Nvidia CFO Colette Kress announced a $2 trillion backlog of unfilled cloud infrastructure orders during the company's second-quarter earnings call.
Kress projected that hyperscaler capital expenditure will reach $1.3 trillion by 2027, according to the company's earnings announcement. The $2 trillion figure represents cumulative unfilled orders, not annual spending, and reflects multi-year purchasing commitments major cloud operators have made as they expand generative AI infrastructure.
Hyperscalers including Amazon Web Services, Google, Microsoft and Meta have pivoted sharply toward AI spending over the past 18 months, driving record demand for Nvidia's H100 and newer-generation GPU clusters. These companies have already committed capital for deployments extending well beyond the current fiscal year, locking in expected hardware demand even as competition in AI accelerators intensifies.
Nvidia's own second-quarter revenue grew to $30.1 billion, up 126 percent year-over-year, driven almost entirely by data center sales. The company's gross margin expanded to 75.1 percent, the highest in the company's history, as hyperscaler demand outpaced competing alternatives and supply constraints eased.
The backlog announcement comes as larger cloud operators face pressure from investors and customers to justify trillion-dollar AI infrastructure spending. Meta alone is expected to spend $37 billion on AI infrastructure in 2025, while Microsoft has committed to multi-year GPU procurement contracts worth tens of billions. These commitments extend further and run deeper than previous public disclosures have indicated.
Nvidia's $1.3 trillion hyperscaler capex forecast by 2027 implies an average annual spend of roughly $650 billion across the global cloud industry if distributed evenly, more than double the current estimated annual hyperscaler capex of approximately $300 billion. Cloud operators would need to sustain spending at historical peak levels for the next two years to clear the backlog if these projections hold.
The number to watch is whether Nvidia updates this backlog figure in future earnings calls or if the company provides granularity on the portion of the $2 trillion backlog already contractually committed versus provisionally forecasted.
