NVIDIA History Fact Check: From Xbox GPUs and CUDA to AI Data Center Dominance

Below is the story checked against NVIDIA's own records, its SEC filings, regulator statements and market research. Where a claim couldn't be verified, it's labeled that way.

The founding: three engineers and a bet on 3D graphics​

NVIDIA's corporate timeline says the company was founded on April 5, 1993, by Jensen Huang, Chris Malachowsky and Curtis Priem. The goal was to bring 3D graphics to gaming and multimedia. Huang has been CEO since day one, which is rare in Silicon Valley.

Britannica adds some background:

• Huang had been a director at LSI Logic and a microprocessor designer at AMD.
• Malachowsky was an engineer at Sun Microsystems.
• Priem was a senior staff engineer and graphics chip designer at IBM and Sun.
• The name combines invidia, Latin for "envy," with "NV" for "next vision," a label the founders used on early files.
• Funding: $20 million in first-year venture money from firms including Sequoia Capital, followed by an IPO in 1999.

The mix matters. Two of the three came from workstation-class companies, where hardware was built for heavy computing. The consumer PC gaming market they targeted cared about speed above almost everything else. That combination explains much of what came next.

Section summary: NVIDIA began as a narrowly focused PC graphics startup. Its founders came from workstation and chip-design backgrounds, and the CEO who started it still runs it.

RIVA, GeForce 256 and the birth of the "GPU"​

According to Britannica, NVIDIA got its foothold in PC gaming with the RIVA graphics processors in 1997, then the RIVA TNT accelerator in 1998. The big moment came in 1999, the same year as the IPO, with the GeForce 256. It moved transform and lighting (T&L) onto consumer graphics hardware.

That matters because transform and lighting were geometry calculations the CPU used to handle. Once the graphics chip did them itself, it stopped being a simple pixel-pusher and became a processor in its own right. NVIDIA's official timeline names 1999 as the year it introduced the GPU. That's NVIDIA's own branding, but the industry adopted the term.

The Xbox connection​

Britannica says the success of those products won NVIDIA the contract to build graphics hardware for Microsoft's original Xbox, with a $200 million advance. The Xbox link is widely known, but the $200 million figure comes from Britannica and couldn't be independently confirmed here, so treat it as Britannica's number. Either way, the deal put NVIDIA inside a Microsoft platform early, years before the two companies became partners in AI infrastructure.

Britannica also says Sony later chose NVIDIA to help design the RSX Reality Synthesizer graphics processor in the PlayStation 3. That made NVIDIA a supplier to both big console rivals.

The 2000s: beyond gaming, then trouble​

In Britannica's account, the 2000s were NVIDIA's first real move outside games:

• A photorealistic Mars simulation built with NASA in 2003
• Graphics chips supplied for several Audi vehicles
• The purchase of graphics card maker 3dfx Interactive in 2000, an early sign of growth by acquisition
• Forbes' "Company of the Year" title in 2007

Then things went wrong. Britannica describes reports of "abnormal failure rates" in some mobile chips and GPUs caused by manufacturing defects, and says NVIDIA wouldn't name the affected products. A class-action lawsuit followed, which Britannica says was settled in September 2010. It also reports a $1.5 billion cross-licensing deal in January 2011 that ended pending litigation. Those legal details come from Britannica alone and weren't independently checked for this piece.

For long-time PC users, the lesson from that era still applies: a hardware company's reputation depends on the parts nobody sees, like solder, packaging and heat. It also depends on being open about problems when they happen.

CUDA: the 2006 decision that changed everything​

If you remember one date from NVIDIA's history, make it 2006. That year NVIDIA released CUDA (Compute Unified Device Architecture), a platform and programming model that let developers write general-purpose software for NVIDIA GPUs. NVIDIA's timeline presents CUDA as the step that opened GPU parallel processing to science and research.

Why does it matter? Britannica gives the usual explanation: GPUs are good at parallel work, meaning many simple operations at once across many cores, while CPUs are described as handling tasks one after another. That's a useful simplification, but only a simplification. Modern CPUs have many cores and do plenty of parallel work. The better way to put it is that CPUs are built for fast, flexible, branch-heavy work on a few threads, and GPUs are built for huge amounts of the same math across very large data sets.

Neural networks are mostly the second kind of work. NVIDIA's timeline marks 2012, when the AlexNet neural network was trained on NVIDIA GPUs, as a milestone for modern AI.

Why CUDA is still the moat (my analysis)​

This is my read based on general industry knowledge, not a quote from any source. NVIDIA's lasting advantage isn't just faster chips. It's close to twenty years of developer habits, libraries and tools built on CUDA. The company's fiscal 2026 annual report calls CUDA foundational to its platform, alongside hundreds of software libraries, frameworks, SDKs and APIs. That's NVIDIA's own description in its filing. Rivals have mostly struggled to match the software side, not the hardware.

Section summary: CUDA turned the GPU from a graphics part into a general computing platform. The software ecosystem around it probably explains NVIDIA's AI dominance better than any single chip does.

The 2010s: mobile, cars and a shopping spree​

Britannica groups NVIDIA's 2010s growth into several areas:

NVIDIA's own timeline adds real-time ray tracing with RTX in 2018, the feature that changed PC gaming graphics. It lists Omniverse as a platform milestone in 2022, which is a later date than Britannica's 2020 beta. The two dates don't necessarily conflict: one is a beta launch and the other is a broader platform milestone.

Not every bet worked out equally well. Tegra never took over the phone market, though a Tegra chip powers the original Nintendo Switch, which you could argue is a nice comeback. The car business looks small next to the data center: NVIDIA's latest proxy statement lists fiscal 2026 automotive revenue of $2.3 billion against $193.7 billion for data center.

The Arm deal: a lesson in regulators saying no​

In 2020 NVIDIA announced a plan to buy Arm from SoftBank for $40 billion. Arm's chip designs are in a huge range of devices. Britannica points to Apple iPhones and iPads, Amazon Kindles and many cars.

Regulators objected. In December 2021 the U.S. Federal Trade Commission voted to challenge the deal. It argued the deal could hurt competition in advanced driver-assistance processors, data-center SmartNICs and Arm-based CPUs for cloud providers. It also warned that NVIDIA could gain access to sensitive information from Arm's licensees. These were the regulator's allegations, not court findings. NVIDIA dropped the deal in February 2022, after the FTC had begun litigation, and it never closed.

Britannica also says NVIDIA received its first antitrust subpoena from the U.S. Department of Justice back in 2006. That detail comes from Britannica and wasn't independently confirmed here.

Why should Windows users care? Arm-based Windows PCs are now a real product category. If one GPU giant had owned the company that licenses the CPU designs to everyone else in that market, it would have raised hard questions about neutral access for every chipmaker building Windows-on-Arm systems.

Fact-checking the headline numbers​

"92% market share"​

Britannica says NVIDIA held 92% of the discrete graphics market in 2025. It's accurate, but the number needs context. NVIDIA achieved a 92% share of the add-in board (AIB) GPU market in the first quarter of 2025, according to data released by Jon Peddie Research. That figure counts shipments of graphics cards only. It isn't all GPUs and it isn't AI accelerators.

The number also changes from quarter to quarter. Jon Peddie Research later put NVIDIA at 94% of all AIB shipments in Q2 2025, then 92% in Q3, and 94% in Q4 2025. A slight inventory adjustment from AIBs could translate to a big increase or decrease in market share from quarter to quarter. Tom's Hardware, citing the same firm, reported that the industry shipped 44.28 million graphics cards in calendar 2025, up from 34.7 million units in 2024, helped by the GeForce RTX 50-series launch.

The outlook is less rosy. According to TweakTown's summary of the research, high memory prices, tariffs, and integrated GPUs are pressuring the AIB market, which is expected to decline nearly 10% in 2026. For PC builders, that means NVIDIA's lead is huge in a graphics card market that is shrinking and getting more expensive.

The ChatGPT supercomputer and the "six-month chip cycle"​

Britannica says OpenAI built its technology on a supercomputer with 10,000 NVIDIA GPUs. It also says NVIDIA sped up its chip development to one new chip every six months, compared with an 18-month industry norm. Neither claim could be independently verified, so treat both with caution. The 10,000-GPU figure in particular mixes up the training of specific models with OpenAI's wider infrastructure.

What is documented is newer and much bigger. On September 22, 2025, NVIDIA and OpenAI signed a letter of intent to deploy at least 10 gigawatts of NVIDIA systems for OpenAI. NVIDIA said it intended to invest up to $100 billion as capacity comes online, with the first gigawatt on NVIDIA's Vera Rubin platform targeted for the second half of 2026. Those were plans the companies announced, not a finished build-out. Microsoft is named in that announcement as one of the partners in OpenAI's wider infrastructure work.

Current scale​

NVIDIA's fiscal 2026 results, for the year ended January 25, 2026, show where the company stands:

• Revenue: $215.9 billion, up 65% year over year
• Data center: $193.7 billion
• Gaming: $16.0 billion, up 41%
• Professional visualization: $3.2 billion
• Automotive: $2.3 billion

Gaming, where NVIDIA started, is now less than a tenth of revenue. These are fiscal-year figures, not calendar-year ones.

What it means for Windows users and IT pros​

• Gamers and PC builders: NVIDIA's graphics card share leaves buyers with few alternatives. Competition from AMD and Intel is the main thing keeping prices and features in check, and that pressure has been weak lately.
• Developers: CUDA is still the default for GPU computing on Windows and elsewhere. Code that depends only on CUDA is tied to NVIDIA hardware, so weigh portability early.
• Enterprise IT: AI capacity in Azure and other clouds relies heavily on NVIDIA systems, and that concentration affects pricing, availability and supply-chain risk.
• Everyone: The company that once built the chip inside Microsoft's first console is now one of the biggest names in computing. Keep an eye on how regulators handle the next big deal.

The bottom line​

In short, NVIDIA built a 3D graphics chip for gamers and then kept widening what that chip could do. CUDA opened it to scientists, AI researchers adopted it, and now data centers buy it at enormous scale. Britannica's profile gets the broad story right. Some of its specific figures need context (the 92% share), and others need confirmation before you repeat them (the 10,000-GPU supercomputer, the six-month chip cycle). The takeaway for Windows users is that the company behind the graphics card in your PC is now mostly a data-center business.