Nvidia
A chip company whose GPUs and CUDA software power gaming, AI and accelerated computing.
Nvidia survived about 90 graphics rivals with a six-month chip cycle, then spent over a decade on CUDA while its stock fell 80%. That bet and full-stack systems took gross margin from 24% to 70% and made it a $1.1 trillion company. Lesson: back a platform with a free, developer-friendly layer early, teach the market, and endure the drawdowns.
- 1Commit $25 millionNVIDIA's K-12 AI commitment
- 2Pick StudyFetchOthers target educators and administrators
- 3Co-create AI courseFor districts and students
- 4Reach high schoolersGoal: 250,000+ on AI literacy path
From NVIDIA's K-12 AI education partnership
What was said about Nvidia
NVIDIA picked StudyFetch because it engages learners, not just administrators. NVIDIA chose StudyFetch for K-12 AI education because most products target educators and administrators, and none had engaged the learner directly. The partnership sits inside NVIDIA's $25 million K-12 AI commitment and aims to put 250,000+ high schoolers on a path to AI literacy.
StudyFetch's consumer engagement skill was the differentiator, and the two are co-creating an AI literacy course for districts and students. The quote from the source: "there has never been so far a successful product that has been able to engage the end user, which is the learner." The lesson for builders is to win partners by owning the end user that others ignore.
How Jensen Works
Jensen runs it flat with 60 direct reports and no one-on-ones, and reads about 100 'top five' emails a day. CUDA work cut gross margin from 45% to 35% and the stock fell over 80%, yet it became the center of AI. Built in 2007 and 2008 to make GPUs easy to program, it cost margin and stock value but became the foundation of Nvidia's AI position. Nvidia taught it through schools, over 100 talks in a year and a textbook that sold tens of thousands of copies.
NVIDIA picked StudyFetch because it engages learners, not just administrators. NVIDIA chose StudyFetch for K-12 AI education because most products target educators and administrators, and none had engaged the learner directly. The partnership sits inside NVIDIA's $25 million K-12 AI commitment and aims to put 250,000+ high schoolers on a path to AI literacy.
StudyFetch's consumer engagement skill was the differentiator, and the two are co-creating an AI literacy course for districts and students. The quote from the source: "there has never been so far a successful product that has been able to engage the end user, which is the learner." The lesson for builders is to win partners by owning the end user that others ignore.
NVIDIA CEO Jensen Huang
Nvidia simulates risk before betting big, as with the Riva 128 emulator, and keeps every chip CUDA compatible across roughly 250 to 300 million GPUs. It is run like a computing stack with mission as the boss, survived 80% stock drops during the CUDA bet, and was valued at $1.1 trillion at recording. CUDA grew out of the earlier Cg layer and was already used in molecular dynamics, CT reconstruction and seismic processing, which tied Nvidia to AI researchers. Gaming chips carried it into a large install base, and Nvidia's market value sank to about two or three billion dollars during the bet. Mellanox brought the networking AI needs to shard one job across many processors, which commodity Ethernet handles poorly. Its Israel team now numbers about 3,200, and Huang calls the deal one of his best strategic decisions. It began about 17 years ago from an engineer's demo of capturing, encoding and streaming a frame buffer. It was Nvidia's first data center product, separating computing from the screen. Huang's current example of a zero billion dollar market, with roughly 40 customers including Amazon and BMW. He still asks himself how hard it can be about it each morning.
Nvidia Part III: The Dawn of the AI Era (2022-2023)
Data center revenue hit $10.3 billion in one quarter, up 141% on the prior quarter, with total revenue of $13.5 billion and gross margin of 70% versus 24% in the commodity days. It spent five years building a full GPU data center stack, sells H100s at $40,000 and DGX boxes from $500,000, and a $11 billion Q2 forecast made it a trillion-dollar company. Nvidia bought it for $7 billion in 2020 when most thought Ethernet was enough. InfiniBand lets many GPUs act as one machine, which proved critical for training large language models. It starts at $37,000 a month for an A100 system that costs about $120,000 to build, a payback of about three months. It runs in Azure, Oracle and Google, and gives Nvidia a direct relationship with enterprises.
Nvidia: The Machine Learning Company (2006-2022)
Spent over a decade building CUDA, which was unusable for six-plus years while the stock fell 80% from a roughly $20 billion peak. AlexNet's 2012 win made it the base of deep learning. Data center revenue tripled in two years to over $10.5 billion, gross margin reached 66%, and revenue grew 60% to $27 billion. Nvidia bought it for about $7 billion in 2020. It let Nvidia sell the whole data center as one box and added the DPU as a third chip type.
Nvidia: The GPU Company (1993-2006)
Founded in 1993, it outlasted about 90 graphics rivals by shipping a chip every six months and then making the GPU programmable. Revenue hit about $1.4 billion in 2001, but gross margin was only 29% in 2004 and it still lacked lasting power.
How to Dominate for Decades | Doug Leone, Sequoia Capital
Sequoia owned about 20% of it. No graphics chip company had survived two cycles before it refounded itself as an AI company under the same founder.
How StudyFetch Used Creator-Led Growth to Hit $10M+ ARR in 3 Years
Chose StudyFetch because it engages learners, which other K-12 AI products have not. The partnership sits inside NVIDIA's $25 million K-12 AI commitment and aims to put 250,000+ high schoolers on a path to AI literacy.
Google Part III: The AI Company. Google is amazingly well-positioned... will they win in AI?
Google's 2014 $130 million GPU order came when Nvidia had about $4 billion in revenue and a $10 billion market cap. Google pays Broadcom about a 50% margin on TPUs versus roughly 80% paid to Nvidia.
How to win in the AI era: Ship weekly, embrace technical debt, ruthlessly cut scope, and more
Partnered with Captions to build the eye contact feature, though it was unsure at first why anyone needed it. Almost every app has since copied the feature.
TSMC founder Morris Chang
In 1997 it was a bankrupt-bound startup of 50 or 60 people when Chang answered Jensen Huang's letter, and its games chip made it a top-five TSMC customer within two or three years. In 2009 Chang settled a 40nm yield dispute with an offer of over $100 million, with a 48-hour deadline.
TSMC
It raised only $20 million and never built a fab, relying on TSMC, which makes all of its chips. It is an example of the fabless model the foundry enabled.
More in Developer tools
Developer toolsClaude Code
Anthropic's agentic coding tool that lets Claude work on code end to end.
15 sources · 46 tips
Developer toolsCursor
An AI code editor that lets engineers build software with AI while staying in control.
33 sources · 47 tips
Developer toolsGitHub Copilot
A code hosting and collaboration platform for software developers, with free and paid plans.
16 sources · 45 tips
Developer toolsLovable
An AI app builder that lets people create software by chatting with an agent.
11 sources · 48 tipsB7Dev
An app development company that builds apps for founders and clients.
5 sources · 28 tips
Developer toolsLinear
A fast, well-designed project and issue tracking tool for software teams.
12 sources · 41 tips
Developer toolsFirebase
Google's backend platform for building apps.
16 sources · 20 tips
Developer toolsVercel
A cloud platform for building, deploying and scaling front-end and AI web applications.
6 sources · 22 tips