TSMC, Nvidia & Microsoft: The History of Modern Computing and the AI Revolution

Introduction: how did a near-bankruptcy become the AI revolution?

A technology journey from the near-bankruptcy crisis of the 1990s to the 2026 RTX Spark N1X super chip

Contents: Nvidia's rescue and leadership, the Microsoft partnership and conflicts, CUDA, Tensor Cores & NVLink, TSMC's 2nm & CoWoS, and the 2026 N1X laptops.

Chapter 1: The 1990s bankruptcy crisis and the problem Microsoft DirectX created

In the mid-1990s, Nvidia was pushed to the brink of bankruptcy because it tried to create 3D graphics in a way that was completely different from every other computer company. That very crisis became the most important turning point in the company's history.

1. The NV1 chip and Quadratic Texture Mapping (QTM) technology (May 1995)

• The technology: Nvidia, founded in 1993 by Jensen Huang, Chris Malachowsky and Curtis Priem, released its first chip, the NV1, in May 1995.
• How it worked: Instead of building 3D objects from flat triangles like other companies, the NV1 chip used a method called Quadratic Texture Mapping (QTM) to draw graphics with curved, four-sided surfaces (quads).
• The commercial failure: Although the technology could create curved shapes from less data, it was extremely difficult for game developers to program for. Apart from a few Sega games, no other software support was available for it.

2. The huge problem caused by Microsoft's DirectX 1.0 standard (September 1995)

In September 1995, Microsoft released DirectX 1.0 (Direct3D) to bring PC games on Windows 95 under a single standard. In it, Microsoft declared only triangle-based polygon rendering as the common standard. Because Direct3D did not support curved surfaces, new Windows games would not run on the NV1 chip, and software emulation made them extremely slow. As computer manufacturers cancelled their orders, Nvidia's technology became obsolete overnight.

3. Cancelling the NV2 chip and Sega's 5 million dollar lifeline (1996)

• A hard decision: Jensen Huang decided to cancel, halfway through development, the NV2 chip that was being built on curved-surface technology for Sega's Dreamcast console.
• Only 30 days of cash left: Even after laying off half the staff, Nvidia had only about 30 days' worth of money left in hand.
• The Sega CEO's generosity: Jensen Huang travelled to Japan, met Sega CEO Shoichiro Irimajiri and openly admitted that NV2 was the wrong path. He asked Sega to pay the remaining 5 million dollars on the contract anyway, so that Nvidia could survive, even though the contract was cancelled. Irimajiri agreed and saved Nvidia.

4. The comeback with the RIVA 128 chip (1997)

• With the 5 million dollars from Sega, Nvidia designed the RIVA 128 (NV3) chip, built in hardware for Microsoft's Direct3D triangle technology.
• Released in 1997, the RIVA 128 sold more than one million units in just four months and rescued Nvidia from bankruptcy.

Chapter 2: The fall of 3dfx and Jensen Huang's leadership qualities

1. How Nvidia pushed 3dfx behind

• All-in-One design (RIVA 128): Because 3dfx Voodoo cards handled only 3D, users had to buy a separate 2D card as well. The RIVA 128 chip delivered both 2D and 3D on a single chip.
• The Glide vs DirectX battle: While 3dfx trusted only its own Glide API, Nvidia supported the common standards, Microsoft's DirectX and OpenGL.
• The world's first GPU (GeForce 256, 1999): In October 1999 Nvidia released the GeForce 256 chip with Hardware Transform and Lighting (T&L) and named it the "GPU" (Graphics Processing Unit). It took the computing load off the CPU and multiplied gaming speed.
• Six-month release cadence: By releasing a new chip every six months, Nvidia drove 3dfx into bankruptcy and bought its assets in 2000.

2. Jensen Huang's leadership approach, forged in crisis

• The "we are 30 days from going out of business" mindset: To keep complacency about success away, he built a culture of always acting with urgency.
• Intellectual honesty about mistakes: Admitting wrong plans immediately and learning from failures without hiding them became Nvidia's core rule.
• Simulation First: He made it permanent practice to test a chip with 100% accuracy in a software emulator before manufacturing it.
• Flat organization: He cut layers of management titles and built a structure in which more than 50 senior leaders report directly to him.

Chapter 3: The history of the Nvidia–Microsoft partnership and conflicts (1995–2026)

Chapter 4: The foundation of the modern AI revolution (CUDA, Tensor Cores & NVLink)

• CUDA software (2006): CUDA made it possible to use a GPU's thousands of cores for scientific and mathematical computing through the C/C++ languages, without knowing graphics programming. In 2012, the AlexNet AI model used CUDA on two Nvidia gaming cards to win the ImageNet competition, laying the foundation for today's AI revolution.
• Tensor Cores: Instead of multiplying just two numbers like ordinary cores, they multiply an entire matrix of numbers at once, completing AI computations at lightning speed with lower bit precision (FP8/FP4).
• NVLink & NVLink-C2C: In data centres, NVLink directly connects hundreds of GPUs so they work like a single super chip. In N1X laptops, the NVLink-C2C link connects the ARM CPU and the Blackwell GPU inside the same chip and turns the 128 GB of unified memory into a pool shared by both. This makes it possible to run large AI models (local LLMs) without an internet connection.

Chapter 5: The history and technology of TSMC (Taiwan Semiconductor Manufacturing Company)

Taiwan, the chip capital of the world: TSMC manufactures more than 90% of the world's most advanced semiconductor chips. In addition, UMC (chip manufacturing), MediaTek (the company that co-designed the N1X chip's ARM CPU with Nvidia), Realtek, Novatek and the chip packaging company ASE Technology also play important roles in Taiwan.

1. TSMC historical milestones (1987–2026)

• 1987 – the birth of the pure-play foundry: TSMC was founded by Morris Chang on February 21, 1987. By creating the model of "not designing chips under its own name, but only manufacturing chips designed by other companies", it laid the foundation for the growth of companies such as Nvidia, Apple, AMD and Qualcomm.
• 1997–1998 – the partnership with Nvidia: After the success of the RIVA 128, TSMC accepted the letter Jensen Huang wrote and began manufacturing Nvidia's chips.
• 2014–2018 – the Apple contract and overtaking Intel: After winning the Apple A8 chip production in 2014, TSMC used ASML's EUV lithography machines from the Netherlands to overtake Intel in 2018 and become the world's number one chip manufacturer.
• 2020–2026 – global expansion (the "Silicon Shield"): Serving as Taiwan's "silicon shield", TSMC is building new factories in the United States (Arizona), Japan (Kumamoto) and Germany (Dresden) to reduce geopolitical risks.

2. TSMC's 2nm (N2) and CoWoS technologies

• 2nm (N2) nanosheet transistors: Instead of the older FinFET design, TSMC uses Gate-All-Around (nanosheet) technology, in which the transistor gate surrounds the current path on all four sides. This reduces power leakage and delivers 10–15% more speed and 25–30% lower power consumption than 3nm chips.
• CoWoS (Chip-on-Wafer-on-Substrate): This 3D packaging technology stacks separate chips such as the ARM CPU, the Blackwell GPU and high-speed memory closely together on a thin silicon base called a silicon interposer, joining them into a single chip. This is what gives the Nvidia N1X chip its monstrous speed.

Chapter 6: The 2026 RTX Spark N1X Windows 11 laptops

The Nvidia N1X chip went on sale on October 16, 2026, exclusively for Windows 11 PCs (it does not work on Mac computers; WSL is recommended for Linux use). Here are the details of the key models with a 20-core ARM CPU, a 6,144-core Blackwell GPU and up to 128GB of RAM (in Australian market prices):

Read more articles