Texas Instruments' CTO on Making Hardware in the Age of AI

In publishing, it’s known as “killing your darlings.” It’s been alternately ascribed to the usual suspects for quotations of uncertain origin — Faulkner, Wilde, Chekov, Eudora Welty.  

It’s universal advice about avoiding myopia that applies across disciplines. Texas Instruments CTO, Ahmad Bahai, puts it well toward the end of our upcoming podcast interview, noting, “The worst thing [that] can happen for a researcher is to fall so much in love with their own ideas, they ignore the reality of what’s going on around them, and say, 'I've worked on this for five years. I have to make it work.' ” 

While the sentiment is transferable, the consequences can be starkly different. In the case of the sorts of technologies Bahai and I discuss, it’s the question of paying attention to one’s surroundings. It’s about reading the research, following the market, and appreciating the sorts of breakthroughs that shatter orthodoxies. Sunk cost is a powerful influence on a personal level, and anyone who has operated in a corporate environment can tell you that even the most progressive-seeming businesses are not especially nimble. 

Here’s another overtired, but largely accurate cliché for you: hardware is hard. Maintaining a successful technology company requires the ability to read the writing on the wall. This requires a much longer horizon when dealing with atoms. 

“Even when you wake up to the fact that AI is going to be big, a new device takes five, six years to develop. It's not like tomorrow,” says Bahai. “If you just wait for the next market to start something, then you're always behind the curve and not having like opportunity to capture these growth opportunities. That's why I think we need to look at these trends and these exponentials from a foundational technology viewpoint and then react to the markets.” 

Even by hardware standards, chip fabrication is a notoriously lengthy process, requiring highly specialized tools, hundreds of steps, and months to years to complete. The semi-conductor industry is notoriously volatile, as new technologies and industry trends dictate market success. Big successes require big swings — and a lot of foresight. As I’ve noted time and again in the pages of Automated, NVIDIA’s overnight success in robotics dates back a decade ago, when the chipmaker took a huge gamble to exit mobile for the category we’ve more recently come to know as “physical AI.” 

Part of looking like a genius in hindsight seems to be the ability to predict trends on more abstract levels. In the case of silicon giants, the specifics of the end user are often of little consequence — that is to say, they have the same or similar underlying compute requirements. This is why, say, companies that have been developing for self-driving cars have a built-in headstart in robotics. 

Given ongoing conversations around the exponential power and resource demands from generative AI usage, it’s unsurprising that data centers are also a key factor. 

“Data center is becoming one of the leading drivers of innovation across the board in different disciplines of the semiconductor,” says Bahai. “Guess what happens now that we have these massive data centers? One of the bottlenecks is how to deliver power to these giant chips. One of the challenges is how these chips can communicate on a super high speed of the terabits per second and so on.”