Nvidia's Demand Is Outpacing Its 70% Growth Forecast, and That Is Just the Beginning

Nvidia management indicated fiscal 2028 revenue could double based on customer demand, although it currently forecasts approximately 70% growth because of supply constraints for manufacturing inputs. CFO Colette Kress said customer forecasts imply growth could be twice as high, while CEO Jensen Huang said agentic AI workloads require 15x to 100x the compute of human-prompted AI. The outlook reinforces a multi-year GPU demand tailwind, supported by S&P Global's estimate of $1.3 trillion in AI spending next year, but execution remains constrained by industry-wide component supply.

Analysis

The investable signal is not incremental NVDA demand optimism; it is the widening value of scarce inputs. If accelerator availability is the binding constraint, price realization and allocation power migrate upstream to HBM and advanced packaging, where MU has materially greater earnings torque than NVDA on a percentage basis. The second-order beneficiary set includes TSMC (TSM), Broadcom’s custom-AI supply chain (AVGO), and power/cooling vendors, while hyperscalers face delayed AI-service revenue recognition and continued depreciation pressure before utilization catches up.

Management’s upside case should not be capitalized as base-case revenue: customer forecasts can reflect strategic capacity reservations rather than deployable, financed demand. The key 1-3 month catalyst is whether NVDA converts the supply narrative into higher backlog, sustained gross-margin guidance, and evidence that lead times—not customer ROI—are limiting shipments. A weaker-than-expected cloud capex outlook, rising cancellation/reallocation rates, or gross margin dilution from supply-expansion actions would compress the scarcity premium quickly, particularly at NVDA’s exceptional valuation.

Contrarianly, supply shortages can be less bullish for NVDA than for memory suppliers if constrained systems force customers to prioritize lower-memory configurations, alternative accelerators, or internally designed ASICs. Over 6-18 months, capacity normalization creates a bifurcation: AI infrastructure spending may remain high, but NVDA’s unit economics and valuation multiple become more sensitive to competition from AMD (AMD), AVGO custom silicon, and hyperscaler chips. The more durable expression is therefore selective ownership of bottleneck suppliers rather than adding unhedged NVDA beta after narrative-driven strength.

Key Decisions for Investors

• Favor a 3-6 month long MU / short NVDA pair, sized beta-neutral: MU has greater sensitivity to sustained HBM scarcity and memory pricing, while the short leg hedges broad AI-capex enthusiasm. Exit if MU fails to raise HBM mix or gross-margin outlook at its next earnings; reduce if NVDA’s supply expansion lifts availability without margin dilution.
• Maintain NVDA core exposure but do not add solely on the conditional upside-demand framing; use a 5-10% pullback or post-earnings confirmation of backlog and gross-margin durability for entry. Thesis invalidation: a guide-down in data-center revenue, material gross-margin compression, or hyperscaler capex reductions.
• Add a 6-12 month basket of TSM and AVGO against an underweight in AI-adjacent software with weak monetization evidence. TSM captures advanced-node/packaging utilization, while AVGO benefits if customers diversify toward custom silicon; principal risk is a broad capex digestion cycle.
• Monitor quarterly cloud capex guidance from MSFT, META, AMZN, and GOOGL as the decisive demand-quality check. If aggregate capex growth decelerates while NVDA lead times shorten, rotate out of the scarcity trade rather than treating increased supply as incremental upside.