A room full of high achievers suddenly aware that everything they’d done to get here — the memorizing, the test-perfecting — had trained them for the wrong century.
Huang — the co-founder and CEO of Nvidia — had not come to this idea through philosophy. He arrived by surviving.
Years earlier, in a low-slung building in Santa Clara, fluorescent light laying a cold sheen on concrete floors, Huang walked the corridors of his office gripping a Styrofoam cup as if caffeine could buy clarity. CUDA was barely more than a rumor in those rooms — a word printed on draft slides and muttered in halls. It stood for Compute Unified Device Architecture, though almost no one outside Nvidia knew or cared. For everyone else, it sounded technical, niche, irrelevant. Yet what CUDA really was — for anyone who has planted a garden or run a kitchen — was a simple kind of power: a way to make thousands of tiny tasks happen at once, instead of one after another.
Most computers, until then, thought like a single cook at a stove — chopping, stirring, flipping — one action per moment. CUDA was like convincing an entire crew of cooks to work in parallel, each handling one slice of the meal, side by side, so the feast arrived in minutes instead of hours. It turned graphics chips — originally built to make video games look pretty — into engines that could think fast enough to train artificial intelligence, simulate weather, model proteins, or diagnose cancer. A tool no one had asked for, until suddenly the whole world needed it.
In 2006, the year Nvidia first released CUDA, the reception inside Silicon Valley bordered on polite confusion. Parallel computing was a phrase that made investors’ eyes glaze. Huang, already weary from nights spent sketching architecture diagrams by hand, scheduled a meeting that would later go unrecorded in formal history but burned itself into the memory of those who were there.
The conference room had a long, fingerprint-stained glass wall facing a row of eucalyptus trees. Rain that morning had turned the trunks nearly black. Engineers sat with laptops half-closed, unsure whether this was a funeral or a rally. One executive leaned forward and asked the question that nearly suffocated the idea before it breathed: “Who will even use this?”
Silence. Then someone coughed.
CUDA, at that moment, was like electricity before the lightbulb. A force waiting for a purpose. Huang stood with his palms flat on the table. His voice did not rise, but the words were a blade: We are going to build the railroad before anyone knows they need a train.
There was no applause. Only the quiet scraping of chairs as the meeting ended. Several left believing they’d just watched a CEO step into delusion.
Months later, when layoffs brushed the edges of the org chart, rumors flared. CUDA was to blame. CUDA had no buyers. CUDA was an ego project – it was proprietary in an era of open source. But Huang did what those destined for retroactive myth always do: he endured humiliation in real time and called it strategy.
He made one more choice — the kind that only matters years later. He refused to pivot.
His leather jacket may look effortless now, but it was not bought with ease. It was bought with refusal.
Back in Cambridge, a student raised her hand — a motion halfway between courage and surrender —