Jefferies: AI Compute Demand Outstrips Supply, Hyperscalers Remain Top Beneficiaries
Summary
- • Jefferies upgrades hyperscaler outlook after Google Cloud Next bullish industry signals
- • GPU scarcity across Hopper and Blackwell pushing enterprises toward committed capacity contracts
- • Cloud capex projected to grow ~64% in 2025 and ~58% in 2026 per Jefferies estimates
- • Enterprises increasingly seeking model-agnostic tooling to manage rising token costs
Details
Jefferies upgrades hyperscaler view post-Google Cloud Next
Jefferies analysts stated they are 'incrementally more positive on AI demand vs our checks in early April,' citing real-world signals at Google Cloud Next that exceeded prior expectations heading into Q1 earnings season.
Google Cloud leads cloud upside skew; AWS high conviction; Azure on Copilot traction
Jefferies ranks cloud upside potential with Google Cloud leading, followed by AWS — where expectations are elevated but conviction remains high — and Azure, where Microsoft is increasingly judged on Microsoft 365 and Copilot traction rather than Azure beat size. Alphabet is positive on trajectory but may be constrained by valuation.
GPU supply tight across Hopper and Blackwell; shift to committed capacity contracts
AI compute scarcity was a dominant theme. Supply is tight across multiple GPU generations including Hopper and Blackwell, with limited spot availability for older chips. This is pushing hyperscalers and enterprises to lock in supply through committed capacity contracts rather than relying on spot markets.
Cloud capex to grow ~64% in 2025 and ~58% in 2026
Jefferies estimates reflect sustained heavy investment in AI infrastructure across hyperscalers, with no near-term signs of spending pullback despite macro uncertainty, underscoring the scale of the ongoing AI buildout cycle.
Enterprise GPU utilization low despite rising adoption; FOMO driving spend
Analysts flagged inefficient enterprise usage — firms are committing to AI deployment cycles driven by competitive pressure and fear of missing out, yet actual GPU utilization is not keeping pace with procurement, suggesting meaningful future efficiency gains remain on the table.
Enterprises want model-agnostic harnesses to control token costs
A key emerging theme: enterprises want orchestration layers that let them route tasks to lower-cost models while maintaining flexibility. Analysts quoted industry feedback: 'enterprises wanted model agnostic harnesses, as they want to optimize token consumption and have control over what model is used.' Frontier models from Anthropic, OpenAI, and Google remain widely used but cost optimization is becoming a strategic priority.
Market Impact = competitive positioning and analyst views; Infrastructure = hardware/capacity dynamics; Financials = spending projections; Insight = behavioral/adoption patterns; Strategy = enterprise tooling trends
What This Means
The AI infrastructure buildout cycle shows no signs of slowing — GPU scarcity, surging capex, and committed capacity contracts all point to a multi-year tailwind for hyperscalers and compute-adjacent players like CoreWeave and Oracle. For AI practitioners and enterprise buyers, the accelerating shift toward model-agnostic orchestration signals that cost optimization and workload flexibility are becoming as strategically important as raw model capability.