Nvidia Says Its Cloud Customers Are Sitting on a $2 Trillion Backlog -- Here's What Nvidia's Cut Is Worth by 2028

Nvidia reported fiscal Q2 revenue up 106% YoY to $96.2B and data center revenue up 117% YoY to $89.0B, underscoring sustained hyperscaler demand. Management guided fiscal 2028 revenue growth to ~70%, explicitly attributing the cap to supply constraints despite “demand much greater than 70%,” and flagged gross margin pressure as memory prices rise (gross margin bottoming at ~71–72% vs 75% in Q2). The hyperscalers’ capex outlook is cited at nearly $800B in 2026 and $1.3T in 2027, implying Nvidia could generate roughly ~$315B in hyperscaler-derived revenue on that run-rate math (about half of Nvidia’s data center revenue), supporting a bullish valuation reset even as the stock trades at ~27x earnings (and ~14x vs fiscal 2028 estimates).

Analysis

This reads less like a one-quarter beat and more like evidence that the AI capex cycle is still in the monetization phase for the best-in-class vendor. The key mechanism is supply allocation: when a platform leader is capacity-constrained, incremental dollars migrate to whoever can actually ship, which supports pricing power and preserves backlog visibility. That is bullish not only for NVDA, but also for the narrow set of bottleneck suppliers in advanced packaging, foundry, networking, and power/thermal infrastructure; by contrast, weaker GPU challengers and custom-silicon programs risk being crowded out unless they can prove materially better perf-per-dollar.

The first-order risk is that the market is extrapolating capex into revenue too cleanly. A meaningful share of hyperscaler spend is non-semiconductor infrastructure, so the revenue capture rate is likely to drift lower if power, land, and data-center construction absorb a bigger slice of budgets. Over 1-3 months, the important catalyst is not demand, but whether any hyperscaler trims 2026/27 spend or shifts mix toward in-house silicon; that would pressure NVDA’s multiple more than near-term reported growth. Over 6-18 months, rising memory and packaging costs matter because they can compress gross margin even if units keep growing.

Contrarian view: the consensus is treating “supply constrained” as uniformly bullish, but scarcity can simply defer revenue rather than create it. If customers re-route budgets into networking, compute adjacency, and custom ASICs, the capex boom can keep expanding while NVDA’s share silently plateaus. The thesis breaks if hyperscaler capex guidance rolls over, if gross margin bottoms meaningfully below the implied range, or if competing accelerators start winning measurable workload share in earnings commentary from AMZN, GOOGL, or META.

Key Decisions for Investors

• Maintain a core long NVDA, but add only on 3-5% pullbacks over the next 2-4 weeks; the stock is still the cleanest expression of AI capex with the best visibility into 2026-27 earnings.
• Pair trade: long NVDA / short AMD for a 6-12 month horizon. The spread should work if hyperscaler spend stays concentrated in the incumbent platform, with AMD acting as the cleaner hedge against share-transfer optimism.
• Add a smaller long in TSM or AVGO as a picks-and-shovels hedge to the NVDA trade. If supply remains the bottleneck, the ecosystem monetizes even when NVDA revenue is capped by manufacturing.
• Watch hyperscaler capex commentary in the next earnings cycle; if AMZN, GOOGL, or META signal a pause or mix shift toward custom silicon, cut NVDA exposure and rotate to networking/power names instead.
• If implied volatility stays reasonable, consider a bullish call spread in NVDA for the next 6-9 months rather than outright stock to express upside while limiting drawdown if the market de-risks on margin concerns.