When veteran engineer and executive Pat Gelsinger returned to Intel as CEO in 2021, the once-great chipmaker was in decline. After failing to adapt to the mobile era and then missing out on several steps in modern microprocessor manufacturing, it was now lagging behind in supplying chips to meet the growing appetite of the tech industry. Artificial intelligence.

With optimism that seemed reckless at times, Gelsinger promised that Intel would make an epic comeback. He vowed to shake up its dormant corporate culture, refocus on core engineering, and deliver a revamped manufacturing plan that would make rivals TSMC and Samsung take notice.

This week, Gelsinger declared Intel’s comeback plan well and truly on track.. He announced a rebrand of the company’s “foundry” business, which produces chips designed by other companies, saying that Intel’s advanced manufacturing process could produce silicon chips as efficient as TSMC later this year. And will enable. Microsoft is the first major customer for this new chipmaking technology—a key coup for Intel as it tries to convince the industry it can offer competitive products fit for the age of AI. .

Pat Gelsinger spoke with Wired senior writer Will Knight about Intel’s AI reboot on Zoom from his home in Santa Clara, California. The conversation has been lightly edited for length and clarity.

Will Knight: You announced this week that Intel will relaunch its business that makes chips for other companies as the “AI-era System Foundry.” what is the meaning of this?

Pat Gelsinger: I started Intel’s strategy two years ago, and for the company, creative AI has been this unexpected growth. This has been the land. NvidiaBut we are a company that has an opportunity to capture 100 percent of the AI ​​market. We know how to connect networks and memory. [provide] Supply chains and all these other elements that we’re finding consumers very excited to take advantage of.

Speaking of AI additions, what did you make? reports suggest. OpenAI CEO Sam Altman wants to raise $7 trillion to develop and manufacture the chips necessary to guarantee advances in AI?

My first reaction was that it was a big number and then I had to do the math. Today, the largest AI models were built on approximately 10,000 GPUs. I believe we probably need to be 10 million for the largest AI models to be developed in the future.

We are already saying that today we can spend a few billion dollars on training the most advanced models. Also, the $7 trillion math includes power and data centers.

This week you said that Intel is on track to deliver its new “18A” manufacturing process, which will rival TSMC’s best offerings. What else are you doing to get ahead?

The entire industry is chasing this next-generation transistor, which we call Ribbon FET. I think everyone [asking] Who is going to develop the next best transistor on the planet?

But what everyone is giving us credit for is backside power, this new way of delivering power to the device, which gives you better current resistance performance, but it’s also improving chip density. . This means that the same wafer, instead of producing 100 chips, can produce 120 chips. This is a great value proposition.

You announced Microsoft as a customer of your Foundry business. But Intel has previously fallen behind the competition in this market. How will you convince customers that things are different this time?