The AI Boom Is Only in the Foothills
AI stocks sold off hard last week, and the Financial Times called it a dress rehearsal for an AI reckoning. Is the AI bubble about to burst? Neil Woodford thinks the AI boom is only in the foothills, but he has a clear view on who is holding the risk if it doesn't pay for itself.
AI stocks sold off hard last week, and the Financial Times called it a dress rehearsal for an AI reckoning. Is the AI bubble about to burst? Neil Woodford thinks the AI boom is only in the foothills, but he has a clear view on who is holding the risk if it doesn't pay for itself.
Anthropic's IPO prospectus gave the first look at the finances of a leading AI lab: an $80bn loss in the first half of 2026, and at least $518bn of computing power committed over the next decade, about 80% of which can't be cancelled. That money is revenue for Google, Amazon and Microsoft. It's a big part of why the largest spenders are expected to put around $785bn into data centres and chips this year, and more of it is now being borrowed.
Neil and Jon follow the money down the chain, from the AI labs to the cloud companies, the neoclouds and the lenders behind them. They look at why the $80bn loss matters less than the headline suggests, what Meta's latest results say about funding the build-out, and the two threats that could undermine all of it: planning refusals for data centres, and cheap open-weight AI models from China.
Then Neil's view: why he thinks AI spending carries on, why it could run slower but for longer, what the telecoms bust of 2000 does and doesn't tell us, and the one part of the AI build-out that worries him most. Finally, what it means for investors, including anyone in an index tracker who may not realise how much of the AI trade they already own.
AI stocks sold off hard last week, and the Financial Times called it a dress rehearsal for an AI reckoning. Is the AI bubble about to burst? Neil Woodford thinks the AI boom is only in the foothills, but he has a clear view on who is holding the risk if it doesn't pay for itself.
Anthropic's IPO prospectus gave the first look at the finances of a leading AI lab: an $80bn loss in the first half of 2026, and at least $518bn of computing power committed over the next decade, about 80% of which can't be cancelled. That money is revenue for Google, Amazon and Microsoft. It's a big part of why the largest spenders are expected to put around $785bn into data centres and chips this year, and more of it is now being borrowed.
Neil and Jon follow the money down the chain, from the AI labs to the cloud companies, the neoclouds and the lenders behind them. They look at why the $80bn loss matters less than the headline suggests, what Meta's latest results say about funding the build-out, and the two threats that could undermine all of it: planning refusals for data centres, and cheap open-weight AI models from China.
Then Neil's view: why he thinks AI spending carries on, why it could run slower but for longer, what the telecoms bust of 2000 does and doesn't tell us, and the one part of the AI build-out that worries him most. Finally, what it means for investors, including anyone in an index tracker who may not realise how much of the AI trade they already own.