Intel faces near-death, as analysts ponder early AI miss and fab’s future

A month has passed since Intel announced it would lay off 15,000 workers, igniting a massive 30% drop in its share price and a 57% decline for all of 2024. The silicon industry stalwart seems to be in free fall.

Things have gotten so bad that the company is now reportedly working with Morgan Stanley to defend itself against activist investors, something that has happened to Intel in 2020 before Pat Gelsinger became CEO a year later. In that earlier example, hedge fund Third Point bought a large stake in Intel and pushed for new strategic directions. 

Of interest, Lip-Bu Tan resigned from Intel’s board on August 19 citing personal reasons after serving just two years there. Still, he said he remains “supportive of the company and its important work.” 

Important work, indeed.

Even if things get much worse at Intel, much of the industry sees Intel as the trusty old horse that is still doing vital work, especially for PCs and laptops (80% of that market) and Intel Xeon CPUs for 70% of data centers. 

That work is on the product side, but Intel also is on a path to becoming a chip foundry for the world, as well as for its own designs. The foundry work is considered so vital that Intel won a preliminary grant for more than $8 billion from the US CHIPS Act, making it the largest grant recipient. 

The grant would support Intel’s ability to build massive fabs in Arizona and Ohio, part of the nation’s initiative to become more chip self-sufficient. A big factor is 90% of all the world’s advanced chip nodes are built by TSMC, based in Taiwan, putting the US and much of the world at risk should China attack Taiwan.

Gelsinger gets much of the credit for pushing hard for US CHIPS Act funds, and, indeed, the strategic direction of getting Intel into the fab business to build its own chips and those of other companies. To get there, he has pushed a cycle of new chip nodes—five nodes in four years—meant to entice new customers. If a node falls behind in the complex production schedule, that’s when the investor activists always take notice. Almost immediately a report of a delay will hurt a chipmaker’s stock.

Early signs at Intel came in April

The 15,000 layoffs announced on August 1 were hinted at as early as April. At that point, Gelsinger announced a $7 billion operating loss for the fab unit in 2023, steeper than the $5 billion loss the year before. Gelsinger saw the clouds clearing for the fab with break even on an operating basis by about 2027.

In effect, the Intel fab strategy is meant to be the company’s saving grace, but it will take three more years to make a difference to Intel’s bottom line.

That three years gives an investor activist plenty of time to act, but Intel’s strategy is still going to remain essential. “Semiconductor needs are so critical to the world that they have to be built in more places” including by Intel, said Bob O’Donnell, chief analyst at Technalysis Research, in an interview with Fierce Electronics.

He added: “I don’t think we’re going to see Intel go away and Gelsinger is still the right guy for the job. Intel has a future that’s a little out there that we are still counting on. Wall Street wants something now of course, or next quarter, but we don’t yet have an answer. Right now, Intel is an absolutely essential cog in the industry, and we don’t see that changing any time soon.”

Recent progress on Intel’s 18A node is encouraging, O ’Donnell and several industry analysts told Fierce Electronics. The company recently announced its first two 18A products are out of the fab and have powered on. AI PC client processor, Panther Lake and server processer, Clearwater Forest, are on track to start production in 2025, with the first external customer expected to tape out on Intel 18A in the first half of next year. 

Intel 18A is one of the final nodes from Intel’s fabled ‘five nodes in four years’ road map. It is the first with PowerVia, a power delivery system, and the RibbonFET gate-all-around transistor architecture. It also uses the Intel open system foundry approach. Intel believes it represents “a significant step in bringing Intel back to process leadership and industry-leading innovation.”

In July, Intel released its 18A process design kit to allow its partners to update their tools and designs.  The 18A node could be used for many designs, including GPUs, CPUs, FPGAs and more. Intel has advantages in packaging that could help a customer design a System on Chip with possibly five components built on different process technologies, then stitched together, O’Donnell noted.

A lot is riding on 18A.

“Five nodes in four years was very ambitious. The stark reality is 18A is likely going to be the only competitive processor. Intel has to start at 18A and 20A,” said Mario Morales, group vice president of enabling tech and semiconductors at IDC. 

The foundry business has received a lot of investment but “it needs scale and needs customers to be signed,” Morales said. “Intel has hinted of a new customer in 2025 and I’m watching to see if they are gaining ground, but they don’t have flagship customers.”

What’s gone wrong at Intel?

Several analysts interviewed by Fierce Electronics fingered a number of problems hurting Intel, many going back to the days before Gelsinger took the helm. Their common refrain: decisions to focus more heavily on AI or explicitly GPUs and AI accelerators should have been made years ago. 

Cutbacks on investments by Gelsinger's predecessor Bob Swan didn’t help, “and we’re still seeing the remnants of that,” Morales added. “A challenge the company faces is being ingrained and insular and it has gotten complacent and forgotten how to execute as before. Listen to how Intel talks about AI now and has trouble articulating its role.” 

Intel workers are now like ‘deer caught in headlights’

The result, since August 1, is workers face voluntary layoffs which will move to involuntary layoffs in mid-November. “Intel is in a lot of pain right now, and most employees don’t know what to do—they’re like deer caught in the headlights—until focus is resumed,” Morales said.

While the hope of the foundry business is great, both the product side and foundry side “need time to get mean and leaner,” Morales added. Part of the value of a foundry business is giving the product side time to evaluate new designs, but that approach has meant Intel is “not really building to be competitive. They have built for industry leadership and not volume, which tells a lot about what it has to do if it wants to compete with TSMC. They have to serve a lot more customers and process technology cannot push just performance levels but they need to address companies in mobile and with balanced performance.”

As a result, data center AI chips “are where AMD is beating Intel up,” while Nvidia is “playing the right long-term play,” Morales said.  Intel’s GPU/accelerator offering has come out “a lot worse than what folks expected.”

Investors worry, but Intel failure is not an option

“Investors may be losing patience with Intel’s plans to reconfigure its market position, but Gelsinger is actually sticking reasonably close to the plan, considering the massive challenge of competing with AMD on CPUs, Nvidia on GPUs, and TSMC on chip fabrication,” said Karl Freund, founder and principal analyst at Cambrian-AI.

“Will Intel succeed? I don’t think the US can let them fail, especially in terms of becoming a competitive fab partner to US industry.”

The next two years will decide.

RELATED: Is Intel too big to fail? Here’s what analysts say

“Intel is going to have a very tough next couple of years,” said Jeff Koch, analyst at SemiAnalysis. “Both 18A and 14A are critical to the company. 18A for foundry customers to test the waters with less crucial designs and 14A to potentially win flagship designs. If either process does not deliver, it could end Intel's ability to fund leading-edge development, at least without outside intervention from the US government.”

(To explain: Intel’s nomenclature uses numbers for chip nodes that are not easily aligned with competitors’ nanometer measurements. The lower the number, the smaller the node. Intel partly took this tack with nomenclature as it enters the angstrom era, with an angstrom at 0.1 nanometers in size. With 14A, Intel is targeting 2027 for production where it will use high numerical aperture lithography in production to allow circuits with less than 10 nanometers and even smaller to be imaged on a microchip.)

“Intel would be in a better spot if they had a competitive AI accelerator chip, but that is really just one of the many after effects of acute mismanagement during the pre-Gelsinger era when financial engineering took precedence over process technology,” Koch added.  “Now we are seeing the resulting long-term pain.”

Koch is still supportive, however. “I applaud Gelsinger and his team for course correcting when they took over in 2021 but it will take years to turn around and the company may have a near-death experience in the meantime.”

If the worst hits Intel’s fab capability down the line, the world’s electronics device makers face a supply crisis for many versions of chips. “Keep it mind that Intel is an Integrated Device Manufacturer and manufactures its own chips in addition to those of outside customers,” said Patrick Moorhead , chief analyst at Moor Insights & Strategy.

The world’s other largest fabs, TSMC and Samsung, “don’t have enough capacity to support a total Intel meltdown," he added.

Gelsinger seems determined to ward off talk of a meltdown.