NVIDIA Corporation (NVDA) Earnings

Management highlights

- **Record Financials**: Delivered record revenue ($96B), operating income, and EPS, driven by a global AI infrastructure build-out across hyperscalers, enterprises, and sovereign customers. - **Supply Constraints vs. Demand**: Revenue growth is currently supply-constrained. While unconstrained demand would be significantly higher, NVIDIA expects ~70% growth in FY28 due to bottlenecks. - **Product Roadmap**: Vera Rubin production has commenced, expected to see the fastest product ramp in company history. It offers 30x higher throughput per megawatt and 35x lower token costs compared to Grace Blackwell Ultra. - **Full-Stack Platform**: NVIDIA emphasizes its 'AI Factory' platform (CPU, GPU, Networking, Software) which captures a larger share of the data center TAM, expanding revenue opportunity from $18B/GW (Hopper) to $40B/GW (Vera Rubin). - **Neocloud & Sovereign AI**: The ACIE segment (Sovereign, Neocloud, Enterprise) grew rapidly, representing roughly half of the data center business. NVIDIA partners with regional clouds (e.g., in Japan, India, Australia) providing land/power/shell support and revenue-sharing models. - **Frontier AI Labs Support**: NVIDIA invested nearly $50 billion in Frontier AI Labs and partnered with major capital providers (Apollo, BlackRock, etc.) to raise $500 billion in third-party financing to support their infrastructure builds, citing limited risk due to the fungibility of NVIDIA compute. - **Ecosystem Dominance**: NVIDIA runs all leading closed (OpenAI, Anthropic) and open models. The CUDA ecosystem ensures durability and fungibility, making it the preferred infrastructure for agentic workloads.

Guidance

- **Q3 FY27 Revenue**: Expected to be $108 billion (+/- 2%). - **FY27 Full Year Revenue**: Preliminary expectation is for approximately 70% year-over-year growth. - **Gross Margins**: Reset expectations due to rising memory costs. Q3 gross margins are expected to be 74% (+/- 50 bps). Margins are expected to bottom in Q4 at 71%-72%, before settling at 72%-73% in FY28 as price increases take effect. - **Operating Expenses**: Q3 OPEX expected to be ~$9.2B (GAAP) / ~$9.0B (Non-GAAP). Full-year FY27 OPEX growth is expected in the low 50s percentage range. - **Tax Rate**: Full-year FY27 GAAP/Non-GAAP tax rates expected between 16% and 18%. - **Forward Visibility**: Management explicitly stated they do not provide long-term guidance beyond the current fiscal year outlook due to the dynamic nature of supply chain constraints.

Segment performance

Total revenue reached $96 billion, more than doubling year-over-year. Data Center revenue was $89 billion (representing approximately 93% of total revenue), up 18% sequentially. This segment is split into two main sub-segments: Hyperscale revenue was $49 billion (growing 13% sequentially), and ACI & Enterprise (ACIE) revenue was $40 billion (growing 25% sequentially and 138% year-over-year). Gross margins for both GAAP and Non-GAAP were 75%, unchanged from the previous quarter.

Risks & headwinds

- **Supply Chain Bottlenecks**: Growth is constrained by supply availability, particularly in memory components (DRAM) and manufacturing capacity. Memory prices have risen significantly faster than anticipated. - **Geopolitical Restrictions**: No China data center compute revenue is included in the forward outlook due to ongoing US government license restrictions. Current Hopper shipments to China are dilutive to margins. - **Customer Custom Silicon**: Some key customers (e.g., OpenAI, Anthropic) are developing custom chips. However, management argues these will still rely on NVIDIA's broader full-stack platform and networking solutions. - **Financing Risks**: While NVIDIA views its investments in AI Labs as low-risk due to asset redeployability, there is inherent risk in supporting balance-sheet-constrained startups with large infrastructure commitments.

Analyst Q&A

• Q: Analyst asked about the basis for the 70% FY28 growth guidance and confidence levels despite supply constraints.

  A: Jensen Huang explained that unconstrained demand would be much higher (potentially 100%+ growth). He highlighted that NVIDIA’s unique position allows it to capture not just hyperscaler spend but also the massive, invisible market of sovereign AI, neoclouds, and enterprises. He noted that securing land, power, and shell infrastructure takes years, giving NVIDIA greater visibility into future demand pipelines upstream and downstream. The 70% figure reflects the reality of supply chain limitations rather than lack of demand.

• Q: Analyst asked about inference market share evolution, specifically regarding agentic AI workloads and the impact of Grok 3 LPX.

  A: Huang emphasized that the AI lifecycle (data prep, training, inference) is becoming increasingly complex, favoring NVIDIA’s fungible architecture. The Vera Rubin platform addresses the entire lifecycle, preserving customer investment longer than specialized single-phase chips. Regarding Grok 3 LPX, he described it as a high-interactivity accelerator that complements the main Vera Rubin/Multi-NVL72 systems. While Grok targets specific high-ASP use cases, the vast majority of data centers will continue to rely on the core Vera Rubin platform for general-purpose agentic inference.

• Q: Analyst asked if NVIDIA’s strategy remains viable given that major AI labs like OpenAI are designing their own custom chips.

  A: Huang asserted that NVIDIA provides an entire 'AI Factory' platform spanning the whole lifecycle, whereas custom chips are often inference-specific or cloud-specific. He expressed 100% confidence that even companies building custom silicon will run on NVIDIA platforms globally due to the superior economics, productivity, and versatility of the CUDA ecosystem. He framed the relationship as symbiotic, noting that NVIDIA’s investments in these labs are designed to scale their success, which in turn drives demand for NVIDIA’s comprehensive infrastructure.

• Q: Analyst asked about the impact of open-source models on NVIDIA’s growth trajectory.

  A: Huang stated that both closed and open models are skyrocketing and nearly all open models run on NVIDIA due to its ubiquitous footprint and fungibility. He argued that open models are vital for enterprises and countries to build proprietary, domain-specific AI and for cybersecurity applications. Rather than cannibalizing sales, the rise of frontier-level open models enables broader adoption and creates more demand for compute, as every entity needs to rent intelligence while building their own alpha. Both model types drive NVIDIA sales simultaneously.

• Q: Analyst asked about the most acute supply constraints (power, DRAM, wafers) limiting the gap between actual and unconstrained demand.

  A: Huang confirmed that the entire supply chain is challenged and running flat out. He did not rank specific constraints but indicated that DRAM scarcity is a significant factor driving margin pressure. He emphasized that capacity comes online gradually throughout the year, improving yields daily. Currently, NVIDIA has enough supply to meet the 70% growth target, but meeting the much higher unconstrained demand requires massive cooperation from the entire supply chain to expand capacity continuously.