Thinking Machine
Nvidia's bet on GPUs as the substrate of deep learning.
Summary
Witt, a journalist who previously wrote How Music Got Free, wrote a biography of Jensen Huang and a history of Nvidia focused on the company's role in the artificial intelligence transition.
The narrative covers the founding, the early graphics competition, the invention of the graphics processing unit, and the decision to build CUDA, the software platform allowing general computation on graphics hardware. Witt treats CUDA as the central strategic bet: a decade of investment with no identified market, against investor objection, which positioned the company when deep learning researchers found that neural network training ran efficiently on the same hardware. He covers the AlexNet result in 2012, in which a neural network trained on Nvidia hardware won an image recognition competition by a large margin, and the subsequent adoption across the field.
Witt has access to Huang and to many people around him, and the portrait is of an unusual manager: extremely flat reporting structure, no formal one-to-one meetings, direct written status reports from throughout the organisation, public criticism, and personal involvement in technical detail at a level unusual for a chief executive of a company this size.
The book covers the current period, including the scale of demand from the largest technology companies, the concentration of manufacturing at TSMC and the geopolitical exposure that creates, and the competitive threats from customers designing their own silicon.
It is written during a period of extraordinary success with the company's cooperation, which shapes the assessment. Witt is more sceptical than a purely authorised account would be, but the reader should recognise that any book about a company at a valuation peak carries that problem. It overlaps substantially with Tae Kim's Nvidia Way, also in this reference.
Key takeaways 7
- CUDA was a decade-long investment with no identified market, and it is what positioned Nvidia when deep learning arrived.
- The 2012 AlexNet result demonstrated that neural network training ran efficiently on graphics hardware, which redirected the entire field.
- Nvidia's advantage rests on the software platform and developer ecosystem more than on the chips themselves.
- Huang operates with an extremely flat structure and direct written reporting from throughout the organisation rather than through a hierarchy.
- Manufacturing concentration at TSMC creates a geopolitical exposure that no amount of design capability offsets.
- The company's largest customers are designing their own silicon, which is the principal structural threat to its position.
- Building a capability before its application exists requires a management structure that can sustain the decision against investor pressure.
Who should read it
Investors and founders in technology who want the account of Nvidia's long-horizon platform bet and the management structure that sustained it.
Similar books 6
Chosen by shared topics, the entries that hold them, and the people who recommended both.
The OpenAI-DeepMind race that set the shape of Nvidia's customer base.
Recommended by: Andrew Wilkinson
Fifteen years of AI through Demis Hassabis and DeepMind, from AlphaGo to AlphaFold.
How Hinton, LeCun and Bengio kept neural networks alive until deep learning took off.
Nvidia from its 1993 founding to the AI boom, from interviews with Jensen Huang.
How chips became the key to economic and military power, made in a few chokepoints.
Recommended by: Andrew Wilkinson
Autonomy from the incumbent and Waymo side.
Where it sits in the reference 3
Topics