The $700 billion in AI-related capital expenditure announced by major tech companies is not speculative. It’s a direct response to $1.5 trillion in committed cloud orders, a figure that doubled since late 2023. This colossal backlog, detailed on FYI, represents a decade of pre-sold compute, primarily to a handful of AI labs.
The fuel for this machine is venture capital. On The AI Daily Brief, Nathaniel Whittemore notes Anthropic’s $200 billion deal with Google Cloud comprises over 40% of Google’s reported backlog. Similarly, OpenAI and Anthropic account for almost half of the total $2 trillion in backlogged orders across Microsoft, Oracle, Google, and Amazon. This creates a closed loop where VC dollars flow to startups, who immediately spend them on cloud infrastructure from their largest potential competitors.
“The AI boom rests on a precarious financial loop… giants like Microsoft and Nvidia enter complex equity and cloud deals with their own customers.”
- Rohit Chopra, Breaking Points
This circularity, warns former CFPB director Rohit Chopra on Breaking Points, creates systemic risk reminiscent of pre-2008 financial linkages. If revenue from these AI startups fails to meet Wall Street expectations - OpenAI has already delayed its IPO - the valuations propped up by this spend could crack. The top 1%’s heavy exposure to these tech stocks, Chopra argues, makes the broader economy vulnerable.
Facing astronomical compute costs, AI labs are ruthlessly prioritizing their highest-margin customers: enterprises using coding agents. As Whittemore explains, OpenAI walked away from a billion-dollar Disney deal and shuttered its Sora app because enterprise API users consuming tokens at scale dwarf subscription revenue. Anthropic’s annualized revenue reportedly surged from $14 billion to $44 billion by focusing on work-related usage.
The outlier is Meta. While the industry chases enterprise seats, Mark Zuckerberg is spending $145 billion this year on infrastructure to build consumer agents for shopping and personal tasks, forecasting integration into Instagram by year’s end. As FYI analysts note, the market punishes Meta’s capex because it lacks a third-party cloud business, but its internal ROI - driving up Facebook video watch time by 8% - may be more substantial than a pure rental model.
“Enterprise customers have pricing power because removing AI tools would make their organizations extremely unproductive.”
- Brett, FYI
The shift is structural, not cyclical. As Brett from ARK Invest argues on FYI, AI agent usage represents a productivity expansion, not just time-saving. A third of the startups he’s invested in already spend as much on AI tokens as on labor. This creates immense pricing power for model providers and suggests the $1.5 trillion backlog is less a bubble and more the foundation of a new industrial layer, one where compute is the essential commodity and Big Tech owns the mine.
