Intel’s CEO Pat Gelsinger has developed a reputation for plainly criticizing tech companies, and that’s not limited to giants like Nvidia or AMD but also includes Intel itself. In an interview with Digit, Gelsinger spoke frankly about what he felt were Intel’s three greatest failures: the company’s failed smartphone business, the cancellation of an early AI-oriented GPU, and the lack of focus on “building a great foundry.”
Certainly, one of Intel’s highest-profile business failures is its smartphone chips. The first Atom CPUs were announced in 2008, and they were largely designed for smartphones, which were dominated by ARM chips. Momentum on Atom fizzled very quickly, with the first Intel-powered smartphone launching in 2012, and the smartphone venture was ultimately canceled in 2016 after very little had been accomplished.
Gelsinger also had an issue with the company canceling Larrabee in 2010, which would have been an early general-purpose GPU. The program was famously axed in the wake of Gelsinger’s exit from the company. Intel had planned to launch Larrabee for both the consumer market and HPC, but first, the consumer plans were scrapped, and then the whole GPU program was axed. Xeon Phi picked up where Larrabee left off, though it wasn’t nearly as ambitious or had the impact we’ve seen with GPUs. Needless to say, if Intel had built out a successful GPU program, it wouldn’t be at such a disadvantage to Nvidia in the current AI arms race.
Speaking of which, Gelsinger also criticized Intel’s acquisition of five AI companies, which he claimed were unnecessary if Intel had just launched Larrabee. He didn’t call out any specific purchases, though if we had to guess, he probably wasn’t a fan of Intel buying Nervana in 2016 just to cancel all of its products four years later. Intel’s $500 million acquisition of DeepLabs in 2014 also hasn’t seemed to meet with much success.
There are a few apparent wins, though: Intel also purchased Habana Labs for $2 billion in 2019, and that lives on through the company’s Gaudi products, serving as a lynchpin for the company’s current AI roadmap — it isn’t likely that Gelsinger chalks that acquisition up as a loss. Intel also spent $650 million for Granulate in 2022, but it’s probably too early to classify that as either a win or loss.
In the Digit interview, Gelsinger had the most to say about Intel’s foundry business and how the company was “fundamentally biased to building a great foundry.” Gelsinger’s underlying thoughts on this matter aren’t entirely clear, but this could indicate that Intel was focused on foundry ahead of all other pursuits.
Gelsinger is still focused heavily on the foundry operations; it is now opening up its fabs to companies that want to create high-performance chips on Intel’s cutting-edge process technologies. That’s essentially the business model of TSMC and Samsung, the world’s other two cutting-edge foundries. Intel had previously kept its cutting-edge nodes to use for its own processors, but Gelsinger has argued in the past that was a mistake.
By opening its fabs to other companies, even rivals like Nvidia, Intel could still at least make money through fabbing. That takes business away from TSMC and Samsung and boosts the amount of capital that Intel can invest in both manufacturing and R&D. In fact, that’s the entire idea behind the company’s new IDM 2.0 initiative that aims to open up Intel’s fabs to external customers.
Gelsinger took care to point out that taking advantage of nascent trends was difficult, saying, “You don’t get them all right” and that he would “bring us back into position on those.” Now that he is at the helm, Gelsinger has charted a new course for Intel that now appears to be bearing fruit. Only time will tell if the ambitious turnaround plan includes catching the next big wave.