Nvidia CFO says about half of its data center business comes from customers beyond hyperscalers

Nvidia’s fiscal Q2 revenue surged to $96.2B (+106% YoY) and Data Center revenue jumped to $89.0B (+117% YoY), with non-GAAP EPS of $2.22. The company guided Q3 sales to $105.8B–$110.1B and projected fiscal 2028 annual sales to rise 70% vs the prior year, reinforcing “demand is accelerating.” Management also argued growth is increasingly diversified beyond hyperscalers, with AICE expected to represent ~half of Data Center revenue, and cited AI VC compute-linked funding exceeding $400B in H1 2026 (vs $265B in all of 2025).

Analysis

This is less about one company beating and more about the AI capex base getting broader, which matters for valuation. If demand is migrating from a few hyperscalers into sovereign, neo-cloud, and enterprise buyers, the market can justify a longer depreciation cycle for the whole AI stack and a lower cyclicality discount on NVDA. The immediate winners are NVDA and the semi infrastructure complex (SMH/SOXX), while the larger cloud platforms (AMZN, MSFT, GOOGL, META) get an indirect boost because their AI spend is now easier to frame as ecosystem leadership rather than one-off arms race expense.

The catch is that non-hyperscaler demand is usually the most financing-sensitive part of the stack. If those customers are leaning on VC, private credit, or lease financing, the risk is not demand collapse today but a 1-3 quarter air pocket if capital markets tighten; that would hit NVDA orders before it shows up in revenue. The key falsifier is not sentiment — it is whether next-quarter data center growth and backlog conversion stay at these rates once easier comp effects fade.

Consensus may be underestimating second-order supply effects: sustained AI spend keeps pressure on power, networking, and memory, but it also delays any meaningful digestion in the semiconductor group because buyers are still under-installing relative to training/inference demand. The overdone part is assuming that every new end-market is equally durable; enterprise software monetization remains the weakest proof point, so a lot of the long-duration upside is still contingent on software budgets expanding, not just model usage. If that monetization stalls, the whole thesis reverts from 'structural' to 'capital-intensive and cyclical' quickly.

Key Decisions for Investors

• Add NVDA on any 3-5% post-earnings pullback over the next 1-2 sessions; target a 10-15% move over 1-3 months if the market re-rates AI demand durability. Invalidate if next quarter guide or data center sequential growth decelerates materially.
• Buy SMH call spreads for a 4-8 week window to express the sector breadth read-through with defined risk; this is cleaner than outright chasing NVDA after a strong print. Cut if semis fail to hold relative strength versus QQQ within 2 weeks.
• Pair trade: long NVDA / short AMD for 1-3 months. Thesis is that the strongest demand visibility and ecosystem lock-in should keep share gains concentrated at the top end; cover if AMD commentary shows meaningful design-win acceleration or better AI attach rates.
• Watch, don’t chase, FIG and other AI-enabled enterprise software names into this narrative. They only work if earnings show real monetization, not just usage growth; if bookings/revenue conversion disappoints, the 'AI adoption is broadening' thesis is overextended.