The AI infrastructure arms race is heating up, but the financial health of the hyperscalers is showing cracks. Microsoft’s Azure grew 48% in the latest quarter, and AWS expanded 39.3%, yet these top-line victories mask a severe liquidity squeeze. Operating cash flow is not keeping pace with the massive capital expenditures required to build out data centers, leading to a significant drop in cash remaining after capex.

The Cash Flow Crisis

The numbers are stark. Microsoft’s cash remaining after cash capital expenditure fell about 27% year-on-year, despite a 29% increase in operating cash flow. Meta reported a negative $2.12 billion in quarterly free cash flow, while Amazon’s trailing twelve-month free cash flow has sunk to negative $42.4 billion. This divergence highlights a critical infrastructure reality: spending on GPUs and data centers is rising faster than the operating cash flow those assets generate.

Valuation Dislocations and Hidden Risks

Analyst Michael Hillaert notes that while negative free cash flow during a buildout isn't inherently fatal, the uncertainty regarding future GPU replacement costs and sustained competition makes estimation difficult. He argues that the difficulty lies in predicting how much spending will be needed after the first wave of data centers is built, as GPUs must be replaced and competition may keep investment high.

Supplier Financing Risks

Broadcom’s involvement in a $63 billion financing arrangement for Anthropic’s infrastructure illustrates how suppliers are now taking on residual-value risk, effectively financing their own customers. NVIDIA has also been moved to 'Conditional dislocation' status, as its exceptional performance remains tethered to customers' ability to continue heavy spending.

Where the Value Is Hiding

Amidst the hyperscaler turmoil, software companies with established data moats are emerging as more stable investments. Intuit remains the top pick at 13.9× forward earnings, followed by RELX and Adobe. These companies benefit from AI integration without bearing the crushing capital expenditure burden of the infrastructure builders. Memory companies like SK hynix and Micron trade at low multiples of 3–5× forward earnings, but their profitability is vulnerable to capacity cycles and contract repricing.

Key Takeaways

  • Hyperscaler free cash flow is negative or declining sharply due to AI capex.
  • Broadcom is taking on residual-value risk in Anthropic's $63B financing deal.
  • Software firms with existing data assets offer better risk-adjusted returns than infrastructure giants.
  • NVIDIA and Broadcom are now classified as 'Conditional dislocations' due to demand uncertainty.

The Bottom Line

We are building the plumbing for the next decade of AI, but we are currently financing it with shareholder equity rather than operating profits. Until these data centers generate sufficient revenue to offset their depreciation and replacement costs, the 'AI boom' remains a capital-intensive gamble with poor near-term returns for investors. This infrastructure phase is unsustainable if it continues to burn cash faster than it generates it. Developers and investors alike need to watch the cash flow statements, not just the revenue headlines.