Analysis
Three numbers from the past week of Pulse's own AI infrastructure coverage tell a consistent story about capital intensity. Oracle's remaining performance obligations -- contracted future revenue not yet recognized -- rose $209 billion in a single quarter, per Oracle's disclosures. Microsoft is targeting 38 gigawatts of data center power by 2032, more than triple its current footprint, with roughly a third earmarked for GPU-dense AI compute.
And the Pentagon's Office of Strategic Capital is in talks to lend a large sum to AI cloud startup Fluidstack (see the figure above) -- what would be the office's largest loan to date, extended to shore up the domestic data center supply chain rather than for any weapons program.
“That backlog arrived alongside $28.5 billion in quarterly capital expenditure and negative free cash flow of roughly $5 billion.”
The number that should worry anyone modeling this sector off backlog growth alone is Oracle's own cash flow. That backlog arrived alongside $28.5 billion in quarterly capital expenditure and negative free cash flow of roughly $5 billion.
Set against a $125 billion debt load, backlog measures demand; it says nothing about whether the company financing the supply to meet it can fund the gap between signing a contract and collecting on it.
The Pentagon loan is the more novel data point. A national-security lender extending capital to a private AI cloud company -- not a defense contractor -- is a tacit admission that domestic compute capacity is now treated as strategic infrastructure, alongside energy and semiconductors. Pulse covered SoftBank's own debt-fueled OpenAI exposure today; the Pentagon backstop is the same financing-intensity story with a government balance sheet attached instead of a corporate one.
