Intel aims to find clients and catch TSMC with new chip fab in Arizona

Intel was once the world’s largest semiconductor company, but its market value has plummeted in recent years as the chipmaker lagged behind Taiwan Semiconductor Manufacturing Co. and spent billions of dollars to catch up.

Now Intel has entered high-volume production of the 18A, its new chip node that it says will turn things around.

The biggest problem? Convincing a major chipmaker to trust Intel with manufacturing on the new node. For now, Intel’s only major customer is itself. The company’s long-awaited Core Ultra series 3 PC processor, codenamed Panther Lake, will come to PCs in January as the first major product produced on 18A.

“It’s become an internal node for now,” Futurum Group CEO Daniel Newman said. “Many companies have invested so heavily in TSMC to guarantee efficiency and guarantee capacity plates that they are not going to switch yet.”

Intel is pinning its hopes of attracting customers on Fab52, its new chip manufacturing facility in Chandler, Arizona, which CNBC took an exclusive on-camera tour of in November. About 50 miles north, in Phoenix, TSMC also has a new factory where it produces chips with 4-nanometer technology. The most advanced 2nm technology is currently produced only in Taiwan.

Intel’s 18A is generally on par with TSMC’s 2nm node in some metrics, such as transistor density. But as Intel worked to fix the problems after years of delays on previous nodes, defects appeared in some 18A wafers, resulting in fewer usable chips per wafer, commonly referred to as yield.

“Yields are always an issue in the advanced node. It’s not an uncommon issue,” said David Yoffie, a Harvard Business School professor who served on Intel’s board of directors from 1989 to 2018. early yield problems with NvidiaBlackwell GPUs at TSMC were quickly fixed.

Intel’s renewed focus on foundry (chip production for external customers) comes as Pat Gelsinger takes over as CEO in 2021. Gelsinger was fired last December and replaced by Lip-Bu Tan in March.

“Over the last few years, the company invested too much, too early, without sufficient demand,” Tan said. memory in July.

With Intel eyeing a major outside customer, the US government stepped in in August and took a 10% stake in the company with an $8.9 billion investment that came primarily from grants promised under the CHIPS Act signed by President Joe Biden in 2022.

days ago, SoftBank It invested $2 billion in Intel, followed in September by a $5 billion investment from Nvidia, which agreed to use some of Intel’s technologies but did not commit to using its own foundry.

Here’s a behind-the-scenes look at Intel’s new chip factory, where it hopes to find big foundry customers and pay them off.

The fall of a giant

Founded in 1968 by Silicon Valley chip pioneers Robert Noyce and Gordon Moore and legendary investor Arthur Rock, Intel introduced the world’s first commercially available microprocessor just three years later.

From the late 1970s to the early 2000s, Intel pumped out increasingly advanced process nodes at a rapid pace, leading to the term “Moore’s Law”; This meant that the components on a chip doubled every few years.

“The 1990s were a time of curiosity and excitement for Intel,” Yoffie said. “We were the largest semiconductor company in the world, the most profitable semiconductor company in the world.”

But Intel largely missed the mobile revolution reject a deal Making Apple’s original iPhone processors. Then there was a smell in the AI.

In 2024, Intel had its worst year ever, losing nearly 60% of its value. The decline follows two of the previous chip nodes. 10nm And 7nmIt was delayed for several years. Analysts say the delays may have been caused by an earlier choice to delay using ASML’s expensive Extreme Ultraviolet Lithography machines.

“I think we’ve lost cycle time discipline,” said Jim Johnson, chief customer information officer who joined Intel more than 30 years ago. “Cycle time requires you to commit and deliver, and so we started talking to ourselves: we could have longer cycle times and try to lift more or do more.”

As Intel struggles to get back on track, it told CNBC that Fab52 will have at least 15 EUV machines.

By 2021, TSMC became the node leader and Intel outsourcing Some pioneering chip manufacturing for the Taiwanese giant. Around the same time, Apple It began replacing Intel chips in Mac computers with its own M series chips, also produced by TSMC.

In his previous role at Intel, more than a decade before he rejoined as CEO, Gelsinger was “given the responsibility of building a GPU that could compete with Nvidia,” Yoffie said. “Unfortunately, this project failed, which ultimately meant that we were unable to play a significant role in the AI ​​revolution.”

Intel may now considering a deal will acquire private artificial intelligence chip design startup SambaNova for $1.6 billion, but the company declined to comment on the matter.

‘We are changing our culture’

The hallmark of Gelsinger’s tenure as CEO was Intel’s focus on chip production. His ambitious roadmap led Intel to return to TSMC. five knots in four years.

Now Tan is the CEO and Naga Chandrasekaran is in charge of the foundry.

“Every month we’re making improvements in yield, we’re making improvements in defect density, and we’re hitting our targets,” Chandrasekaran told CNBC in an interview in November. he said. “That’s why I believe we’ve turned the corner.”

Chandrasekaran joined Intel last year after more than two decades as a leading memory maker Micron. He said his biggest goal is to find foundry customers.

“I need to be a part of them and convince them that they can trust Intel Foundry,” said Chandrasekaran. “That’s number one. And we’re changing our culture to do that. We’re bringing a big application focus internally to Intel Foundry.”

Chandrasekaran told CNBC that Fab52 has the capacity to launch more than 10,000 18A wafers per week. There are more than a million square feet of cleanroom manufacturing space in Arizona, and all five factories are connected by 30 miles of rails that move sheets between them. The sixth factory, Fab62, is expected to be ready around 2028.

The 18A also uses RibbonFet, Intel’s all-around gate architecture that improves power control by completely surrounding the transistor, unlike previous designs that only touched the top and sides. Chandrasekaran said the 18A offers “more than a 15% performance increase per watt” over the Intel 3.

Perhaps the biggest feature where Intel stands out is advanced packaging, the assembly and connectivity of chips into end systems where they appear in real-world applications..

CNBC went to Intel’s advanced packaging lab in Chandler to see several steps in the process, such as protecting the chips with a polymer-based seal and exposing them to a liquid that detects any defects. Intel’s advanced packaging “could help reduce some power consumption issues,” Yoffie said.

“One of the biggest problems for anyone making chips for data centers today is the power it consumes,” Yoffie said.

The Arizona plant runs on almost 100% renewable energy, Chandrasekaran said. When it comes to water, Intel’s Arizona facilities used more than 3 billion gallons of water in 2024 and returned 2.4 billion gallons to local resources through its on-site water recycling facility.

‘No blank cheque’

Tan’s message It is clear to employees when it comes to future expenses on foundry nodes: “No more blank checks.” The company needs customers.

Intel’s massive new Ohio chip factory has been delayed until at least 2030, and Tan made major cost cuts in July, cutting 15% of the workforce and canceling projects in Germany and Poland.

“This was what the company needed,” said Futurum’s Newman. “He needed to be faster. He needed to be leaner. He needed to be more focused. He needed someone who was a little more crafty.”

Tan is waiting to see how demand shapes up before revealing concrete details about Intel’s next node, 14A. It will be developed first in Oregon, with a target of volume production in 2028, Chandrasekaran told CNBC.

Finding customers for 18A will not be easy. Unlike TSMC, which only produces chips for outside customers, Intel also produces devices powered by its own chips, positioning it as a rival to some of the customers it hopes to win over.

“If I’m an Nvidia or AMD or Qualcomm or broadcomDo you really want to put your secret sauce into a manufacturing operation where you allow Intel access to that secret sauce?,” Yoffie said.

He suggests splitting the foundry into a different company.

“If you really separate the two, you give Intel a much better chance of succeeding,” Yoffie said. “You would also give the United States a much stronger position as the home of a major semiconductor manufacturing organization.”

For now, Intel is hoping Panther Lake will be a big proving point when it launches on PCs from major companies like SAMSUNGDell, HP, LenovoAsus and Acer in January. Intel’s next data center chip, the Xeon 6+, is also built at 18A.

“If you’re a large company looking to bet on a process node, you’ll feel much more comfortable if you see Intel pushing the heart of its customer product line to high volume on that process node,” Johnson said.

Microsoft And Amazon It signed early deals last year committing to use Intel’s foundry for some in-house custom chips.

“That’s a good sign, but of course their volumes are very small relative to Nvidia and other major chip companies,” Yoffie said.

Recent reports indicate that AMD is considering manufacturing at Intel and an analyst He predicts Apple could once again produce Mac chips at Intel by 2027.

Meanwhile, Intel received a lifeline with a 10% stake from the US government.

“This demonstrates the US government’s confidence in Intel and its belief that we should have leading R&D and manufacturing on US soil,” said Chandrasekaran.

Government investment comes days after President Donald Trump He called for Tan to resignthen reversed course.

“I worry sometimes about the expansion of scope here and how the United States might decide to take a cut of all kinds of things,” Newman said. “But there are industries that we have allowed to leave the United States to the extent that they put us at untenable risk, and we need to bring them back.”

Approximately 92% of the world’s most advanced chips are produced in Taiwan. Decades-long decline in percentage of chips produced in the USA

“The risks for Intel, the United States, and the world are incredibly high,” Yoffie said. “The idea that the world’s most advanced products are tied to a single location on an island a few miles off the coast of China is a terrible situation that the entire world must deal with.”

Chandrasekaran, on the other hand, is determined to turn Intel into an advanced chip manufacturer.

“As a semiconductor community, we must enable this solution for the world to move forward with AI,” he said. “There is no choice but to succeed.”

WRISTWATCH: Inside Arizona chip factory is the key to Intel’s salvation