Jensen Huang says Nvidia will sell twice as many chips next year

Nvidia CEO Jensen Huang said the company expects to sell twice as many chips next year as this year, signaling sustained AI-driven demand growth over the next six quarters. The outlook follows Nvidia's forecast for roughly 70% growth in the fiscal year ending January 2028, implying about $673 billion in revenue and prompting higher analyst expectations. Huang attributed demand to broad global investment in AI, while also emphasizing that unsafe AI products should be withheld until further engineered.

Analysis

The key market mechanism is not unit growth alone but mix: a doubling of chip volumes can coexist with lower blended ASP if incremental shipments skew toward inference, enterprise, edge, or sovereign deployments rather than frontier-training systems. That would still deepen NVDA’s ecosystem moat and attach opportunity in networking/software, but it makes consensus revenue and gross-margin extrapolation more fragile than the headline implies. The cleanest second-order beneficiaries are capacity-constrained inputs—TSM, SK Hynix, MU and CoWoS-adjacent supply chains—plus AI-networking/power infrastructure names AVGO, ANET, VRT and CRDO.

Over the next 1-3 months, the upside catalyst is a further sell-side reset to forward demand assumptions and evidence that non-hyperscaler demand is becoming material. The principal falsifier is not a weaker shipment number; it is evidence of lower system-level monetization: cloud capex moderation, declining accelerator lead times, a gross-margin guide below expectations, or rising customer concentration/receivables. Export restrictions and power-grid bottlenecks remain the most credible 6-18 month constraints, because they can redirect demand into lower-margin geographies or delay data-center commissioning despite chip availability.

Consensus is likely treating every incremental AI deployment as equally accretive to NVDA. A more nuanced outcome is that broad adoption expands total units but shifts profit pools toward memory, networking, optics and data-center electrical equipment as AI clusters scale. GOOG is directionally helped by cheaper, more abundant compute, but its equity sensitivity depends on whether AI usage converts into Search/Cloud revenue faster than inference costs increase; the announcement alone is not a standalone GOOG catalyst.

Key Decisions for Investors

• Maintain NVDA overweight into the next earnings cycle, but express incremental upside through a 3-6 month long NVDA / short SOXX hedge rather than unhedged beta. Thesis requires another upward revision to revenue or backlog visibility; exit the relative long if NVDA guides gross margin materially down or lead times normalize.
• Buy a 6-12 month basket of TSM, MU and SK Hynix exposure versus a short in broad semiconductor beta (SOXX) where mandate permits. These names gain from higher accelerator content regardless of whether NVDA’s incremental unit mix carries lower ASPs; key risk is HBM/advanced-packaging capacity expansion arriving faster than demand.
• Add selectively to VRT and ANET on pullbacks rather than chase NVDA’s immediate reaction. Their revenue realization should lag semiconductor shipments by roughly one to three quarters as clusters are commissioned, offering a later-cycle catalyst; invalidate if hyperscaler capex guidance turns down or data-center project schedules slip.
• Do not add GOOG solely on this signal. Set an alert for Cloud growth acceleration paired with stable or improving capex-to-revenue conversion; absent that evidence, rising AI compute availability can be as much a cost-pressure signal as a monetization catalyst.