Jensen Huang is the powermonger of the AI age - literally. His company Nvidia slogged away for decades in the foothills of the semiconductor industry. Now it is the most valuable business in the world, selling the microchips that power AI computing. New Yorker columnist Stephen Witt’s book ‘The Thinking Machine’ digs into this remarkable transformation. It asks what shaped Huang and what makes his company different. Here is what stood out for me.

Share

1. Jensen Huang is a study in relentlessness

Huang has a modest backstory. As a Taiwanese migrant arriving in the US in the 1970s, Huang was bullied at school in Oneida, Kentucky. But over the course of a four-decade career he has worked his way through the tech industry, co-founded Nvidia in 1993, and led it through multiple setbacks to its current position. This year Nvidia became the first company in the world with a market capitalization above $4 trillion. Huang is the longest-tenured major tech CEO. And absolutely no one has earned more from AI than Huang.

Behind this rise is a relentless drive to succeed. It is exhibited in his full-on work ethic and his management ethos, constantly pushing his employees with long days and short deadlines (and working ‘at the speed of light’). It is visible in his hassling of suppliers to deliver ever faster. There’s also his pursuit of the competition: Huang was nicknamed ‘Darth Vader’ by employees of a company he was trying to buy.

Then there are his personal interactions, working with many direct reports, responding to emails from the whole workforce, and regularly showing off his famously volcanic temper - the ‘Wrath of Huang.’ His doggedness comes out particularly strongly in the accounts of the early days, when product failures left the company in crisis - ‘running on fumes’ and ‘30 days from going out of business.’ Huang kept pushing.

2. Nvidia is founded on the unfashionable

Time after time, Huang swam against the tide, targeting marginal or offbeat opportunities - what he calls ‘zero billion dollar’ markets that are not yet in existence. Until recent years, this approach generated incomprehension and disdain from competitors and investors alike.

This contrarian streak comes across in the company’s focus on the niche sector of gaming graphics chips, a market neglected by the semiconductor gorilla of the time, Intel. Particularly from the late 90s, Nvidia became synonymous with accelerating chip performance for cult games such as Quake, whose 3D graphics demanded increasing processing power. The new breed of chips were called ‘Graphics Processing Units’ or GPUs.

Then there was Nvidia’s pursuit of parallel computing, a deeply discredited concept until Nvidia made it work. In Witt’s telling summary, Nvidia had created massively powerful processors that acted like a turbo-charged blender for big data, a far cry from the traditional chef’s knife of a normal chip.

Finally, there was CUDA, Nvidia’s framework for turning GPUs into a high-performance computing resource for the world beyond gamers. CUDA was initially conceived in 2006, and pushed hard by Huang over years despite widespread market indifference. ‘Huang was bringing supercomputing to the masses. But the masses didn’t want it.’

But no one was talking about AI yet.

3. Nvidia ignited the possibilities of AI neural networks

At the same time that Nvidia was turning on the taps of processing power, AI was languishing in obscurity. In March 2013 at the Nvidia GTC (GPU Tech Conference), Huang’s world-famous leather jacket made its first public appearance, but neither AI nor neural nets were mentioned.

But something was stirring in image recognition, a classic computing challenge that had outfoxed waves of researchers. With the deep learning and neural network breakthroughs of the likes of Fei-Fei Li, Andrew Ng and Geoffrey Hinton, AI had compelling new momentum - and an extremely data-intensive approach in search of processing power. Witt is at his most lyrical on the coming together of Nvidia’s product and AI: ‘Neural nets running on parallel computers [became] the twin strands of DNA, for a new and powerful organism looking to consume all the data in the world.’

Huang understood the potential, fixating on this ‘OIALO’ (once-in-a-lifetime-opportunity), and ramping up his hours yet again. Soon GPUs became indispensable in powering the neural nets at the heart of the AI revolution, gaining further impetus with Generative AI, and still rolling out today. Nvidia was at the very center of this, and the soaring valuation followed.

4. Nvidia is a software company

Like most other major microchip companies, Nvidia doesn’t get its hands dirty with manufacturing. That’s outsourced to the likes of Taiwan’s TSMC, in the same way that Foxconn manufactures for Apple.

But Nvidia goes further, and the book shows that in many ways Nvidia has moved beyond manufacturing. Huang himself has declared the death of Moore’s Law, the 60-year-old article of faith that ever more will be crammed into a microchip. Instead, progress is being driven by software, innovative programming and mathematics. ‘The reason that Nvidia succeeded was not that its circuits were better, the reason is that its software was better.’

At the heart of this is CUDA, the architecture that surrounds Nvidia chips, and on which hundreds of free but proprietary tools have been developed for specialist programmers, in areas ranging from finance to automotive to cybersecurity. Software drives continuing Nvidia advances, and keeps competitors in their place.

5. Huang doesn’t do AI doom

As the book concludes, it moves to a more questioning stance towards Huang and the role of his company. Witt asks some of the key policy questions fitting for a company at the forefront of the AI revolution, particularly the impact of energy-hungry AI factories on climate change, and the geopolitics of China and semiconductors.

These lines of inquiry could have been taken further, not least because Nvidia’s influence is at an all-time high. In August Donald Trump announced a revenue-sharing deal with Nvidia and AMD, in return for access to the Chinese market. This raises national security concerns for some, but for all it showcases Huang’s profile, making waves even in the world of realpolitik.

But Witt’s concerns are focused on the high-profile scenarios of AI upheaval and takeover. He reveals his own misgivings about the future of his profession, and humanity more broadly, as step by step AI becomes more powerful. But Huang doesn’t engage on the big questions. His flippant bottom line on AI doom is that AI is no more dangerous than a microwave.

Witt tries to discuss his doubts in person with Huang, who is coldly dismissive and defensive: ‘I don’t like these probing questions.’ Witt notes he is surrounded by staff who are not inclined to challenge him, being ‘more afraid of Jensen yelling at them than they were of wiping out the human race.’

It is hard to avoid the conclusion of prior AIB reviews (The AI Con / How to Think About AI), that many of the leaders of big tech are conflicted and temperamentally unsuited to address the big social and political questions around AI. And in Witt’s story of the dressing down, it seems there are just a few steps from genial tech titan to Bond villain.

The Thinking Machine is a highly readable portrait. It is the story of Huang’s energy as he ricochets through setbacks, clashes and breakthroughs, all the while constantly pushing for more. It reveals some of the complexities of his character, and shows there is more to ask of the company, particularly as Nvidia and Huang remain so influential. One other question hangs in the air to be answered, posed by the author himself: What happens to Nvidia in the future - after Huang? Until then the Jensen show continues.