NVIDIA's $216B Year: Rubin Ships, Groq Absorbed, and the AI Factory Machine Accelerates

NVIDIA closed fiscal 2026 with $216B in revenue and $120B in net income, then unveiled the Vera Rubin platform and absorbed Groq's inference IP for $20B. With Q1 FY2027 guided at $78B, the company is now selling complete AI data center stacks — from GPUs to CPUs to networking to inference co-processors.

NVDA · Information Technology · April 15, 2026

S&P 500 Position

NVIDIA holds the largest single-stock weight in the S&P 500, a historic record since index weight data began in 1981. Within Information Technology, it dwarfs semiconductor peers: its $4.8T market cap exceeds Broadcom ($1.5T), AMD (~$200B), and Intel (~$130B) combined. The top 7 S&P 500 companies account for roughly 21% of the index, with NVIDIA as the single largest contributor. The company's dominance raises concentration risk for passive index investors — a significant NVIDIA drawdown would visibly drag the entire S&P 500.

Index Weight: ~7-8% | Rank: #1 in the S&P 500 by market capitalization

Company Overview

NVIDIA is executing a platform-level expansion that goes far beyond GPU sales. The Vera Rubin platform, launched at CES 2026 and detailed at GTC 2026, comprises six co-designed chips — the Rubin GPU, Vera CPU, NVLink 6 Switch, ConnectX-9 SuperNIC, BlueField-4 DPU, and Spectrum-6 Ethernet Switch — all tightly integrated into rack-scale NVL72 and NVL144 systems. The Rubin GPU, fabricated on TSMC's 3nm process with 336 billion transistors and HBM4 memory, delivers 50 PFLOPS of FP4 inference — a 5x leap over Blackwell. AWS, Google Cloud, Microsoft Azure, and OCI will be the first to deploy Vera Rubin instances in H2 2026. The $20 billion Groq licensing deal, closed in December 2025, added SRAM-based LPU inference technology directly into the Rubin stack, with the Groq 3 LPX rack delivering a claimed 35x tokens-per-watt improvement over Blackwell alone for trillion-parameter models. The company simultaneously took a $5 billion stake in Intel, securing custom x86 CPUs with NVLink integration for its DGX Rubin NVL8 systems — a hedge that ensures x86 continuity for enterprise data center workflows. NVIDIA's networking division, now generating $11B+ in annual revenue after 263% YoY growth in Q4, has made NVIDIA the world's largest networking company by AI infrastructure revenue. The CUDA software moat, Dynamo inference orchestration engine, NIM microservices, and Nemotron model family together form a full-stack lock-in that competitors like AMD and Google TPU struggle to match at the platform level. Jensen Huang's vision at GTC 2026 was blunt: $1 trillion in cumulative orders for Blackwell and Vera Rubin through 2027, with agentic AI driving compute demand 100x beyond what was projected a year ago. The company is not just selling chips — it is selling AI factories as vertically integrated products, from silicon to software to systems.

Products & Revenue

NVIDIA's revenue is overwhelmingly driven by Data Center, which encompasses GPU accelerators (Blackwell, Hopper), networking (NVLink, Spectrum-X, InfiniBand), DPUs (BlueField), and enterprise AI software (CUDA, NIM, AI Enterprise). Hyperscalers account for just over 50% of Data Center revenue, with the remaining split among AI model makers, enterprises, sovereign AI projects, and cloud partners. Gaming remains the second-largest segment but is increasingly supply-constrained by DRAM shortages. Professional Visualization saw explosive 154% YoY growth in Q4, driven by DGX Spark. Automotive continues steady growth on DRIVE Hyperion autonomous vehicle platforms.

Data Center (89.7%): GPU accelerators (Blackwell/Hopper), NVLink networking, Spectrum-X Ethernet switches, BlueField DPUs, and AI Enterprise software. Includes rack-scale systems like NVL72 and emerging Groq LPX inference racks.

Gaming (7.4%): GeForce RTX 50 Series GPUs, DLSS 4.5, GeForce NOW cloud gaming. Revenue of $16B for FY2026, up 41% YoY but facing supply headwinds from DRAM shortages.

Professional Visualization (1.5%): RTX workstation GPUs, Omniverse platform, DGX Spark desktop AI supercomputer. Surged 154% YoY in Q4 on DGX Spark demand.

Automotive & Robotics (1.1%): DRIVE Hyperion autonomous vehicle platform, AGX Thor SoC, Isaac robotics framework. $2.35B for FY2026, up 39% YoY.

OEM & Other (0.3%): Legacy OEM graphics, crypto mining processors, and miscellaneous revenue. $619M for FY2026.

Based on NVIDIA FY2026 10-K filing (fiscal year ended January 25, 2026). Revenue percentages calculated from full-year segment data.

Leadership

Jensen Huang

CEO since 1993. Co-founded NVIDIA in 1993 and has served as President and CEO since inception. Holds a MSEE from Stanford and a BSEE from Oregon State. Huang invented the GPU category in 1999, drove the CUDA parallel computing pivot, and personally steered NVIDIA's transformation from a gaming graphics company into the dominant AI infrastructure platform — presiding over revenue growth from $11B to $216B in four fiscal years.

Dr. Bill Dally, Chief Scientist & SVP of Research: Former Stanford CS department chair. Leads NVIDIA's long-term research in parallel computing, deep learning architectures, and processor design. His work underpins GPU memory efficiency and the transformer engine innovations in Blackwell and Rubin.

Colette Kress, EVP & Chief Financial Officer: Joined in 2013. Architected NVIDIA's capital allocation strategy through the AI boom, managing $41B in shareholder returns in FY2026 while maintaining a 0.07 debt-to-equity ratio. Key voice on earnings calls navigating the Groq deal accounting and China export uncertainty.

Michael Kagan, Chief Technology Officer: Former Mellanox CEO who joined via NVIDIA's $7B Mellanox acquisition in 2020. Leads networking technology strategy including NVLink, Spectrum-X Ethernet, and InfiniBand — the division that grew 263% YoY in Q4 FY2026.

Ian Buck, VP of AI and HPC: Original author of CUDA's precursor Brook GPU computing language at Stanford. Leads the integration of Groq LPU technology into the Rubin platform and oversees AI factory systems architecture.

Debora Shoquist, EVP of Operations: Manages NVIDIA's global supply chain and manufacturing, including the critical TSMC relationship for CoWoS packaging. Led the expansion of Blackwell production to TSMC's Arizona fabs and Foxconn's Mexico assembly plants.

The AI Angle

The operating system of AI factories

NVIDIA does not have an AI strategy — NVIDIA is the AI strategy. The company sits at the intersection of every major AI trend: training, inference, agentic AI, physical AI, and autonomous systems. Its hardware stack now spans from the Rubin GPU (50 PFLOPS FP4 inference) down to the Vera CPU (88 custom Olympus Arm cores for agentic reasoning) to the Groq 3 LPU (SRAM-based decode co-processor for low-latency token generation). The GTC 2026 keynote made the full-stack ambition explicit: NVIDIA sells complete AI factories — the DGX SuperPOD, the networking fabric, the storage architecture, and the software to orchestrate it all. The Groq 3 LPX rack, paired with Vera Rubin NVL72, claims 35x inference throughput per megawatt over Blackwell alone using Attention-FFN Disaggregation to split prefill (GPU) from decode (LPU) workloads. On the software side, NVIDIA Dynamo is an open-source inference orchestration engine designed to maximize token revenue generation across thousands of GPUs. NIM microservices package optimized model inference into containerized endpoints. The Nemotron 3 model family — including reasoning, speech, and vision models — ships as open-source foundations for enterprise agent development. NVIDIA ACE brings AI-driven NPCs to games, while the Cosmos world model and Isaac GR00T frameworks target physical AI and humanoid robotics. The CUDA ecosystem, with 6 million+ developers, remains the deepest moat: A100 systems are still fully utilized six years after shipping because CUDA backward compatibility extends hardware useful life. NVIDIA's AI research team, led by Chief Scientist Bill Dally, continues to produce architecturally significant work. The third-generation Transformer Engine in Rubin includes hardware-accelerated adaptive compression — a first in production silicon. The company also launched NVIDIA Ising, open-source quantum AI models for quantum processor design, signaling early positioning in quantum-classical hybrid computing. The NVQLink architecture enables tight coupling between GPU compute and quantum processors, adopted by over a dozen supercomputing centers. The risks are real but bounded. AMD's MI455X with HBM4 is competitive on paper. Google's TPU v6 continues to improve. Hyperscalers building custom silicon (AWS Trainium, Google TPU, Microsoft Maia) could reduce NVIDIA dependency over time. The Groq deal faces FTC scrutiny over whether it constitutes a de facto acquisition. China export restrictions have made H200 sales negligible. But NVIDIA's annual architecture cadence — Blackwell (2024) → Rubin (2026) → Feynman (2028) — combined with software lock-in through CUDA, creates a compounding advantage that no single competitor can replicate across the full stack.

Financial Snapshot

Revenue (TTM): $215.9B — FY2026 (ended January 25, 2026) | Net Income: $120.1B net income

Margins: Gross 75.0% (Q4), 71.1% (FY2026 GAAP), operating ~60%, net 55.6%

NVIDIA generated $97 billion in free cash flow for FY2026 and returned $41.1 billion to shareholders through buybacks and dividends, with $58.5 billion remaining under its repurchase authorization. The company is guiding Q1 FY2027 to $78 billion in revenue — implying a $300B+ annual run rate — with GAAP gross margins expected at 74.9%. Management has committed to holding margins in the mid-70s as Rubin ramps, a signal that pricing power remains intact despite the Blackwell-to-Rubin transition and intensifying competition from AMD.

1-Year Performance

Current price $198.87. The stock has risen roughly 72% over the trailing twelve months, though it sits about 6% below its 52-week high of $212.19.

NVIDIA's stock surged through mid-2025 on Blackwell demand and AI infrastructure spending commitments from hyperscalers. A macro-driven correction in early 2026 — driven by the Iran conflict, oil price shocks, and broader tech sector rotation into defensives — pulled the stock from $207 to as low as $165 before a sharp 10-day rally into mid-April recovered most losses. The Information Technology sector is down 6% YTD in 2026, but NVIDIA has outperformed at roughly +3.5% YTD, buoyed by the GTC 2026 product cycle and Q4 earnings beat.

Recent News

Fun Fact: NVIDIA's $20 billion Groq deal was structured as a 'non-exclusive licensing agreement' rather than an acquisition — a deliberate legal architecture designed to avoid the multi-year antitrust review that killed NVIDIA's $40B Arm bid in 2022. Groq founder Jonathan Ross (creator of Google's original TPU) and key engineers joined NVIDIA, Groq shareholders received cash as if it were an acquisition, and unvested equity converted to NVIDIA RSUs — yet on paper, Groq remains an independent company under a new CEO. The FTC is currently investigating whether this structure constitutes a de facto acquisition, making it the most scrutinized 'non-deal deal' in semiconductor history.