Follow us on:
Corporate Case Study | June 2026
Three engineers met at a Denny’s restaurant near San Jose, California. They had $40,000 between them. They believed the personal computer needed dedicated hardware for 3D graphics. On April 5, 1993, they incorporated Nvidia Corporation with that conviction. Thirty-three years later, Nvidia is the most valuable publicly traded company in the world, with a market capitalisation above $3.4 trillion. In fiscal year 2026, it generated $215.9 billion in revenue and $117.0 billion in net income. In Q1 FY27, announced May 20, 2026, it generated $81.6 billion in a single quarter, more than most chip companies generate in a year, and guided Q2 FY27 revenue at $91 billion. In June 2026, it entered the personal computer chip market for the first time with the RTX Spark Superchip alongside Microsoft, and confirmed its next-generation Vera Rubin platform has entered full production. This is the full story of how a gaming chip company became the infrastructure of the artificial intelligence economy, updated with the latest developments as of June 2026.
$215.9B
Nvidia FY26 revenue (fiscal year ending January 25, 2026), up 65% from $130.5 billion in FY25. Data center alone: $193.7 billion. Source: Nvidia 8-K filed February 26, 2026.
$117.0B
FY26 GAAP net income $117.0 billion, up 57.5% from $74.3 billion in FY25. GAAP diluted EPS: $4.90. Non-GAAP diluted EPS: $4.77. Non-GAAP net income: $113.3 billion. GAAP operating margin: 60.4%.
$81.6B
Q1 FY27 revenue (quarter ending April 26, 2026), announced May 20, 2026. Up 85% year-on-year from $44.1 billion in Q1 FY26. Q2 FY27 guidance: $91 billion (plus or minus 2%).
91.5%
Data Center segment share of total Q4 FY26 revenue. Data center revenue was $62.3 billion in Q4 FY26 alone, up 75% year-on-year. Compute: $51.3B. Networking: $11.0B.
Table of Contents
- Part I: The Founding April 5, 1993, the Denny’s booth, $40,000 in starting capital, Jensen Huang’s background, and what the three founders actually believed
- Part II: The Early Years and Near-Death The NV1 failure in 1995, the near-bankruptcy, the RIVA 128 rescue in 1997, the GeForce 256 in 1999, and the IPO on NASDAQ
- Part III: Inventing the GPU and the Gaming Era The GeForce 256 as the world’s first GPU, the 3dfx acquisition, the Xbox contract, and dominating PC gaming through the 2000s
- Part IV: CUDA and the Pivot That Changed Everything 2006: CUDA launches and opens the GPU to science. 2012: AlexNet proves neural networks. The decade-long bet that nobody understood until it paid off.
- Part V: The AI Explosion ChatGPT launches November 2022. Every hyperscaler wants H100s. Nvidia becomes the most valuable company in the world. The Blackwell architecture.
- Part VI: The Financials Annual results FY22 to FY26, quarterly breakdown for FY26, segment breakdown, gross margins, EPS, and the Q1 FY27 record
- Part VII: Business Segments Data Center, Gaming, Professional Visualization, Automotive, and OEM. Revenue, growth rates, and strategic importance of each.
- Part VIII: Risks and the China Headwind US export controls on advanced chips, $4.6 billion in H20 revenue lost in Q1 FY27, competition from AMD and Intel, and the customer concentration risk
- Part IX: Latest Developments, June 2026 Vera Rubin enters full production, the $200 billion agentic AI TAM, and Nvidia’s first-ever entry into the PC chip market with RTX Spark
- Frequently Asked Questions
Part IThe Founding
Three Engineers, $40,000, and a Denny’s Booth
Nvidia Corporation was officially incorporated on April 5, 1993, in Sunnyvale, California. The three founders were Jensen Huang, Chris Malachowsky, and Curtis Priem. The commonly cited story of their founding meetings happening at a Denny’s restaurant near San Jose is accurate. They met regularly at the diner before incorporating the company.
The founding capital was $40,000. Sequoia Capital and Sutter Hill Ventures subsequently invested approximately $20 million in the company’s early venture round, with Mark Stevens of Sequoia joining the board in 1993. Without this early institutional capital, Nvidia would not have survived its first product failure.
The founding conviction was specific and technically grounded. CPUs, the general-purpose processors that ran personal computers, were poorly designed for the massively parallel calculations required to render real-time 3D graphics. Every pixel on a 3D screen requires simultaneous arithmetic on thousands of triangles, textures, and lighting effects. A CPU handles tasks sequentially. What 3D graphics needed was a chip that could run thousands of calculations simultaneously. The three founders believed that dedicated parallel-processing silicon was the future of personal computing, and that whoever built it first would own a large market.
Jensen Huang: The CEO Who Never Left
Jensen Huang was born as Jen-Hsun Huang on February 17, 1963, in Tainan, Taiwan. When he was five, his family moved to Thailand. At age nine, his parents sent him and his older brother to the United States, where a family miscommunication placed them at Oneida Baptist Institute, a Christian boarding school in rural Kentucky that served troubled youth. The experience of being a small Taiwanese child in a reform-school environment shaped his resilience. He later described cleaning dormitory toilets as one of his assigned responsibilities.
After reuniting with his family in Oregon, Huang excelled academically and graduated with a degree in electrical engineering from Oregon State University, where he met his future wife Lori Mills. He worked at AMD and then at LSI Logic, where he rose to lead a division with $250 million in revenue. While at LSI Logic, he completed a master’s degree in electrical engineering from Stanford University, finishing in 1992. In 1993, at the age of 30, he left LSI Logic to co-found Nvidia. He has been its President and CEO since founding, a tenure of 33 years and counting.
Lisa Su, the current CEO of AMD and Nvidia’s most prominent semiconductor competitor, is Jensen Huang’s cousin. This connection between the CEOs of the two leading GPU companies is not widely known.
Jensen Huang
BornFebruary 17, 1963, Tainan, Taiwan
EducationBS Electrical Engineering, Oregon State; MS Electrical Engineering, Stanford (1992)
Prior employersAMD, then LSI Logic (led $250M revenue division)
Role at NvidiaCo-Founder, President and CEO (1993 to present)
Notable relationCousin of Lisa Su, President and CEO of AMD
Chris Malachowsky and Curtis Priem
Malachowsky bornMay 2, 1959
Malachowsky educationBS University of Florida; MS Santa Clara University
Malachowsky prior roleEngineer at Hewlett-Packard, then Sun Microsystems
Malachowsky current roleSenior Vice President and Fellow, Nvidia (still active)
Priem prior roleEngineer at IBM, then Sun Microsystems. Resigned from Sun December 31, 1992, prompting Huang and Malachowsky to also resign and found Nvidia.
Why the company name is Nvidia: The founders initially called their company NVision, but discovered that name was already registered. They switched to NVIDIA. The name combines “NV,” which the founders used as shorthand for “next version” in early file naming, with “invidia,” the Latin word for envy. The company’s official history, published on nvidia.com, states the founding date as April 5, 1993, and records the original vision as bringing 3D graphics to the gaming and multimedia markets.
Part IIThe Early Years and Near-Death
1995: The NV1 and the First Near-Bankruptcy
Nvidia’s first commercial product, the NV1, was released in 1995. It was an ambitious chip that combined 2D and 3D graphics acceleration, audio processing, and game controller ports on a single card. Sega, the Japanese gaming company, had contracted with Nvidia to build graphics hardware for its Saturn gaming console port for PC. The NV1 was the product of that relationship.
The NV1 used a proprietary approach called quadratic texture mapping to render 3D objects. The rest of the industry used triangles as the fundamental unit of 3D geometry. Microsoft had built its DirectX graphics API around triangle-based rendering. Games written for DirectX would not run well on the NV1’s quadratic architecture. When Sega switched to PowerVR technology for its Dreamcast console, Nvidia lost its primary hardware customer. The NV1 was a commercial failure. Nvidia came close to shutting down.
Jensen Huang described this period as the moment that defined Nvidia’s culture. The company scrapped the NV1 architecture entirely, abandoned the NV2 which was already in development, and started over with a DirectX-compatible triangle-based architecture. This was a bet-the-company decision: spend the last of the venture money rebuilding from scratch on a timeline that almost no one believed was achievable.
1997: The RIVA 128 Saves the Company
The RIVA 128, released on August 25, 1997, was the product that rescued Nvidia. RIVA stood for Real-time Interactive Video and Animation. It was Nvidia’s first product built on industry-standard triangle rendering, making it compatible with DirectX and OpenGL. It sold over one million units in its first four months. This was the fastest-selling graphics card in the industry at the time. It generated the revenue that funded Nvidia’s continued operation and the development of its next products.
The RIVA TNT (1998) and RIVA TNT2 (1999) followed, further cementing Nvidia’s position in the consumer graphics market. These products competed directly with 3dfx Interactive’s Voodoo graphics cards, which had defined early 3D gaming. The competition between Nvidia and 3dfx through 1998 and 1999 was fierce. 3dfx’s Voodoo cards were beloved by PC gamers. Nvidia’s engineering pace was faster.
1999: The GeForce 256 and the IPO
On January 22, 1999, Nvidia went public on the NASDAQ exchange at $12 per share. The company listed under the ticker NVDA. In the same year, it launched the GeForce 256 on August 31, 1999. This was the product Nvidia marketed as “the world’s first GPU,” a term it coined for the occasion. The GeForce 256 introduced hardware Transform and Lighting (T&L) processing. T&L calculations had previously been done by the CPU. Offloading them to the graphics chip freed CPU capacity for game logic and other tasks. This was a genuine step-change in how personal computers handled 3D graphics.
Part IIIInventing the GPU and the Gaming Era
Acquiring 3dfx and Building the GeForce Brand
In 2000, Nvidia acquired the key assets of 3dfx Interactive, including its Voodoo graphics card intellectual property and most importantly its engineering talent. 3dfx, which had once led the consumer graphics market, had struggled to compete with Nvidia’s development pace. The acquisition removed Nvidia’s most significant direct competitor and consolidated the gaming GPU market in Nvidia’s favour. Also in 2000, Nvidia was selected by Microsoft to design and supply the GPU for the original Xbox gaming console. This provided a significant revenue anchor and validation of Nvidia’s technical capabilities.
The GeForce brand became the most recognised name in consumer graphics through the 2000s. Each generation of GeForce cards offered substantially more performance than the previous at competitive prices. The company also built out the Quadro brand for professional visualization users: graphic designers, CAD engineers, architects, and film production studios who needed the highest accuracy and reliability in their rendering hardware.
The fabless model: Nvidia operates as a fabless semiconductor company. It designs chips but does not manufacture them. Manufacturing is contracted out, predominantly to TSMC (Taiwan Semiconductor Manufacturing Company) and to a lesser extent Samsung. This model eliminates the enormous capital costs of building and maintaining semiconductor fabrication plants, which cost tens of billions of dollars each. The fabless model allows Nvidia to direct all of its capital toward research and design rather than manufacturing infrastructure. It also means Nvidia is dependent on TSMC’s production capacity and technology roadmap, which became a significant strategic consideration as chip demand surged during the AI boom.
Part IVCUDA and the Pivot That Changed Everything
2006: CUDA Opens the GPU to Science
In 2006, Nvidia introduced CUDA: Compute Unified Device Architecture. CUDA was a parallel computing platform and programming model that allowed developers to write software that ran on Nvidia GPUs for non-graphics purposes. Before CUDA, GPUs were single-purpose hardware. They rendered graphics. That was all. CUDA made the GPU a general-purpose parallel computing engine. Scientists, researchers, and engineers could now write programs in standard programming languages like C that ran on GPU hardware and exploited its thousands of parallel processing cores.
The significance of CUDA was not immediately obvious. Most of the industry in 2006 saw it as a niche product for scientific computing. A small community of researchers began using GPUs for computational fluid dynamics, molecular dynamics simulations, financial modelling, and signal processing. Nvidia built a developer ecosystem around CUDA, offering free development tools, documentation, and academic partnerships. It took six years for the real impact of CUDA to become apparent.
2012: AlexNet and the Proof of Concept
In September 2012, a team at the University of Toronto led by Geoffrey Hinton, with PhD students Alex Krizhevsky and Ilya Sutskever, submitted an entry to the ImageNet Large Scale Visual Recognition Challenge. Their model, which they called AlexNet, used deep convolutional neural networks trained on two Nvidia GTX 580 GPUs using CUDA. AlexNet won the competition by a margin so large that it shattered the assumptions of the computer vision research community. The winning model had a top-5 error rate of 15.3 percent. The second-place entry had 26.2 percent. The gap was not a refinement. It was a discontinuity.
AlexNet demonstrated that deep neural networks trained on GPUs using CUDA could perform tasks that had previously required hand-crafted algorithms, and could do so at a level of accuracy that no prior approach had achieved. The result ignited the deep learning revolution. Researchers across the world realised that GPU-accelerated neural network training was the tool they had been waiting for. Nvidia had built that tool eight years earlier without knowing exactly what it would enable.
The CUDA moat: CUDA is not just software. It is an ecosystem. By 2024, CUDA had been available for 18 years. An entire generation of AI researchers, machine learning engineers, and data scientists had learned to code in CUDA. Every major deep learning framework, including PyTorch and TensorFlow, was built with CUDA as the primary hardware acceleration target. Libraries like cuDNN, cuBLAS, and NCCL, all Nvidia-developed and CUDA-native, were deeply embedded in the AI research pipeline. Switching from Nvidia to a competitor GPU is not a hardware change. It is a complete rewrite of every library, tool, workflow, and optimisation that the AI community has developed over 18 years. This is why AMD’s MI300 chips and Intel’s Gaudi accelerators, despite competitive hardware specifications, have struggled to take meaningful market share from Nvidia in AI training workloads.
-
1993April 5, 1993Nvidia Incorporated in Sunnyvale, California
Jensen Huang, Chris Malachowsky, and Curtis Priem incorporate Nvidia with $40,000 in starting capital. Sequoia Capital’s Mark Stevens joins the board after an early $20 million venture round. The goal: build dedicated 3D graphics silicon for personal computers.
-
19951995NV1 Released and Fails Commercially
Nvidia’s first product, the NV1, uses quadratic texture mapping that is incompatible with Microsoft’s DirectX standard. Sega abandons the platform. Nvidia nearly goes bankrupt. Jensen Huang scraps the architecture entirely and begins again with DirectX-compatible triangle rendering.
-
1997August 25, 1997RIVA 128 Sells 1 Million Units in Four Months
The RIVA 128 saves Nvidia. Built on DirectX-compatible triangle architecture, it sells over one million units in its first four months. Revenue from the RIVA 128 funds Nvidia’s continued operation and the development of the GeForce line.
-
1999January 22, 1999Nvidia IPO on NASDAQ at $12 per Share
Nvidia lists on NASDAQ under the ticker NVDA at $12 per share. Later the same year, on August 31, 1999, Nvidia launches the GeForce 256, which it markets as the world’s first GPU. Hardware T&L offloads calculations from the CPU to the GPU for the first time.
-
200020003dfx Acquired; Xbox GPU Contract Won
Nvidia acquires the assets of rival 3dfx Interactive, eliminating its primary consumer graphics competitor. Nvidia is also selected by Microsoft to supply the GPU for the original Xbox console, providing significant volume and revenue validation.
-
20062006CUDA Launched: The GPU Becomes a General Computing Platform
Nvidia releases CUDA (Compute Unified Device Architecture), a parallel computing platform that opens the GPU to non-graphics workloads. Scientists and researchers begin using CUDA-enabled GPUs for simulations, financial modelling, and early machine learning experiments.
-
2012September 2012AlexNet: The Proof That Started the AI Era
AlexNet, trained on two Nvidia GTX 580 GPUs using CUDA, wins the ImageNet competition with a 15.3% error rate versus 26.2% for the second-place entry. The gap proves that GPU-accelerated deep learning is a step change. The AI research community pivots to GPU computing almost overnight.
-
2020September 2020$40 Billion Arm Acquisition Announced (Later Blocked)
Nvidia announces a deal to acquire UK-based chip designer Arm Holdings from SoftBank for $40 billion. Regulators in the US, UK, EU, and China block the deal on competition grounds. It is terminated in February 2022 with no penalty. Arm subsequently IPOs on NASDAQ in September 2023.
-
2020April 2020Mellanox Acquired for $6.9 Billion
Nvidia completes the acquisition of Mellanox Technologies for $6.9 billion. Mellanox builds InfiniBand and Ethernet high-speed networking technology that connects GPUs in data centers. This acquisition gives Nvidia control over the networking layer of AI infrastructure, not just the compute layer.
-
2022November 30, 2022ChatGPT Launches: Everything Changes for Nvidia
OpenAI releases ChatGPT. The product reaches 100 million users in two months. Every major technology company announces aggressive AI investment. All of them need H100 GPUs. Nvidia’s order backlog extends to over a year. The company enters a revenue growth trajectory unlike anything in semiconductor history.
-
2026May 20, 2026Q1 FY27: $81.6 Billion in a Single Quarter
Nvidia announces Q1 FY27 (April 26, 2026) revenue of $81.6 billion, up 85% year-on-year. GAAP net income of $58.3 billion, up 211% year-on-year. Data center revenue: $75.2 billion, up 92%. Q2 FY27 guidance: $91 billion. Quarterly dividend raised from $0.01 to $0.25 per share (a 25-fold increase). New $80 billion share buyback authorisation approved.
Part VThe AI Explosion
From ChatGPT to $215.9 Billion in Revenue in Three Years
When OpenAI launched ChatGPT on November 30, 2022, Nvidia was already the dominant provider of AI training hardware. But ChatGPT changed the scale of demand. Before ChatGPT, AI investment was concentrated at a handful of research laboratories and large technology companies. After ChatGPT, every enterprise, every government, every startup, and every university wanted to build AI capabilities. All of them needed GPUs. The vast majority of AI training workloads ran on Nvidia GPUs using CUDA. The result was a demand surge that Nvidia’s supply chain could not immediately satisfy.
The H100 GPU, built on Nvidia’s Hopper architecture and released in 2022, became the most sought-after piece of hardware in technology history. Companies placed orders months or years in advance. The H100 was priced at approximately $25,000 to $35,000 per unit. Large hyperscalers purchased clusters of tens of thousands of GPUs at a time. Microsoft, Amazon, Google, and Meta collectively committed hundreds of billions of dollars to AI infrastructure, with Nvidia GPUs as the primary component.
Blackwell: The Architecture That Drove FY26
Nvidia’s Blackwell architecture, which succeeded Hopper, began shipping in volume during FY26. The Blackwell platform includes the GB200 and B200 GPUs and the NVLink Switch fabric that connects multiple GPUs into coherent supercomputing clusters. The GB300, the next iteration of Blackwell (known as Blackwell Ultra), became the leading architecture across all customer categories in Q4 FY26. By Q4 FY26, Blackwell GPU revenue had exceeded Hopper GPU revenue. The transition from Hopper to Blackwell contributed to a temporary gross margin compression in FY26 relative to FY25, as Blackwell’s full rack-scale solutions had higher content cost than standalone H100 units. FY26 GAAP gross margin was 71.1 percent, down from 75.5 percent in FY25.
China and export controls: The US government has progressively tightened export restrictions on advanced semiconductor products to China since October 2022. Nvidia developed the H20 GPU as a product compliant with export control rules for the Chinese market. In Q4 FY25, Nvidia recognised a $5.5 billion inventory charge related to H20 chips following new export restrictions announced in April 2025. In Q1 FY27 (quarter ending April 26, 2026), no Data Center compute revenue was recognised from China, compared to $4.6 billion in Q1 FY26. The company’s Q2 FY27 guidance of $91 billion explicitly assumes no Data Center compute revenue from China. This represents a material and permanent revenue headwind from one of the world’s largest markets for AI infrastructure.
Part VIThe Financials
Annual Financial Performance: FY22 to FY26
Nvidia’s fiscal year ends in late January. FY26 ended January 25, 2026. FY25 ended January 26, 2025. All figures are from Nvidia’s official SEC filings (8-K and 10-K forms).
| Fiscal Year | Revenue ($B) | GAAP Net Income ($B) | GAAP EPS (Diluted) | GAAP Gross Margin | Key Development |
|---|---|---|---|---|---|
| FY22 (ended Jan 2022) | 26.9 | 9.8 | $0.39 | 64.9% | Gaming peak; data center growing; Arm deal announced but blocked |
| FY23 (ended Jan 2023) | 26.9 | 4.4 | $0.18 | 56.9% | Gaming downturn; cryptocurrency collapse; inventory write-downs; H100 launches; ChatGPT launches end of FY23 |
| FY24 (ended Jan 2024) | 60.9 | 29.8 | $1.19 | 72.7% | AI demand explosion post-ChatGPT; H100 order backlog extends to 12+ months; first $20B revenue quarter |
| FY25 (ended Jan 2025) | 130.5 | 74.3 | $2.99 | 75.5% | Blackwell architecture launched; data center 88% of revenue; 114% YoY revenue growth |
| FY26 (ended Jan 2026) | 215.9 | 117.0 | $4.90 (GAAP) | 71.1% | 65% YoY growth; Blackwell Ultra dominates; data center 89.7% of revenue; gross margin compressed by Blackwell transition costs |
Annual Revenue: FY22 to FY26 (USD Billion) | Source: Nvidia SEC 10-K Filings
Note: FY22 and FY23 both registered $26.9 billion, reflecting the gaming downturn and pre-AI-boom flat growth. The 126% jump in FY24 reflects the ChatGPT-driven AI infrastructure demand surge. FY26 Q1 revenue alone ($81.6B) exceeded the entire FY22 and FY23 annual revenue.
Quarterly Results: FY26 and Q1 FY27
| Quarter | Revenue ($B) | GAAP Net Income ($B) | GAAP EPS (Diluted) | Data Center Revenue ($B) | GAAP Gross Margin |
|---|---|---|---|---|---|
| Q1 FY26 (ended Apr 2025) | 44.1 | 18.8 | $0.76 | 39.1 | 60.5% |
| Q2 FY26 (ended Jul 2025) | 45.3 | 16.6 | $0.67 | 40.2 | 70.1% |
| Q3 FY26 (ended Oct 2025) | 57.0 | 19.3 | $0.78 | 51.2 | 73.4% |
| Q4 FY26 (ended Jan 2026) | 68.1 | 42.96 | $1.76 | 62.3 | 75.0% |
| Q1 FY27 (ended Apr 2026) | 81.6 | 58.3 | $2.39 | 75.2 | 74.9% |
Reading the Q1 FY26 gross margin anomaly: GAAP gross margin of 60.5% in Q1 FY26 was significantly lower than the 75.5% achieved in FY25. This compression was caused by inventory charges related to Hopper products and early Blackwell supply preparation costs. The company recognised charges related to supply commitments for products that had not yet ramped in volume. By Q4 FY26, as Blackwell achieved full production ramp, the margin recovered to 75.0%. The Q1 FY26 compression was a transition cost, not a structural deterioration.
Key Balance Sheet and Return Metrics (FY26)
Total Assets (FY26)
$206.8 Billion
As of January 25, 2026. Up from $85.2 billion at the end of FY25.
Total Equity (FY26)
$144.3 Billion
As of January 25, 2026. Reflects retained earnings and share buyback programme.
Cash Returned to Shareholders
$41.1 Billion
Total capital returned in FY26 through share buybacks and dividends. Q1 FY27: $20 billion in a single quarter.
GAAP Operating Margin (FY26)
62.7%
GAAP operating income of $130.4 billion on revenue of $215.9 billion. A level of profitability rarely seen in any industry at this scale.
Employees (FY26)
42,000
As of end of FY26. Nvidia generates over $5 million in revenue per employee annually, among the highest ratios of any large company in the world.
Q1 FY27 Free Cash Flow
$48.6 Billion
GAAP free cash flow in Q1 FY27 alone (quarter ended April 26, 2026). Nearly half returned to shareholders in the same quarter.
Part VIIBusiness Segments
Five Segments, One Dominant Revenue Driver
Data Center
FY26 Revenue: $193.7B (89.7% of total) | Q4 FY26: $62.3B (+75% YoY)
Nvidia’s primary revenue engine. Includes H100, H200, and Blackwell GPUs for AI training and inference; Grace CPU for data center compute; NVLink Switch for GPU interconnect; Spectrum-X and InfiniBand networking from the Mellanox acquisition. In Q4 FY26, compute was $51.3 billion and networking was $11.0 billion. Customers include Microsoft Azure, Amazon AWS, Google Cloud, Meta, and thousands of enterprises and AI startups. Blackwell Ultra is the current leading platform as of Q4 FY26. In Q1 FY27, data center revenue reached $75.2 billion, up 92% year-on-year.
Gaming
FY26 Revenue: $16.042B (up 41% YoY) | Q4 FY26: $3.727B (+47% YoY, -13% QoQ)
GeForce GPU cards for consumer PC gaming. Nvidia’s original business and still the largest consumer electronics market for its products. The 47% year-on-year growth in Q4 FY26 was driven by Blackwell-generation GeForce RTX 50 series products. The -14% sequential decline reflected normalisation of channel inventories after the RTX 50 series launch. Gaming is now approximately 7.4% of Nvidia’s total revenue, down from over 40% in FY22. It remains important for maintaining the consumer GPU ecosystem that feeds into data center demand for CUDA-compatible hardware.
Professional Visualization
FY26 Revenue: $3.191B (up 70% YoY) | Q4 FY26: $1.321B (+74% QoQ, +47% YoY)
Quadro and RTX professional GPU products for architects, engineers, designers, and visual effects studios. Q4 FY26 saw 154% year-on-year growth, driven by the launch of DGX Spark (a desktop AI workstation) and Blackwell-generation professional graphics products. DGX Spark allows AI researchers and developers to run large language models on a desktop machine using Blackwell GPU hardware.
Automotive
FY26 Revenue: $2.386B (up 39% YoY) | Q4 FY26: $570M (+6% YoY)
Nvidia DRIVE platform for autonomous vehicles and advanced driver assistance systems (ADAS). Partners include Mercedes-Benz, BYD, Toyota, Volvo, and other major OEMs. Growth in this segment is driven by the adoption of Nvidia’s DRIVE Orin and DRIVE Thor systems-on-chip in production vehicles. This segment is expected to grow significantly as robotaxi and fully autonomous vehicle programmes accelerate, with Nvidia projecting multi-billion dollar automotive revenues in FY27 and beyond.
OEM and Other
FY26 Revenue: $0.619B (up 59% YoY) | Small but growing from crypto mining trough
Cryptocurrency mining processors (CMPs) and other OEM products. This segment has declined significantly as cryptocurrency mining demand collapsed. It represents a minimal share of Nvidia’s current revenue and is not strategically significant.
Part VIIIRisks and the China Headwind
The Key Risks Every Investor Must Understand
The US government has progressively restricted the export of advanced AI chips to China since October 2022. Nvidia developed the A800, H800, and subsequently the H20 as products designed to comply with the export control rules while still serving Chinese customers. In April 2025, the US government announced that H20 exports to China would also require licences. Nvidia recognised a $5.5 billion inventory charge in Q1 FY26 related to H20 chips that could no longer be sold as planned.
In Q1 FY27 (quarter ending April 26, 2026), Nvidia recognised zero Data Center compute revenue from China, compared to $4.6 billion in Q1 FY26. Nvidia’s Q2 FY27 guidance of $91 billion explicitly assumes no Data Center compute revenue from China. China had historically represented a meaningful portion of Nvidia’s data center revenue. The loss of this market is permanent unless US export control policy changes. Nvidia has stated it will continue to engage with US government officials on export control policy but has provided no guarantee of future China revenue.
AMD’s MI300X and MI325X chips have achieved adoption at some hyperscalers, particularly for AI inference workloads where CUDA compatibility matters less than for training. AMD’s advantage is that it can offer competitive performance at a potentially lower cost for specific workloads. Intel’s Gaudi accelerators have gained limited traction in data centers. Neither AMD nor Intel has been able to replicate the CUDA ecosystem advantage that keeps most AI training workloads on Nvidia hardware.
The more significant competitive threat is custom silicon developed by Nvidia’s largest customers. Google’s TPUs (Tensor Processing Units) have been used internally for years and are available through Google Cloud. Amazon’s Trainium and Inferentia chips are deployed across AWS. Microsoft has developed the Maia chip for Azure. Meta has developed MTIA for its own inference workloads. If hyperscalers shift a meaningful portion of their AI compute to custom silicon, Nvidia’s revenue growth could slow significantly. These customers currently account for a large proportion of Nvidia’s data center revenue. The risk is not immediate. Custom silicon development is expensive and time-consuming. But it is a structural competitive threat over a three-to-five-year horizon.
Nvidia designs its chips but does not manufacture them. TSMC manufactures the vast majority of Nvidia’s GPU products using its most advanced process nodes. This creates a significant concentration risk: Nvidia is dependent on a single manufacturer in Taiwan for the production of hardware that generates over $190 billion in annual revenue. Any disruption to TSMC’s operations, whether from geopolitical events related to Taiwan’s status, natural disasters, or manufacturing problems, would directly impair Nvidia’s ability to supply its products.
Nvidia has taken some steps to diversify, including working with Samsung on some products and supporting the US CHIPS Act initiatives to expand domestic semiconductor manufacturing. The new TSMC fab in Arizona is partly a response to this risk. But the advanced node processes that Nvidia’s highest-performance GPUs require are not available at any other manufacturer at scale. This dependency is a structural feature of the fabless semiconductor model that cannot be quickly changed.
A handful of hyperscalers account for a very large proportion of Nvidia’s data center revenue. Microsoft, Amazon, Google, and Meta are all among Nvidia’s largest customers. In any given quarter, it is estimated that these four companies collectively account for a majority of data center GPU purchases. This concentration means that any moderation in their AI capital expenditure programs could have an outsized impact on Nvidia’s revenue. If cloud providers pause or slow their data center GPU purchases to integrate and utilise existing infrastructure rather than adding new capacity, Nvidia’s sequential revenue growth could slow or reverse in a given quarter.
Nvidia’s response to this risk is to expand beyond hyperscalers into enterprise customers, sovereign AI programs (governments building national AI infrastructure), and the automotive sector. Enterprise and sovereign customers tend to have smaller but more diversified and potentially more stable purchasing patterns than hyperscalers. Jensen Huang has described the expansion of the enterprise customer base as a strategic priority for the FY27 and FY28 period.
Part IXLatest Developments: June 2026
Vera Rubin Enters Full Production
On the Q1 FY27 earnings call held May 20, 2026, Jensen Huang announced that Nvidia’s next-generation Vera Rubin platform had entered full production. Vera Rubin succeeds the Blackwell architecture and is built around a new central processing unit called Vera, designed specifically for agentic AI workloads, paired with the Rubin GPU. Huang described Vera Rubin as the most ambitious engineering effort in Nvidia’s history. He stated that the supply chain built for Vera Rubin is twice the size of the supply chain built for Grace Blackwell, and that assembly time for a single AI rack has fallen from two hours to five minutes.
Huang also disclosed that the Vera CPU and the broader Vera Rubin platform opens a new total addressable market of approximately $200 billion, distinct from the GPU compute market that Blackwell and Hopper served. This figure refers to the addressable market for CPU based agentic AI infrastructure, a category Nvidia had not previously competed in at scale. Vera Rubin is expected to be the primary growth driver for Nvidia’s data center segment through FY27 and into FY28.
Nvidia Enters the PC Chip Market: RTX Spark and the Microsoft Partnership
At the Computex conference in Taipei on June 1, 2026, Jensen Huang announced Nvidia’s entry into the personal computer chip market for the first time, a segment it had never directly competed in before. The centrepiece of this announcement was the RTX Spark Superchip, a new processor designed for Windows based laptops, desktops, and workstations built around on device AI capability.
The announcement was made jointly with Microsoft. Huang stated that Nvidia and Microsoft are working together to reinvent the PC, with new Windows based systems built specifically to run agentic AI experiences directly on the device rather than relying entirely on cloud based AI compute. Huang described this as the first major architectural reinvention of the personal computer in 40 years. He compared the moment to the transition from feature phones to smartphones, suggesting that PCs with embedded AI supercomputing capability could eventually become as central to daily life as a smartphone.
The market reaction was immediate. Shares of AMD, Intel, and Qualcomm, the three companies that have historically dominated PC processor supply, fell on the day of the announcement as investors recognised that Nvidia was extending its competitive reach beyond the data center and into the consumer and enterprise PC market, an area where Nvidia had previously only supplied graphics cards rather than complete system processors.
Why the PC chip move matters strategically: Until June 2026, Nvidia’s AI dominance was almost entirely a data center and cloud phenomenon. The RTX Spark announcement signals an attempt to extend the same AI compute advantage to the edge: laptops, desktops, and workstations that run AI workloads locally without needing a connection to a cloud data center. This is a similar strategic logic to Apple’s adoption of Arm based chips for the iPhone in 2007 and Amazon’s introduction of Arm based Graviton processors for AWS data centers in 2018. If Nvidia can establish RTX Spark as a standard component of next generation Windows PCs through its Microsoft partnership, it would create a second major hardware franchise alongside its data center GPU business.
New Networking and Robotics Announcements
Alongside Vera Rubin and RTX Spark, Huang used the Computex keynote to confirm wider adoption of Nvidia’s Spectrum-X Ethernet networking platform, which the company markets as the world’s first Ethernet fabric purpose built for AI data center traffic. Nvidia confirmed it will release new Spectrum-X networking products annually going forward, reinforcing the strategic importance of the networking layer, originally built around the 2020 Mellanox acquisition, to Nvidia’s overall AI infrastructure stack. Huang also introduced Nvidia Cosmos 3, an open world foundation model for physical AI and robotics applications, built on what Nvidia describes as a mixture of transformers architecture combining vision reasoning, world generation, and action prediction within a single system.
What this means for the financial picture going forward: None of the June 2026 announcements have yet appeared in Nvidia’s reported financial results, since the most recent reported quarter (Q1 FY27, ended April 26, 2026) predates the Computex keynote. Vera Rubin’s full production status and the RTX Spark PC chip launch will first show up in Nvidia’s Q2 FY27 results, expected around late August 2026. The $200 billion TAM figure Huang cited for the Vera CPU and agentic AI infrastructure category is a market opportunity estimate from management, not a revenue figure or company guidance, and should be read as directional commentary on the size of the next growth phase rather than a confirmed financial outcome.
From a Denny’s Booth to the Most Valuable Company in the World
Nvidia’s story is one of the most remarkable in the history of business. It is not a story of overnight success. It is a story of three engineers who identified a correct technical insight in 1993, nearly went bankrupt in 1995, survived through a pivot that took two years, built a gaming chip company for a decade, made a second bet in 2006 that almost no one else understood, and then watched that bet pay off sixteen years later in a manner that no one, including Jensen Huang himself, could have fully predicted.
The numbers are now almost incomprehensible. $215.9 billion in revenue in FY26. $117.0 billion in net income. $81.6 billion in a single quarter in Q1 FY27. A Q2 FY27 guidance of $91 billion with no China data center revenue assumed. A dividend raised 25-fold in a single announcement. Free cash flow of $48.6 billion in one quarter.
These numbers are the result of the CUDA moat. An 18-year head start in building a programming ecosystem that every AI researcher, every deep learning framework, and every data center operator has integrated deeply into their operations. AMD can build a chip with similar specifications. Intel can build a chip. Google, Amazon, and Microsoft can build custom silicon. None of them can replicate 18 years of CUDA ecosystem development in a short timeframe. That is why Nvidia’s pricing power remains extraordinary, and why its gross margins have held at 71 to 75 percent even as revenue has grown from $27 billion to $216 billion in three years.
The risks are real. China export controls have permanently removed a large revenue market. Custom silicon from hyperscalers poses a structural long-term threat. TSMC concentration is a geopolitical vulnerability. Customer concentration at the hyperscaler level creates revenue volatility. And the question of whether AI capital expenditure, which is Nvidia’s primary demand driver, continues to grow at the pace required to sustain 65 to 85 percent annual revenue growth is the central uncertainty for any investor modelling the company’s trajectory.
But the technology foundation, the software ecosystem, and the leadership position that Jensen Huang built over 33 years are not going to dissolve in a year or two. Nvidia is the infrastructure of the AI economy. The AI economy is just beginning to scale. Whether that combination produces another decade of extraordinary returns for shareholders is unknowable. What is certain is that no company in the current era has built a more strategically entrenched position in a more consequential technology market.
Frequently Asked Questions
Nvidia Corporation was officially incorporated on April 5, 1993, in Sunnyvale, California. The three co-founders were Jensen Huang, Chris Malachowsky, and Curtis Priem. Jensen Huang, born on February 17, 1963, in Tainan, Taiwan, had previously worked at AMD and LSI Logic and completed a master’s degree in electrical engineering from Stanford University in 1992. Chris Malachowsky, born on May 2, 1959, had worked at Hewlett-Packard and Sun Microsystems. Curtis Priem had worked at IBM and Sun Microsystems. The three met regularly at a Denny’s restaurant near San Jose, California, to discuss their founding vision: that personal computers would soon need dedicated hardware for 3D graphics and that whoever built it would own a large market. The company was incorporated with $40,000 in starting capital, and Sequoia Capital and Sutter Hill Ventures subsequently provided approximately $20 million in early venture funding, with Sequoia’s Mark Stevens joining the board. The company name combines “NV,” used as shorthand for “next version” in early file naming, with “invidia,” the Latin word for envy. Nvidia’s official corporate timeline, published on nvidia.com, records the founding date as April 5, 1993.
For fiscal year 2026, which ended January 25, 2026, Nvidia reported total revenue of $215.9 billion, up 65 percent from $130.5 billion in FY25. GAAP net income was $117.0 billion, up 57.5 percent from $74.3 billion in FY25. GAAP diluted EPS was $4.90. Non-GAAP diluted EPS was $4.77. GAAP gross margin for FY26 was 71.1 percent, down from 75.5 percent in FY25 due to costs associated with the Blackwell architecture ramp and an inventory charge related to H20 chips in Q1 FY26. GAAP operating income was $130.4 billion, representing a 60.4 percent operating margin. Data center revenue for FY26 was $193.7 billion, representing 89.7 percent of total revenue. Nvidia returned $41.1 billion to shareholders through buybacks and dividends in FY26. Total assets as of January 25, 2026 were $206.8 billion. For Q1 FY27, which ended April 26, 2026 and was announced May 20, 2026, revenue was $81.6 billion, up 85 percent year-on-year from $44.1 billion in Q1 FY26. GAAP net income was $58.3 billion, up 211 percent year-on-year. GAAP diluted EPS was $2.39. Data center revenue was $75.2 billion, up 92 percent year-on-year. Nvidia guided Q2 FY27 revenue at $91 billion plus or minus 2 percent. The company raised its quarterly dividend 25-fold from $0.01 to $0.25 per share, effective June 26, 2026, and authorised a new $80 billion share repurchase programme. All figures are from Nvidia’s SEC filings (Form 8-K and Form 10-Q).
CUDA (Compute Unified Device Architecture) is a parallel computing platform and programming model that Nvidia introduced in 2006. It allows developers to write software in standard programming languages like C that runs on Nvidia GPU hardware, exploiting the thousands of parallel processing cores in a GPU for tasks other than graphics. Before CUDA, GPUs were single-purpose graphics hardware. CUDA made them general-purpose parallel computing engines. The significance of CUDA became apparent in 2012 when AlexNet, a deep neural network trained on two Nvidia GTX 580 GPUs using CUDA, won the ImageNet computer vision competition by a historically large margin, proving that GPU-accelerated deep learning was transformatively more powerful than prior approaches. By 2024, CUDA had been available for 18 years. Every major deep learning framework (PyTorch, TensorFlow) is built with CUDA as the primary hardware target. Nvidia-developed libraries including cuDNN, cuBLAS, and NCCL are deeply embedded in AI research pipelines globally. Competing GPU hardware from AMD and Intel requires developers to use different programming interfaces and optimise their code differently. For most AI training workloads, switching from Nvidia is not a hardware upgrade decision: it requires rewriting years of optimised CUDA code. This dependency creates a structural switching cost that AMD and Intel have not been able to overcome despite offering hardware with competitive specifications. CUDA is Nvidia’s most durable competitive advantage because it is an 18-year legacy of ecosystem development that no competitor can replicate quickly.
Nvidia’s GAAP gross margin fell to 71.1 percent in FY26 from 75.5 percent in FY25. Two factors drove this compression. First, in Q1 FY26 (quarter ending April 2025), Nvidia recognised a $5.5 billion inventory charge related to H20 GPU chips following the US government’s announcement of additional export controls in April 2025 that restricted H20 sales to China. This charge reduced gross margin significantly in Q1 FY26 (GAAP gross margin was 60.5 percent in Q1 FY26). Second, the transition from selling standalone Hopper H100 GPUs (which have a relatively high margin because they are sold individually) to selling full-scale Blackwell rack systems (which include more components and have higher bill-of-materials costs per revenue dollar) temporarily compressed margins during the Blackwell ramp. By Q4 FY26, as Blackwell achieved full production ramp and the inventory charges were behind the company, gross margin recovered to 75.0 percent. In Q1 FY27, GAAP gross margin was 74.9 percent, broadly in line with Q4 FY26. The FY26 full-year margin compression was therefore primarily a transition effect rather than a structural deterioration in Nvidia’s pricing power.
The US government has progressively restricted Nvidia’s ability to sell advanced AI chips to Chinese customers since October 2022. Nvidia developed H20, A800, and H800 chips designed to comply with the performance thresholds set by US export control regulations. In April 2025, the US government announced that H20 exports would also require licences, effectively closing the Chinese market to Nvidia’s data center products. Nvidia recognised a $5.5 billion inventory charge in Q1 FY26 related to H20 chips that could no longer be sold as planned. In Q1 FY27 (quarter ending April 26, 2026), Nvidia reported zero Data Center compute revenue from China, compared to $4.6 billion in the same quarter a year earlier. Nvidia’s Q2 FY27 guidance of $91 billion explicitly assumes no Data Center compute revenue from China. China had represented a significant portion of Nvidia’s data center revenue, and the loss of this market is treated by management as permanent under current US export control policy. Nvidia has stated that it will engage with US regulators on export control policy but has made no representations that China revenue will recover. The export control restrictions represent a permanent structural headwind to Nvidia’s addressable market.
Disclaimer: This article is for informational and educational purposes only and is current as of June 21, 2026. All financial figures are sourced directly from Nvidia Corporation’s official SEC filings: the Form 8-K filed February 26, 2026 (Q4 FY26 and FY26 annual results), the Form 8-K filed May 20, 2026 (Q1 FY27 results), the Form 10-Q filed for Q1 FY27, and Nvidia’s official corporate timeline published at nvidia.com/en-us/about-nvidia/corporate-timeline/. Historical product and company history facts are sourced from Nvidia’s official corporate history. Founding details sourced from Nvidia’s official history page and Sequoia Capital’s published account. Details on Vera Rubin and the RTX Spark PC chip launch are sourced from Nvidia’s Q1 FY27 earnings call (May 20, 2026) and the Computex Taipei keynote (June 1, 2026). All figures are GAAP unless explicitly noted as non-GAAP. This article does not constitute investment advice or a recommendation to buy or sell any security. fiscalzenith.com accepts no liability for decisions made in reliance on information in this article.
