The free cash flow machine is dead. Tech's largest companies are undergoing a violent structural shift from asset-light software into capital-intensive industrial utilities. According to Chamath Palihapitiya on All-In, Microsoft, Google, and Amazon are signing decade-long power contracts at more than double the spot electricity rate just to guarantee supply for their AI data centers. This trillion-dollar capital expenditure cycle, led by Amazon's $200 billion, Microsoft and Google's $190 billion each, and Meta's $145 billion in 2026 guidance, is dismantling shareholder returns.
Amazon’s free cash flow cratered 97%. The spend is a direct response to a physical compute bottleneck. On FYI, Nick Grous detailed that the backlog of committed enterprise cloud spend across the Big Three has ballooned to roughly $1.5 trillion, nearly doubling since late 2023. Microsoft alone carries $600 billion in performance obligations. The market is supply-constrained, not demand-limited, and every token produced is sold out.
This scarcity has turned compute capacity into a strategic weapon. David Sacks argued on All-In that Anthropic's most advanced model, Opus 4.7, is currently "compute-gated," forcing users to older versions. Sam Altman's early bet on massive data center spend gave OpenAI a supply lead. Even if OpenAI missed consumer targets, it now wins enterprise deals simply by having tokens to serve.
"The hyperscalers in five years are going to look like highly leveraged, debt on the balance sheet, industrial businesses."
- Chamath Palihapitiya, All-In
The industry is moving past the model wars into the era of the harness. Nathaniel Whittemore explained on The AI Daily Brief that the runtime environment - the persistent memory and tool dispatch system - is now the primary location of intelligence. Benchmarks show the same model performs drastically better in a superior harness. Startups like Cursor are selling "Harness as a Service," allowing developers to rent agent runtimes like cloud compute, which democratizes creation but commoditizes the underlying models.
Worker behavior confirms AI is now essential infrastructure, not a time-saver. Brett from ARK's team noted that once employees adopt agentic workflows, reverting to manual processes feels like a capability loss. A third of the startups he invests in already spend as much on AI tokens as on labor. This creates immense pricing power for providers; removing these tools would make organizations deeply unproductive.
"If you took the AI tools away from those customers, their organizations would be so extremely unproductive."
- Brett, FYI - For Your Innovation
The financial circularity underpinning the boom introduces systemic risk. On Breaking Points, former CFPB director Rohit Chopra warned of a "circularity" where giants like Microsoft and Nvidia enter complex equity and cloud deals with their own customers, inflating valuations without traditional cash flow. He compared the lack of oversight to the pre-2008 era. If revenue expectations for frontier labs miss targets, the entire interconnected chain could crack.
The arms race is set. The companies that secure the physical infrastructure - chips, power, land - will own the economic value per token for the next decade, even as their balance sheets transform beyond recognition.
