AI summary by TwinMind
Nvidia ft. Jensen Huang - An overnight success story 30 years in the making
Sequoia Capital
38m 54s
Company Origins and Founding (1993)
- Founded by Jensen Huang, Chris Malachowsky, Curtis Priem in 1993
- All three worked in workstation industry (Sun Microsystems, LSI Logic)
- Met at Denny's restaurant to brainstorm company ideas
- Ordered one bottomless cup of coffee, worked for 4 hours
- Jensen previously worked as dishwasher/busboy/waiter at Denny's
- Recognized PC Revolution opportunity (1993-1995)
- PCs reaching consumer price points and ease of use
- Identified gaming as primary consumer application
- PCs had no graphics, sound, multimedia capabilities at the time
- Target: 3D graphics chips for PC gaming
- Market research showed "$0 billion market" for 3D graphics in 1993
- Had to buy Gateway 2000 PC to learn Windows/DOS programming
Early Funding and Challenges
- Jensen pitched to Don Valentine at Sequoia Capital
- Wilfred Corrigan (LSI Logic) made introduction
- Valentine's response: "If you lose my money I'll kill you"
- Mark Stevens joined as Sequoia board representative in 1993
- Competitive landscape challenges:
- Crowded 2D graphics market (Xilinx, Altera, Cirrus Logic, Chips & Technologies)
- Gaming market considered "not serious" - teenagers only
- Much smaller than movies/other media markets
NV1 Chip Failure (1994-1995)
- First product: NV1 chip - 18-24 months to develop
- "Swiss army knife" approach - multiple functions:
- 3D graphics, video processing, audio wave table, IO ports, game acceleration
- Used quadrangles architecture (not industry standard triangles)
- Came with dongles, felt like "octopus"
- Partnership with Diamond Multimedia:
- Sold 250,000 NV1 chips
- 249,000 returned due to poor retail sales
- Nearly bankrupted company
- Lessons learned: product-market fit, pricing, positioning, competition analysis
NV2 Crisis and Sega Partnership (1995)
- Developing NV2 in partnership with Sega for next-generation console
- Microsoft released Windows 95 with DirectX API
- Used triangles architecture (inverse texture mapping)
- Nvidia stuck with quadrangles (forward texture mapping)
- 50 competitors adopted DirectX standard
- Existential choice: finish Sega contract and die, or abandon and die immediately
- Jensen's decision: approached Sega CEO Irimajiri
- Requested release from contract but full payment ($5 million)
- Sega agreed despite getting nothing in return
- Gave Nvidia enough money to survive
RIVA 128 Breakthrough (1996-1997)
- "One shot" mentality - company would die if chip failed
- Innovative development process:
- Worked backwards from perfect first tape-out
- Developed all software in advance
- Completed in 6-8 months "on complete fumes"
- Purchased emulator from bankrupt company (ICE)
- Size of refrigerator, cost of failed company's inventory
- First PC chip ever emulated before manufacturing
- RIVA 128 (NV3) succeeded:
- Worked perfectly on first tape-out
- Sold 1 million units in first 4 months
- Revolutionized chip design methodology
- Enabled 2x faster development cadence than competitors
Market Dominance and GPU Creation (1997-1999)
- By 1999, far ahead of competition
- Quake 3 Arena (1997) demonstrated superior performance
- GeForce 256 (1999): world's first programmable GPU
- Added programmability to acceleration
- Created "accelerated computing" concept
- 1000x more efficient than CPU for specific tasks
- Went public in 1999
- Microsoft chose GeForce for Xbox console
CUDA and AI Pivot (2006-2012)
- 2006: Released CUDA programming interface
- Enabled GPU use beyond gaming:
- Molecular dynamics, seismic processing, CT reconstruction
- Graduate students using gaming cards for research
- Personal supercomputers from retail gaming cards
- Key researcher interactions:
- Taiwan quantum chemist with closet full of GeForce cards
- Andrew Ng at Stanford (2008-2009) using GPUs for deep learning
- Ian Goodfellow built GPU server for neural network experiments
- 10x-100x speedups for parallel processing vs sequential CPU
AI Market Bet (2012-2015)
- Market risk: AI was "$0 billion market" again
- AI had multiple false starts over 40 years
- Company was now successful, public, multibillion-dollar
- Crucible decision: stay in gaming lane or bet on unproven AI market
- "Wholesale pivot" across entire company
- Sought out every AI researcher globally
- Early positive feedback from researcher adoption
AI Revolution Success (2015-Present)
- Huang's Law: computing advancing 1 million times every 10 years (vs Moore's Law 10x every 5 years)
- Compounded exponentials across chip, system, algorithm, and AI levels
- Since ImageNet/AlexNet: 1 million times computing advancement
- Problems previously requiring 30 years now solvable in 5 years
- Platform expansion beyond AI: digital twins, climate modeling, drug discovery
- ChatGPT launch (November 2022): 1 million users in 5 days, 100 million in 8 weeks
Leadership Philosophy and Lessons
- 30-year "time span of discretion" - thinking in decades
- "30 days from going out of business" - constant urgency
- Multiple "diving catches" throughout company history
- Belief in solving problems nobody imagines company can solve
- Overnight success that took 30 years to achieve
