Post

Analysts estimate that 70%+ of Microsoft, Google and Amazon's present and future AI revenues are from Anthropic and OpenAI's compute spend and model sales. Outside of two unsustainable AI labs, demand barely exists, and they've massively overbuilt capacity. www.wheresyoured.at/the-ai-deman...

Last week saw Amazon, Google, and Microsoft congratulated for strong quarterly numbers — with outlets patting them on the back for how their AI bets were “paying off,” despite none of them disclosing their actual AI revenues. www.wheresyoured.at/the-ai-demand-bubble/

Hyperscalers have conned investors by not disclosing their AI revenue concentration, all to hide that revenue growth is increasingly coming from Anthropic and OpenAI, and that their infrastructure was near-entirely built to support two unprofitable AI labs. www.wheresyoured.at/the-ai-demand-bubble/

According to analyst letters from UBS, Wells Fargo and Barclays, Anthropic and OpenAI make up 70% or more of Google, Microsoft and Amazon's AI revenues, which means that outside of these AI labs, there isn't demand for compute or AI software at scale. www.wheresyoured.at/the-ai-demand-bubble/

Hyperscalers have sold the media and investors on the idea that $990bn in capex would create diverse AI demand. Instead they've destroyed their free cash flow to prop up unsustainable AI labs who can only pay when hyperscalers hand them the money to do so. www.wheresyoured.at/the-ai-demand-bubble/

This affirms what I’ve said again and again — there really isn’t an AI industry without Anthropic and OpenAI, as demonstrated by the fact that there’s very little AI revenue beyond these two labs. Without them, there's no justification for further capex. www.wheresyoured.at/the-ai-demand-bubble/

These analyst estimates also show that there’s very little demand for the AI products of the hyperscalers themselves - barely $10bn in FY2027 for Microsoft 365 Copilot, per Wells Fargo - giving them no hope of recouping their investment. www.wheresyoured.at/the-ai-demand-bubble/

Anthropic and OpenAI are ticking time bombs and existential parts of the growth of major hyperscalers. And to make matters worse, to hit these estimates, both AI labs must spend upwards of $150bn a year in 2027 onward for hyperscalers to keep growing. www.wheresyoured.at/the-ai-demand-bubble/

The fact that the only demand for AI compute stems from OpenAI and Anthropic means that these companies cannot fail AND must grow to ridiculous heights. When they fail, hyperscalers will have to revise guidance, slash RPOs, and admit capex was wasted. www.wheresyoured.at/the-ai-demand-bubble/

Neither OpenAI nor Anthropic can actually afford their compute spend, and the only way they'll have the demand for said compute is if they grow at the fastest pace in tech history, all while raising $500bn+ in venture capital in the next four years. www.wheresyoured.at/the-ai-demand-bubble/

Hyperscalers have built — and are building — too much capacity, and only continue to do so because they are dependent on Anthropic and OpenAI's compute spend to hit their growth targets. I worry that 90% of data centers go unused and their debts unpaid. www.wheresyoured.at/the-ai-demand-bubble/

Microsoft, Google, and Amazon are selling their investors a lie about their AI strategies, and in a properly-regulated market would force them to file disclosures that document the heavy revenue concentration of Anthropic and OpenAI’s compute spend. www.wheresyoured.at/the-ai-demand-bubble/