Nvidia's rise from 90 competitors to GPU dominance, 1993–2006

Executive overview

In 1993, Nvidia entered a brutally commoditised market with 90 funded competitors all racing to build 3D graphics chips for PCs. The company nearly went bankrupt twice in its first decade, surviving each time by betting everything on a contrarian technical move.

Jensen Huang's core insight: in a business governed by Moore's law, standing still is slow death. Nvidia's survival required complete reinvention — not once, but repeatedly.

The company that learned to ship a new chip generation every six months, while competitors took 18–24 months, laid the only foundation capable of dominating the GPU era.

Founding and early missteps (1993–1996)

• Jensen Huang, Chris Malachowski, and Curtis Prem co-founded Nvidia at a Denny's in late 1992
• Thesis: bring SGI-quality 3D graphics to consumer PCs, starting with gaming
• Sequoia (Don Valentine) and Sutter Hill invested $2M at a $6M post-money valuation
• First chip used quadrilaterals as its polygon primitive — the industry standardised on triangles via Microsoft's Direct3D
• Sega deal for arcade/console graphics collapsed when Sega also switched to triangles
• Left with nine months of runway, 100+ engineers, and an architecture nobody wanted

Near-death and the six-month chip cycle

• Jensen cut 70% of staff, dropping to ~35 people
• To ship a new chip in nine months — half the normal cycle — the team used software emulation, running at one frame every 30 seconds instead of 60 fps
• They taped out to production without a physical prototype, spending $1M (a third of remaining cash) on the emulator
• The resulting Riva 128 (1997) was overpowered and buggy — only two-thirds of Direct3D blend modes worked
• Jensen personally lobbied developers to use only the working subset; they agreed because raw performance was all that mattered to consumers
• 1 million units sold within four months; the six-month ship cycle became permanent Nvidia practice

Becoming a real GPU and landing TSMC

• After the Riva 128's success, Jensen wrote a physical letter to Morris Chang at TSMC — he couldn't get through sales channels
• Morris called Jensen directly; the Nvidia office froze; a multi-year foundry partnership followed
• In 1999, Nvidia rebranded its product line as GeForce and coined the term GPU (graphics processing unit)
• The GeForce 256 delivered 5x better performance than any competitor
• The GPU label was a strategic declaration: this is not a sound card waiting to be commoditised into Intel's motherboard

Programmable shaders and the Xbox

• The GeForce 3 introduced programmable shaders — for the first time, developers could write code that ran on the GPU itself, enabling dynamic lighting and custom visual styles
• Nvidia developed CG (a C extension for GPU programming) in partnership with Microsoft
• Microsoft chose Nvidia to supply the GPU for the original Xbox; the deal was worth $500M/year with a $200M advance — half of Nvidia's total revenue at the time
• Revenue trajectory: $158M (FY1999) → $375M → $735M → $1.4B (FY2002), making Nvidia the fastest semiconductor company ever to reach $1B in revenue
• Added to the S&P 500 in the company's ninth year of existence

Intel's threat and Nvidia's positioning

• Intel's standard playbook: let peripheral markets develop, then absorb them into the motherboard chipset (Sound Blaster cards, networking cards, etc.)
• Intel launched its own discrete GPU cards around 1999 — they were poorly reviewed and quickly abandoned
• Intel integrated graphics captured the "good enough" market (laptops, basic PCs) but could never satisfy the performance ceiling that gamers and professionals always pushed upward
• Nvidia's bet: the demand for graphical performance is effectively unbounded — there will always be a market for a discrete GPU

Seeds of scientific computing

• A Stanford quantum chemistry researcher discovered that porting molecular models into CG cut computation time from weeks to hours, matching supercomputer results
• Nvidia began attracting researchers who repurposed GeForce cards for massively parallel non-graphics workloads
• Nvidia invested in Keyhole — the company that became Google Earth — because it demonstrated GPU-powered simulation of the real world
• The perception that Nvidia was drifting from gaming toward niche research contributed to AMD approaching Nvidia about an acquisition; Jensen refused unless he became CEO of the combined entity, killing the deal

Strategic position by 2006

• Gross margins: 29% in 2004, far below the 66% they would later achieve — pricing power was being competed away
• AMD acquired ATI instead, creating the rival GPU line that exists today
• Intel announced Project Larrabee, a serious GPU push
• Xbox contract revenue was winding down; Xbox 360 used an AMD/ATI chip
• Nvidia had product-market fit but not yet durable power — switching costs were low, process advantage was compressing, CG was not the lock-in that CUDA would later become
• The company was investing heavily in GPU-accelerated scientific computing with no clear near-term revenue justification — a bet that defined the next era