# NVIDIA (NVDA) — Qualitative Narrative — FY2027 Q1 Fiscal Q1 FY2027 ended April 26, 2026. **Sources** - Latest 10-K — accession **0001045810-26-000021** (FY2026, year ended Jan 25, 2026). - Latest 10-Q — accession **0001045810-26-000052** (Q1 FY2027, ended Apr 26, 2026). --- ## Business — what NVIDIA does *(Source: 10-K, accession 0001045810-26-000021)* NVIDIA is an accelerated-computing company that has become a **data-center-scale AI infrastructure** supplier. Its moat sits on the **CUDA** development platform (introduced 2006) plus hundreds of domain libraries, SDKs, and frameworks (CUDA-X, NVIDIA AI Enterprise, Omniverse, DRIVE). The company has invested **over $76.7 billion in R&D since inception**, has **over 7.5 million CUDA developers**, and says **more than half of its engineers work on software**. The 2020 Mellanox acquisition added networking and the DPU. **Two reporting segments:** - **Compute & Networking** — Data Center accelerated-computing and networking platforms, AI software, plus Automotive (autonomous/EV, DRIVE). This is the growth engine. - **Graphics** — GeForce GPUs for gaming/PCs, and Quadro/NVIDIA RTX GPUs for enterprise workstation graphics. **Four end markets:** Data Center, Gaming, Professional Visualization, Automotive. **How it makes money / product cadence:** - **Data Center** is the dominant business — rack-scale systems, GPUs, Grace CPUs, DPUs, NVLink switches, and end-to-end **InfiniBand / Spectrum-X Ethernet** networking. NVIDIA moved to a **one-year architecture cadence**: Hopper → **Blackwell** (launched FY2025) → **Blackwell Ultra / GB300** (scaled in FY2026) → **Rubin** (announced FY2026, production shipments expected 2H FY2027, targeting up to ~10x lower cost per token vs Blackwell). It also introduced **NVLink Fusion** to let hyperscalers/custom-ASIC designers integrate their own CPUs/XPUs. - **Gaming** — GeForce RTX 50-series (Blackwell), DLSS, GeForce NOW cloud gaming. - **Professional Visualization** — RTX PRO workstation GPUs; DGX Spark launched in FY2026. - **Automotive** — DRIVE Hyperion / DRIVE AGX / DRIVE OS full-stack AV platform. - Software monetization via paid **NVIDIA AI Enterprise** (NIM, NeMo, AI Blueprints), vGPU, and open model platforms (**Nemotron** for agentic AI, **Cosmos** for physical AI, **Dynamo** inference software). Leadership: co-founder **Jensen Huang** (President & CEO), **Colette Kress** (EVP & CFO). Incorporated in Delaware; HQ Santa Clara, CA. --- ## Risk Factors (substantive) *(Source: re-extracted from raw 10-K, accession 0001045810-26-000021, with current-period updates from 10-Q Item 1A, accession 0001045810-26-000052)* **Export controls / China (the dominant, fast-moving risk).** - April 2025: USG required a license to export the China-specific **H20** to China. NVIDIA took a **$4.5B charge in Q1 FY2026** for H20 excess inventory and purchase obligations as demand collapsed. - August 2025: USG granted licenses to ship certain H20 to certain China customers; NVIDIA generated only ~**$60M** of H20 revenue under them. USG officials expressed an expectation of receiving **15% or more of the revenue** from licensed China sales (not codified in regulation). - February 2026: USG granted licenses to ship **small amounts of H200** to specific China customers, but **no revenue generated to date**; H200 must pass a **U.S. inspection** before shipment and is subject to a **25% tariff** on U.S. importation. - NVIDIA states it is **effectively foreclosed from China's data-center compute market** — it cannot build a product that satisfies both the USG and the Chinese government. **No Data Center Hopper shipments to China in Q1 FY2027** (vs $4.6B in Q1 FY2026). Management warns this foreclosure lets competitors build larger developer/customer ecosystems worldwide. - **China antitrust overhang:** On **Sept 15, 2025**, China's regulators issued a preliminary finding that NVIDIA's export-control-driven product degradation discriminated against Chinese customers and **violated the terms of China's approval of the Mellanox acquisition** — possible financial penalties or restrictions on the networking business. - Regulatory volatility: the **AI Diffusion IFR** (Jan 2025, worldwide tiered licensing) was slated to be rescinded (May 2025) with a still-uncertain replacement; the **GAIN AI Act** and **Remote Access Security Act (RASA)** are pending and could further restrict sales/cloud services. **Supply / demand mismatch & long lead times.** Manufacturing lead times have exceeded 12 months; NVIDIA places **non-cancellable, non-returnable purchase orders** and prepays to secure capacity. Both shortages and excess inventory have hurt results (e.g., low-yield Blackwell inventory provisions dinged Q2 FY2025 gross margin; the H20 write-down). Purchase obligations and prepaids are growing as a share of total supply. **Supplier / foundry concentration.** NVIDIA does not own fabs — it depends on third-party **foundries (TSMC), assembly/test/packaging subcontractors, and memory suppliers**, with a **limited number and geographic concentration** of these partners (heavily Asia: Taiwan, South Korea). Product warehoused/distributed largely from Hong Kong. Any disruption to Taiwan/South Korea supply would hit results. **Customer concentration.** Revenue is concentrated in a few direct and indirect customers. **FY2026:** one direct customer was **22%** of total revenue and another **14%** (both Compute & Networking). **Q1 FY2027:** three direct customers were **21%, 17%, and 16%** of revenue. One AI research/deployment company is estimated to drive a meaningful amount of revenue indirectly via cloud purchases. **Competition.** Customers are building their own **ASICs/XPUs**; some competitors run their own fabs and have deeper IP/resources; cloud rivals compete with NVIDIA's own AI cloud offerings. Rising **open-source foundation models** (incl. DeepSeek, Qwen, Kimi) could shift demand to competitors' platforms if adopted there. **Product defects.** Complex hardware/software can ship with defects or low yields, triggering warranty, recall, inventory write-off, and reputational costs. **Demand cyclicality / new use cases.** GPU demand can spike and drop with "mercurial" applications (historically crypto mining), complicating forecasting; gray-market resale adds noise. **Broader regulatory scrutiny.** Antitrust/competition inquiries from the EU, US, UK, South Korea, Japan, and China (incl. the French Competition Authority); the **EU AI Act**; new U.S. state AI laws (effective Jan 1, 2026). Also flagged: the **Groq IP license** (significant nonrefundable payments that may not translate into commercially viable products); ecosystem-investment counterparty risk; data-center energy/capital availability constraining customer buildouts. --- ## MD&A — FY2026 results & drivers *(Source: 10-K, accession 0001045810-26-000021)* - **Revenue $215.9B, up 65% YoY**, driven by Data Center accelerated computing and AI. Blackwell was the majority of Data Center revenue. - By market: **Data Center +68%**, **Gaming +41%**, **Professional Visualization +70%** (Blackwell + DGX Spark launch), **Automotive +39%**. - Within Compute & Networking: **Data Center computing +59%**; **Data Center networking +142%** (NVLink fabric for GB200/GB300, plus Ethernet & InfiniBand). - **Gross margin 71.1%, down from 75.0%** — the model shifted from selling Hopper HGX systems to full **Blackwell data-center solutions**, plus the **$4.5B H20 charge**. Total inventory/purchase-obligation provisions were **$7.2B** in FY2026 (incl. the $4.5B) vs $3.7B in FY2025; net unfavorable gross-margin impact 2.6%. - **Operating expenses +41%** (compensation/headcount, compute & infrastructure; R&D compute/infra +79%). - **Income tax expense $21.4B**, effective rate **15.1%** (up from 13.3%); reflects OBBBA (enacted July 2025) and a $711M state valuation-allowance release. - Other income aided by unrealized gains on non-marketable and publicly-held equity securities, including gains on the **Intel common-stock investment**. - **Capital return:** added **$60B buyback authorization** (Aug 26, 2025); repurchased **282M shares for $40.4B** in FY2026; **$58.5B remaining** at year-end; paid **$974M** dividends. - **Ecosystem investments:** **$17.5B** into private companies/infrastructure funds; **$3.5B** in land/power/shell guarantees to early-stage partners. **Finalizing an investment and partnership agreement with OpenAI** (not yet signed/assured). - **Capex $6.1B** (vs $3.4B FY2025), expected to rise in FY2027. - **Liquidity:** $62.6B cash/equivalents/marketable securities; $8.5B senior notes; commercial-paper program raised to $25B (none outstanding). Under IRS exam for FY2023–2024; $4.0B unrecognized tax benefits. --- ## Current quarter — Q1 FY2027 (ended Apr 26, 2026) *(Source: 10-Q, accession 0001045810-26-000052)* - **Revenue $81.6B, up 85% YoY and up 20% sequentially.** - **Data Center revenue $75.2B (+92% YoY, +21% sequential)** — driven by the ramp of **Blackwell 300 (GB300)** and InfiniBand/Spectrum-X Ethernet/NVLink. Hyperscalers held at ~**50% of Data Center revenue**; the other half came from AI Clouds, industrial, enterprise, and sovereign customers. **No Data Center Hopper shipments to China** (vs $4.6B in Q1 FY2026). - **Edge Computing revenue $6.4B (+29% YoY, +10% sequential)** — robust Blackwell workstation demand, partly offset by softer consumer PC demand amid elevated memory/system prices. - **Gross margin 74.9%**, up from 60.5% a year ago (prior-year quarter carried the $4.5B H20 charge); roughly flat sequentially. Q1 FY2027 inventory/purchase-obligation provisions were **$1.1B** (net unfavorable margin impact 1.2%). - **Operating expenses +52% YoY** (headcount/comp, compute & infrastructure, engineering development materials; R&D compute/infra +112%, eng-dev materials +204%). - **Income tax expense $11.6B**, effective rate **16.6%** (up from 14.3%). - **Other income:** large mark-to-market gains — **$13.4B unrealized gains on publicly-held equity securities** and **$2.6B on non-marketable equity securities**. - **Ecosystem investing:** **$18.6B** into private companies/infrastructure funds in the quarter. - **New reporting framework introduced this quarter:** two market platforms — **Data Center** (sub-markets **Hyperscale** and **ACIE** = AI Clouds, Industrial, & Enterprise) and **Edge Computing** (PCs, consoles, workstations, AI-RAN, robotics, automotive). Prior periods recast. - **Capital return:** repurchased **108M shares for $20.2B**; **$38.5B remaining** authorization at quarter-end. - **Liquidity:** **$50.3B** cash/equivalents/marketable debt securities plus **$30.2B** marketable equity securities. - **Rubin** platform still expected to start shipping **2H FY2027**. **Litigation (10-Q Note 10):** In the long-running securities class action (*In re NVIDIA Corp. Securities Litigation*, channel-inventory/crypto-demand allegations, 2017–2018), the district court **granted class certification on March 25, 2026** (class = purchasers Aug 10, 2017 – Nov 15, 2018); **NVIDIA filed a Rule 23(f) petition to appeal on April 8, 2026.** Related derivative suits remain stayed. No contingent liabilities accrued (loss deemed reasonably possible but not probable/estimable). --- ## Subsequent events — post Apr 26, 2026 *(Source: 10-Q Note 12, accession 0001045810-26-000052. No standalone subsequent-events note exists in this 10-Q; verified against the raw filing.)* - **May 18, 2026:** Board approved an **additional $80.0 billion share-repurchase authorization** (no expiration). - **May 18, 2026:** Board **increased the quarterly cash dividend from $0.01 to $0.25 per share** — a 25x raise — for shareholders of record **June 4, 2026**, payable **June 26, 2026**. No acquisitions, divestitures, or new financings were disclosed as subsequent events in the 10-Q. (Small tuck-ins would typically surface here; none appear.)