In early 2026, the competition between Intel 18A and TSMC Arizona focuses on Intel’s efforts to ramp up high-volume manufacturing with back-side power, while TSMC relies on its own proven 2NM node. Intel 18A is achieving yields of 60% to 75%, which are competitive for 2NM but still trail TSMC’s mature yields of over 90%.  

Intel-18A (Arizona Fab 52 and 62): 

• Status: now in high-volume manufacturing for Panther Lake to reach over 50% yield by mid-2026.  

• Advantages: Column uses backside power delivery, PowerVia, and RibbonFET GAA transistors, which are expected to improve performance and effectiveness.  

• This: lower early yields lead to higher costs per chip, which affects profit margins in early 2026.  

• Yields: Early production is showing yields of up to 65-75%. This is considered good for a new complex node but still below TSMC’s mature high-volume yields.  

TSMC (Arizona FAB): 

• Status: TSMC is currently focused on 4nm and 3nm production in Arizona; 2nm production is expected to ramp up, but the most advanced nodes are still available in Taiwan.  

• Advantage: TSMC has proven yield maturity of over 90% and benefits from a large, established ecosystem.  

• Challenges: TSMC faces geopolitical risks and depends heavily on advanced manufacturing outside the US.  

The Yield War in 2026: 

• Performance vs. Maturity: 18A delivers 2nm-level performance, but Intel still faces trust issues due to past delays.  

• Cost factor: Lower early yields for 18A mean higher costs for chips made in America. TSMC is likely to remain the top choice for maximum yield per wafer until Intel’s 18A yields improve.  

• Outlook: If 18-year yields reach their targets by late 2026, Intel could challenge TSMC’s dominance in the 2NM market, especially for US-based secure and specialized AI chips.  

Intel used to be the world’s largest semiconductor company. Still, its market value dropped sharply in recent years as it fell behind Taiwan’s Superconductor Manufacturing Company and spent billions trying to recover.  

Now Intel has started making large quantities of its new 18A chip node, which it believes will help the company recover.  

Another major chipmaker has entrusted Intel with manufacturing on the new node. Right now, Intel is its own main customer. Its long-awaited Core Ultra Series 3 PC processor, called Panther Lake, will be the first major product made with 18A and will launch in PCs in January.  

Let’s become an internal node for now, said Daniel Newman, CEO of Futurum Group. So many companies have made massive investments in TSMC to ensure yield and capacity, so they just will not make the switch just yet.  

Intel hopes to attract new customers with its new factory, Fab52, in Chandler, Arizona, which CNBC toured in November. About 50 miles from Phoenix, TSMC has also set up a new factory to make chips using 4-nanometer technology. TSMC’s most advanced 2NM chips are still only made in Taiwan.  

Intel’s 18A chip is similar to TSMC’s 2nm node in some respects, such as transistor density. However, as Intel fixes issues from past delays, some 18A wafers have defects, which lowers the number of usable chips per wafer, a measure known as Yield.  

Yields are always an issue at the advanced node. This is not an uncommon problem, said Harvard Business School professor David Yoffie, who served on Intel’s board from 1989 to 2018. He pointed to early yield issues with Nvidia’s Blackwell GPUs at TSMC, which were quickly resolved.  

Intel again began focusing on making chips for other companies when Pat Gelsinger became CEO in 2021. Gelsinger left the company last December and was replaced by Lip-Bu Tan in March.  

For the past several years, the company has invested too much, too soon, without adequate demand, Tan wrote in a July memo.  

While Intel waits for a major outside customer, the US government invested in the company in August, taking a 10% stake with $8.9 billion. Most of this money comes from grants under the CHIPS Act, which President Biden signed in 2022.  

A few days before Softbank invested $2B in Intel. In September, Nvidia invested $5B and agreed to use some Intel technology but did not commit to using Intel’s foundry.  

Here’s a look inside Intel’s new chip factory, where the company hopes to attract major foundry customers and regain its standing.  

Fall of a Giant 

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

From the late 1970s to the early 2000s, Intel quickly introduced more advanced process nodes, leading to the term Moore’s Law, which states that the number of components on a chip doubles approximately every 2 years.  

However, Intel missed out on the mobile revolution, including turning down a deal to make Apple’s processors for the first iPhone. It also struggled to keep up with AI.  

In 2024, Intel had its worst year ever, losing about 60% of its value. This drop followed years of delays for its 10nm and 7nm chip nodes. Some analysts think these delays happened because Intel chose not to use ASML’s expensive ultraviolet lithography machines earlier.  

We lost the discipline of cycle time, said Jim Johnson, head of client computing, who joined Intel over 30 years ago. Cycle time requires you to commit and deliver, and we started telling ourselves we could have longer cycle times and try to do more.  

As it hustles to get back on track, Intel told CNBC there will be at least 15 EUV machines in Fab 52.  

In 2021, TSMC led the industry in chip technology, and Intel started outsourcing some advanced chip production to the Taiwanese company. At the same time, Apple switched Macs from Intel chips to its own M-series chips, which are also made mainly by TSMC.  

In his earlier stint at Intel, more than a decade before joining as CEO, Gelsinger was given the responsibility to build a GPU to compete with NVIDIA, Yoffie said. Unfortunately, the project failed, and that ultimately meant we did not play an important role in the AI revolution, he said.  

Intel is said to be considering buying the custom AI chip design start-up SambaNova for $1.6B, but the company has not commented on the possible deal.  

Changing our Culture 

During Gelsinger’s time as CEO, Intel focused heavily on chip manufacturing. His ambitious plan aimed to help Intel catch up to TSMC by launching five new chip nodes in four years.  

Now Tan serves as CEO, and Naga Chandrasekaran leads the Foundry Division.  

We are improving yield and reducing bug density month over month and hitting our goals, Chandrasekaran told CNBC in an interview in November. So we have turned the corner.  

Chandrasekharan joined Intel last year after spending 20 years at Micron, a top memory maker. His main goal now is to attract new foundry customers.  

I have to become part of their team and convince them that they can trust Intel, ” Chandrasekaran said—that’snumber 1, and to do that, we are changing our culture. We are bringing a huge execution focus internally to Intel Foundry.  

San Jose Current told CNBC that Fab 52 can handle over 10,000 18A wafer starts each week in Arizona. Intel has more than a million square feet of clean room space, with five factories linked by 30 miles of overhead tracks that move wafers between them. A sixth factory should be ready around 2028.  

The 18A process uses Intel’s RibbonFET, a get-all-around design that improves power control by surrounding the transistor. Earlier designs only touched the top and sides. Naga Chandrasekaran said 18A delivers more than a 15% performance-per-watt improvement compared to Intel 3.  

One of Intel’s main strengths is advanced packaging, which entails assembling and connecting chips to use in actual systems.  

CNBC visited Intel’s Advanced Packaging Lab in Chandler to observe operations such as sealing chips with a polymer and using a liquid to detect defects. Yoffe said Intel’s advanced packaging can help mitigate some power consumption problems.  

One of the biggest problems today for everyone making chips for data centers is the power they consume, Yoffie said.  

Chandrasekaran said the Arizona factory runs on nearly 100% renewable energy. In 2024, Intel’s Arizona sites used over 3 billion gallons of water and returned 2.4 billion gallons to the local supply through an on-site recycling plant.  

No Blank Checks 

Tan has made it clear to employees that there will be no more blank checks for future foundry projects. The company needs to secure customers.  

Intel’s large new chip factory in Ohio is now delayed until at least 2030. TAN has also cut costs by reducing the workforce by 15% in July and cancelling projects in Germany and Poland.  

That’s what the company needed, said Newman of Futurum. It needed to be faster, leaner, and more focused. It needed someone who could be more shrewd.  

Intel is waiting to see how demand develops before sharing more details about Intel’s new chip node 14A. Chandrasekaran told CNBC that 14A will first be developed in Oregon with large-scale production starting in 2028. Customers for 18A won’t be easy to find. Unlike TSMC, which only makes chips for outside clients, Intel also makes devices powered by its chips, casting it as a competitor to some of the customers it hopes to win.  

If I were an NVIDIA, AMD, Qualcomm, or Broadcom, would you really want to put your secret sauce into a manufacturing operation where you are giving Intel access to it? Yoffie said.  

He suggests that Intel should spin off its foundry business into a separate company.  

You actually separated the two. I think you’d give Intel a much better shot at being successful. Yoffie said. And you’d also give the United States a much stronger position as the home of a major semiconductor manufacturing organization. For now, Intel hopes Panther Lake will serve as strong evidence of its progress when it launches in pieces from major brands like Samsung, Dell, HP, Lenovo, Asus, and Acer in January. The company’s next data center chip, the Xeon 6+, is also built on 18A.  

If you are a major company that wants to bet on a process node, you are going to feel a lot more comfortable if you see Intel ramping the heart of their client product line to high volume on that process node, Johnson said.  

Microsoft and Amazon signed early deals last year, committing to use Intel’s Foundry for some of their in-house custom chips.  

It’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 thinking about manufacturing chips at Intel, and one analyst predicts Apple might make some Mac chips at Intel again by 2027.  

Meanwhile, Intel received a boost when the US government took a 10% stake in the company.  

It shows the U.S. government’s confidence in Intel and the belief that we need leading-edge R&D and manufacturing on U.S. soil, Chandrasekaran said.  

Government investment came days after President Donald Trump called TAN to resign, then reversed course.  

Worry sometimes about the scope creep here and how the US could decide to take stakes in all kinds of things, Neumann said. But you have industries we have let leave the US to an extent that puts us at an indefensible risk, and we need to bring them back.  

Some 92% of the world’s most advanced chips are made in Taiwan, following decades of decline in US chip production.  

The stakes are incredibly high for Intel for the US and for the world. Yoffe said: “The whole idea that the world’s most advanced products are dependent on a single location on an island a few miles off the Chinese coast is a terrible situation for the whole world to have to deal with.”  

Chandrasekaran, for his part, is devoted to turning Intel into a manufacturer of advanced chips.  

As a satellite conductor community, we have to enable this solution for the world to move forward with AI. He said there’s no other option than to be successful.

Source: Inside Intel’s new Arizona fab, where the chipmaker’s fate hangs in the balance