Nvidia CFO: Cloud industry backlog exceeds $2T, capex to hit $1.3T in 2027

The AI infrastructure gold rush just got a price tag. During Nvidia’s Q1 FY2027 earnings call on May 20, 2026, CFO Colette Kress disclosed that aggregate backlogs across major cloud providers have climbed to roughly $2 trillion, representing multi-year contracts that hyperscalers have already committed to fulfilling. Hyperscaler capital expenditure is forecast to reach $800 billion in 2026 and $1.3 trillion in 2027.

The numbers behind the supercycle

Kress pegged 2026 hyperscaler capex at somewhere between $650 billion and $775 billion, with some projections pushing above $800 billion once all providers are included. The 2027 number, at $1.3 trillion, would represent a spending level the industry has never approached before.

These commitments reflect contractually committed backlogs at Microsoft, Amazon, Google Cloud, and Oracle, all of which have locked in infrastructure deals to meet surging customer demand for AI compute.

AI infrastructure spending is projected to reach between $3 trillion and $4 trillion annually by the end of the decade.

Nvidia itself is the clearest evidence that this spending is real and accelerating. The company recorded $81.6 billion in revenue for Q1 FY2027, with its data center segment alone contributing $75.2 billion of that total. Q2 FY2027 guidance came in at $96.2 billion in total revenue, with data center revenue hitting $89.0 billion.

What Jensen Huang is seeing from the inside

CEO Jensen Huang cited visibility of at least $1 trillion in cumulative revenue through 2027 stemming from Nvidia’s Blackwell and Rubin platform families.

When ChatGPT launched in late 2022, it set off a procurement frenzy that hyperscalers were caught underprepared for. What followed was a multi-year scramble to build out GPU clusters, data centers, networking, and cooling infrastructure at a pace the supply chain had never had to sustain. That scramble is now showing up as locked-in backlog on balance sheets across the industry.

What this means for the broader market

The risk is that the backlog numbers reflect commitments made during peak enthusiasm, and actual utilization could lag if enterprise AI adoption stumbles or if a more compute-efficient model architecture changes the economics of inference.

Cloud providers are not speculating on future demand: they are responding to paying customers who are already consuming AI services and asking for more capacity. The $2 trillion backlog isn’t a forecast of what might be ordered. It’s a tally of what has already been contracted.