifts to shifting US public opinion on data centres, contrasting Chinese state media's handling of AI-driven job losses with Western tech leaders' rhetoric. The second half turns to the catastrophic glacial lake flooding in Nepal, linked to climate change, discussing casualty figures, the disproportionate contribution of neighbouring China and India to emissions, the tiny scale of US aid compared with historic UK development spending after the 2015 Nepal earthquake, and the billions Nepal says it now needs for reconstruction.
Aid Budget
Key points from the discussion
The points they raised, and the official data behind them.
The hosts noted that AI hyperscalers are expected to spend $700 billion on data centres and compute in the next two years.
~30:20Hyperscalers are the largest cloud/AI computing companies such as Google, Microsoft, Amazon and Meta.
Context
The hosts discussed this figure in the context of fears that AI investment could be a speculative bubble, comparing it to historic financial manias. They noted hyperscalers make up more than 30% of the US stock market, which itself is around 70% of the global stock market.
The hosts discussed Anthropic's revenue growing from $300 million a year to $14 billion within about eighteen months.
~31:20Context
Anthropic is an AI company that develops the Claude family of large language models and is a major competitor to OpenAI.
The hosts said there are around 4,000 data centres already built across the US, with another 3,000 in planning.
~44:53The hosts said US tech companies are expected to put $7 trillion into data centres by the end of the decade.
~45:00The hosts cited a poll from August last year showing 43% of US voters supported data centres being built near them, against 42% opposed.
~45:50