What You Can Learn From Our $20B Deal
Chamath tells the Groq story — a $10m check in 2015, seven years without product-market fit, a $13B Nvidia licensing deal closed in weeks after an NVLink Fusion experiment. Doubles as a manifesto on backing technical founders over org-builders.
Key points
- Groq was born when Chamath became fascinated with Google's TPU after a 2015 Sundar Pichai CNBC mention, tracked down the inventor Jonathan Ross, and wrote a $10m check for roughly a third of the company.
- The company spent seven years with no product-market fit — taped out a chip, failed to win government contracts, kept trundling. LLM inference changed everything because Groq's SRAM-heavy architecture was accidentally perfect for the decode phase.
- The Nvidia deal consummated astonishingly fast: NVLink Fusion experiment in May 2025 → Saturday call with Jensen → Groq team in Australia woken up → deal terms around Dec 17 → '$13 billion in the account on Tuesday, get it done.'
- Chamath's heuristic when a founding team fractures: always back the technical person the company wouldn't exist without. Builders over managers. Non-negotiable.
- Broadside against modern org design: VP of Engineering, CTO, CFO with VPs of Audit, Tax, FP&A — all 'vestige of some crappy enterprise software.' He points to Jensen running with 60-80 direct reports and Elon ping-ponging across 5-7 companies as counter-examples.
- A-players are meaningless in isolation — only A-teams of high-agency, self-directed people matter. They get motivation from the work itself, aren't one-foot-out-the-door insecure politickers.
- Framework for AI silicon: training is systems-heavy; inference is north-south traffic, pre-fill and decode as distinct problems. The architectural choice (SRAM for decode) that crushes inference does not make sense for training — and this asymmetry is the opportunity.
Notable quotes
There'll be 13 billion in the account on Tuesday. Get it done.
Org building is stupid and worthless... organizational charts accelerate and celebrate the politician or the maneuverer or the kisser, but very rarely do they accelerate and highlight the technical genius.
A-players are meaningless and irrelevant and useless. But there are teams that can perform at an A level.
We didn't have product market fit for seven years... most of these businesses will go through many years of darkness — five, six, seven years where nothing is working until it all of a sudden works.
Themes
- The $13B phone call: how Groq's Nvidia deal closed in weeks
- Back the builder, not the org-builder
- Inference silicon as the asymmetric bet
- Seven years of darkness is the job