Intel’s years of missteps leave it fighting for survival in the Nvidia-dominated AI era

Henry V had Agincourt. Gen. Robert E. Lee had Gettysburg. And Intel (INTC) CEO Pat Gelsinger has Chandler. That would be Chandler, Ariz., outside Phoenix, where Intel is investing nearly $30 billion to build two state-of-the-art semiconductor plants, or fabs, that will be the first to use the company’s newest chipmaking process. It’s here, in Chandler, where Gelsinger’s fate—and likely that of the company he leads—will be decided.

Gelsinger, who was named CEO in 2021, has essentially bet the company on 18A, a new chipmaking process. He hopes it will position Intel as a viable alternative to Taiwan Semiconductor Manufacturing Co. (TSMC), the world’s leading contract manufacturer of chips.

The reason for Intel’s struggles is clear: It fell victim to a classic innovator’s dilemma—not once but twice. First, early in the 21st century, its preoccupation with producing chips for PCs and data centers led it to miss the smartphone revolution. Then, in the past decade, it missed the emergence of chips designed for artificial intelligence.

Intel rival Nvidia took a type of chip originally designed for the demands of video games, the graphics processing unit (or GPU), and turned it into the workhorse for training and running AI models. Now the generative-AI boom has made Nvidia one of the world’s most valuable companies, worth more than $3 trillion, compared with Intel’s relatively paltry $84 billion.

Gelsinger is racing to reverse Intel’s slide by repositioning the company around manufacturing excellence, while also trying to establish Intel as a player in the market for AI chips. Many are skeptical he can pull it off and fear the company may be in permanent decline.

Intel wound up in such dire straits owing to missteps in its core business for central processing units (CPUs), in which it was once the unrivaled king. Production delays and problems in its own fab facilities have let rival AMD steal significant market share.

Distracted while trying to fix these issues, Intel failed to see the extent to which graphics chips would come to dominate the market for AI. Instead, it thought AI would be run on systems that still had CPUs at their heart.

“The strategy has been to fix the core business and don’t worry about the ancillary stuff,” says Alan Priestley, a Gartner vice president analyst. “GPUs were the ancillary stuff.”

Even after Intel belatedly recognized the fast-growing market for AI-specific chips, it bungled its efforts to get into the AI game. In 2019, it announced its own GPU design for AI called Ponte Vecchio. But the design was complicated, requiring three different fab processes to make, and expensive. Worse, its performance couldn’t match that of Nvidia’s chips. This year, Gelsinger shelved Ponte Vecchio in favor of a new design due in 2025.

Story continues

Intel also bought AI chip startups but struggled to turn their products into blockbusters. In 2016, it purchased Nervana for $350 million, but took years to roll out chips based on its technology. When it did, they had already been eclipsed by Nvidia. Then in 2019, it bought Israeli AI chip startup Habana Labs for $2 billion. Intel says Habana’s chips beat Nvidia’s. But these claims have not been independently verified, and Intel expects only $500 million in sales from Habana’s latest chips this year, compared with Nvidia’s tens of billions of dollars in revenue.

Intel did not respond to a request to make Gelsinger or other executives available to comment for this article. But Gelsinger publicly acknowledges the impossibility of displacing Nvidia’s dominance in chips for training large AI models anytime soon. “In that race, you know, [Nvidia] are so far ahead,” the CEO said at a technology conference in August.

Instead, Gelsinger thinks Intel can compete in the market for AI inference— running AI models that have already been trained. He also says another new Intel chip will be a hit for use in laptops and PCs that run AI applications.

But Intel’s board has reportedly voiced concerns about Gelsinger’s AI strategy not being aggressive enough to ensure Intel will play a significant role in what is now the fastest-growing market for chips.

Gelsinger’s immediate problem is that Intel is hemorrhaging cash. The 18A gamble is expensive. The company has committed $185 billion to build new fabs and upgrade existing ones.

View this interactive chart on Fortune.com

Meanwhile, costs for Intel’s planned Arizona fabs and two more in Ohio have soared past initial projections. The Arizona plants, which Intel is counting on having online in 2025, have been hit by construction delays. And in September, Reuters reported that Broadcom, which makes networking and radio chips, had tested Intel’s process and concluded it was not yet ready for full production.

Amid this bad news, Intel’s foundry business has notched a few precious wins: In September, Amazon’s AWS agreed to manufacture a next-generation AI chip at Intel’s 18A fabs. Intel previously struck a similar deal with Microsoft.

Finding the money to pay for 18A, though, is rapidly becoming an existential crisis. Intel’s annual sales have flagged—down $24 billion, or 30%, since Gelsinger took over. In 2022, the company’s free cash flow turned negative. It has worsened since then. As of late June, Intel was burning through $12.6 billion more cash than it was taking in on an annual basis. Investors have reacted to the cascade of troubles by punishing Intel’s stock, driving its shares down almost 60% so far this year.

To reassure Wall Street, Gelsinger has been forced into painful decisions. In August, he announced $10 billion in cost cutting that included laying off 15,000 employees—15% of Intel’s workforce. He also reduced capital spending by $5 billion and suspended Intel’s dividend.

Then, in September, he announced more radical action: Intel’s foundry business will be formally spun off as a separate subsidiary. The hope is that this will reassure potential foundry customers that their design secrets won’t leak to Intel’s product division and that Intel won’t prioritize its own production needs over theirs. Critically, it will also let the foundry business raise capital from outside investors, possibly alleviating Intel’s cash woes. Gelsinger also shelved plans for a new $32 billion fab in Germany, as well as new facilities in Poland.

The company has turned to the U.S. government for help, too. Intel has been counting on money from the U.S. CHIPS Act, signed by President Biden to boost domestic chip production, to assist in paying for some of its new 18A fabs. In March it was awarded $8.5 billion in direct CHIPS funding and $11 billion in loans. But the money is tied to Intel hitting certain construction milestones, and it has yet to receive any funds.

At some point Gelsinger may be forced to choose between preserving Intel’s core chip-design division—and selling off the foundry business entirely, an admission that his 18A strategy has failed. Whatever Intel ultimately does, drastic measures are likely necessary if Gelsinger still hopes to snatch victory from the jaws of defeat.

This article appears in the October/November 2024 issue of Fortune with the headline “A wounded Intel battles for survival in the AI era.”

This story was originally featured on Fortune.com