NVIDIA Projects $1T Hyperscaler Capex in 2027; Vera Rubin Supply Constrained Through Lifecycle
During its Q1 FY2027 earnings call, NVIDIA provided a striking projection: hyperscaler AI capex will climb to $1 trillion in 2027, doubling the ~$750 billion deployed in 2026. This forecast is not speculation—it reflects signed commitments and backlog from the five largest cloud providers (Amazon, Alphabet, Microsoft, Meta, Oracle), plus sovereign AI investments (Saudi Arabia, Japan, Australia). CEO Jensen Huang stated bluntly that 'demand has gone parabolic' and 'agentic AI has arrived,' signaling that the AI factory buildout is accelerating faster than even aggressive 2026 estimates anticipated.
NVIDIA's forward guidance of $91 billion in revenue for Q2 FY2027 (up from ~$39 billion in Q1) reflects this intensity. Gross margins held at 75%, demonstrating sustained pricing power despite competition from AMD custom chips and hyperscaler in-house accelerators. However, Huang explicitly warned that NVIDIA 'will be constrained throughout the entire life of Vera Rubin'—the company's next-generation AI supercomputer system launching in 2026–2027. Vera Rubin, composed of 72 Rubin GPUs and 36 Vera CPUs, is claimed to deliver 10x more performance per watt than Grace Blackwell and is projected to be 'even more successful' than Grace Blackwell. Yet the supply signal is unmistakable: NVIDIA expects demand to outpace supply for years.
This has downstream implications for the entire AI stack. NVIDIA's data center segment now accounts for 90%+ of revenue (gaming is <8%), making this constraint a limiting factor for any cloud provider or startup trying to scale AI capacity in 2026–2027. Broadcom, Marvell, Micron, and SK Hynix—all suppliers of networking, memory, and interconnect components—will feel demand pressure from NVIDIA's supply constraints cascading into their own capacity planning. For architects evaluating compute procurement strategy, the message is: expect Vera Rubin allocation scarcity in 2026–2027 and plan alternative fallbacks (H200 refresh, custom chips, neo-cloud GPU rental) if your buildout timeline is aggressive.
Sources
- Primary source
- cnbc.com
“CEO Jensen Huang: 'Demand has gone parabolic. The reason is simple: Agentic AI has arrived. Blackwell sales are off the charts, and cloud GPUs are sold out.' Vera Rubin expected to be even more successful; Huang said he anticipates NVIDIA 'will be constrained throughout the entire life of Vera Rubin.'”
- nvidianews.nvidia.com
“NVIDIA during Q3 FY2026: data center revenue majority of sales, gross margin 75%, Vera Rubin deliveries started; partnership announced with Anthropic (1GW capacity), OpenAI (10GW), and others driving capacity bottlenecks”
- fool.com
“During Nvidia's latest conference call, it projected that 2027 AI hyperscaler capital expenditures will rise to $1 trillion next year”