How Micron is building the biggest chip fab in U.S. history despite a China ban and smartphone slump

Published

11 months ago

October 23, 2023

admin

Memory chips are at the center of all devices, helping store and access data in smartphones, computers and the servers training generative artificial intelligence models.

Just three companies make more than 90% of the world’s dynamic random-access memory, or DRAM, chips. With Samsung and SK Hynix both headquartered in South Korea, Idaho-based Micron is the only manufacturer in the U.S. — that has made it the latest target of China’s bans on U.S. technologies.

About a quarter of Micron’s revenue comes from China, and “about half that revenue is at risk,” Micron CEO Sanjay Mehrotra told CNBC in an interview.

Meanwhile, Micron is doubling down on U.S. manufacturing. Its current leading-edge chips are made in Japan and Taiwan, but Micron is aiming to bring advanced memory production to the U.S. starting in 2026 with a new $15 billion chip fabrication plant in Boise, Idaho. Micron celebrated its 45th anniversary in October by pouring the first cement at the new fab.

The facility is located next to Micron’s huge research and development facility, where CNBC got a behind-the-scenes tour.

Micron’s existing research and development facility in Boise, Idaho, shown here on Oct. 6, 2023.

Ben Farrar

“Memory is very cost-sensitive and we have to get economies of scale to mass produce our chips on a level that meets the market demands,” said Scott Gatzemeier, Micron’s corporate vice president of front end U.S. expansion.

DRAM and NAND memory chips are a cheaper type of semiconductor than the high-powered central processing units from Intel and AMD and graphics processing units that sparked Nvidia’s growth. But multiple memory chips are needed to support each GPU or CPU, so making memory requires more fab space.

That’s why Micron is planning the biggest chip project in U.S. history, spending $100 billion over 20 years to build four 600,000 square foot fabs in upstate New York.

Mehrotra told CNBC that Micron’s goal is to vastly increase the U.S. share of DRAM production, which he said currently sits at just 2%. That production comes from Micron’s fab in Manassas, Virginia. The company is getting assistance from the federal CHIPS and Science Act, which offers billions of dollars to incentivize domestic production.

“With Micron’s investments through CHIPS support in Boise, Idaho, as well as in Syracuse, New York, that 2% over the course of nearly 20 years will be changing to about 15% of the worldwide production coming from the U.S.,” Mehrotra said.

The U.S. share of overall chip manufacturing has plummeted from 37% to 12% in the last three decades, largely because it costs at least 20% more to build and operate a new fab in the U.S. than in Asia. Labor is also cheaper there, the supply chain is more accessible and government incentives have been far greater. That’s why the CHIPS and Science Act set aside $52.7 billion for companies that manufacture in the U.S.

Senate Majority Leader Chuck Schumer, D-N.Y., co-sponsored the bill.

“When it came to chips so essential to everything we do, we had lost that edge,” Schumer told CNBC in an interview. “And if we didn’t get back that edge, not just on chips but on science broadly, we would no longer be the No. 1 economic power in the world.”

Micron and at least 460 other companies have applied for funds from the CHIPS Act. States are also offering incentives to entice chip companies. Micron told CNBC it’s eligible for up to $5.5 billion from the state of New York for the four fabs it’s building just north of Syracuse. New York Gov. Kathy Hochul signed the state’s Green CHIPS Act into law last year.

“If they hadn’t passed the CHIPS and Science Act first, I don’t think it would have been as many incentives as necessary,” Hochul said. “I knew I had to woo them, talk about our incentives, but also we get out of it 50,000 jobs. That’s a good deal for us any day of the week.”

These promises come on the heels of a major price slump for memory chips, which led to layoffs at Micron and SK Hynix, and resulted in Samsung slashing production. Now, Micron is betting big that the memory market will grow.

“The large language learning models and other things like that continue to increase large demand,” Gatzemeier said.

“We’re now moving into things like FaceTime, higher resolution images, movies on demand,” he said. “All of that requires more and more memory to be made available.”

Micron says construction in New York will begin at the end of 2024 and chip production there will start in 2027. With both Idaho and New York fabs online, Mehrotra told CNBC that Micron plans to increase the share of chips it makes in the U.S. from 10% to nearly 60% in the next two decades.

Micron CEO Sanjay Mehrotra shows CNBC’s Katie Tarasov a 300mm silicon wafer at the memory company’s San Jose office on Oct. 2, 2023.

Kent Kessinger

‘Feast or famine’

Micron was founded in 1978 by three chip engineers, along with one of their twin brothers, in the basement of a dental office in Boise. By 1980, it was building its first fab and a year later was pumping out a revolutionarily small 64K DRAM chip. These chips, used for storing bits of data that can be quickly accessed by a CPU, ended up in many of the early PCs.

Gatzemeier, who joined as an intern in 1997, explained the two main kinds of memory: DRAM and NAND.

DRAM is “volatile memory, which means that when the power is removed, it loses all of its information. It’s very fast but has to be, and it sits near the CPU and it’s used for real-time processing,” he said. “NAND flash memory is what’s in your SSDs or your storage cards. And NAND flash is nonvolatile, meaning it’ll still store your memory even when the power’s removed.”

Micron went public in 1984. Memory was a crowded field, but over the years, it has whittled down to just three top players.

“The name of the game is high performance and low cost at the same time,” said Patrick Moorhead, CEO of Moor Insights and Strategy. “Otherwise, you’re going to be blasted out of the market.”

When it comes to the biggest type of memory, DRAM, Samsung is by far the leader, followed by SK Hynix and then Micron. Micron has made 11 acquisitions since 1998, including Texas Instruments‘ memory division, Numonyx, Elpida and Inotera.

“For a very long period, they had not invested in a new fab,” said Gaurav Gupta, an analyst at Gartner. “But they were still able to retain their market share by acquiring other smaller memory firms, which were either going out of business or bankrupt.”

Unlike many kinds of chips, memory wasn’t in short supply during the chip shortage. Micron and its competitors saw a major upswing in the pandemic-fueled boom in consumer electronics. Micron’s profits then fell significantly due to weakened demand for PCs and smartphones and a chip oversupply that led to lower prices. It’s a downturn that has affected much of the chip industry.

“When I look at this market over the past 30 years, it’s always feast or famine,” Moorhead said. “We have an oversupply now. But guess what? Give it a couple of months and we will be in an undersupply and prices will go up.”

Even amid the downturn, Mehrotra is optimistic about the growth of Micron’s smartphone business. It supplies memory in phones from Apple, Motorola, Asus and more.

“The mix of smartphones is going more and more toward higher-end smartphones, toward the flagship smartphones, which require more memory as well,” Mehrotra said. “When we look ahead at 2024, we actually expect that year-over-year total worldwide smartphone unit sales will increase.”

Micron is also focused on rapid growth markets such as automotive and AI. The next generation of its most advanced product, High Bandwidth Memory, is set for volume production next year. HBM helps AI models such as ChatGPT remember past conversations and user preferences to generate more humanlike responses.

“It is able to pack 50% more memory capacity in a memory cube,” Mehrotra said. “It is able to give you 50% faster performance and is able to give you about 2.5 times better power and performance efficiency. And these are all the elements that are critically important in AI applications.”

Banned in China

Micron is facing one major specific challenge. In May, China’s cybersecurity administration banned some of its sales to key China infrastructure projects, saying it failed a security review. Last year, the U.S. barred chip companies from supplying China with certain key technologies.

“Micron is absolutely just a pawn in this game right now,” Moorhead said. “They weren’t the first and they were not the last.”

Mehrotra offers a more diplomatic approach.

“It’s very important for U.S. and China to provide an environment to the businesses so that they can invest in a predictable manner,” he said. “And what I can also tell you is that Micron, of course, is totally committed to bringing the value of its technology and products and manufacturing scale to the benefit of our customers across various end markets in China.”

Meanwhile, Micron has started construction on a $2.75 billion assembly and test facility in India.

“Micron is obviously trying to diversify its base,” Gartner’s Gupta said. “It has testing and packaging facilities in China. And obviously they are trying to move, diversify out of China.”

China can still rely on chips from Samsung, SK Hynix and smaller Chinese memory makers. That’s because memory is considered a commodity, meaning it’s relatively easy to switch between products from different companies. But that’s not guaranteed to last.

“When we get back to the boom days and Hynix and Samsung can’t fulfill all the volumes, you might see China diving back into Micron and suddenly lifting any restrictions,” Moorhead said.

Moorhead added that China’s cybersecurity risk accusation about Micron is “a front.”

“Compared to a CPU or a GPU system, it’s pretty hard to embed something nefarious into something like storage or memory,” he said. “That would be technology that I have never heard of.”

Schumer led a delegation of senators to visit China in October for a rare meeting with President Xi Jinping, in part to discuss the ban on Micron.

“We think China was being very nasty about this to Micron,” Schumer told CNBC ahead of the visit. “China’s upset with the Biden administration’s very smart prohibition of selling certain types of chip manufacturing equipment to China. But we’re going to stick up for Micron.”

This also isn’t the first time Micron has been at the center of U.S.-China tensions. In 2018, the U.S. accused Chinese chip company Fujian Jinhua of stealing intellectual property from Micron, a claim the Chinese company denied.

With no slowdown in geopolitical tension, Micron is instead focusing on U.S. expansion. Water and power were both significant reasons Micron settled on New York for its biggest project.

A rendering of Micron’s planned four memory chip fabs it will build north of Syracuse, New York, spending $100 billion over the next 20 years.

Micron

“Not just the Finger Lakes, but two Great Lakes: Lake Erie and Lake Ontario,” Hochul said. “There’s plentiful water and low-cost power generated primarily by hydroelectric and wind and solar. So we’re ready for it. We know it’s going to be a transition, but that’s what we want to do.”

Micron said each of its new fabs will use the equivalent of 25 Olympic-size swimming pools worth of water each day, with a goal of reusing or recycling 75% of that. Micron will also use the same amount of energy required to power some 25,000 homes.

“The energy costs are, interestingly enough, lower in the United States than most parts of the world,” Moorhead said. “People are more expensive in the United States, and so is the materials and the cost to build that factory. But that gap is narrowing over time.”

In Arizona, the world’s advanced chip leader, Taiwan Semiconductor Manufacturing Company, recently blamed a shortage of skilled labor for delays to its massive $40 billion fab under construction.

“That won’t happen in New York because we already have a legacy,” Hochul said. “We have Wolfspeed, we have GlobalFoundries. So this is not a new industry to us.”

Micron runs a Chip Camp in Boise for middle schoolers, which Gatzemeier’s daughter attended over the summer, and is investing in university programs to feed the pipeline for future semiconductor engineers.

“We’re actively starting our hiring ramp now,” Gatzemeier said. “We’ve started aggressively targeting all the universities. We’re also really going to draw on the global resources that Micron has across the world and bring in some of that semiconductor expertise to help train these new team members.”

Technology

Intel’s wild week leaves Wall Street more uncertain than ever about chipmaker’s future

Published

6 hours ago

September 21, 2024

admin

Intel's wild week leaves Wall Street more uncertain than ever about chipmaker's future

Intel CEO Patrick Gelsinger speaks at the Intel Ocotillo Campus in Chandler, Arizona, on March 20, 2024.

Brendan Smialowski | AFP | Getty Images

It was quite a week for Intel.

The chipmaker, which has lost over half its value this year and last month had its worst day on the market in 50 years after a disappointing earnings report, started the week on Monday by announcing that it’s separating its manufacturing division from the core business of designing and selling computer processors.

And late Friday, CNBC confirmed that Qualcomm has recently approached Intel about a takeover in what would be one of the biggest tech deals ever. It’s not clear if Intel has engaged in conversations with Qualcomm, and representatives from both companies declined to comment. The Wall Street Journal was first to report on the matter.

The stock rose 11% for the week, its best performance since November.

The rally provides little relief to CEO Pat Gelsinger, who has had a tough run since taking the helm in 2021. The 56-year-old company lost its long-held title of world’s biggest chipmaker and has gotten trounced in artificial intelligence chips by Nvidia, which is now valued at almost $3 trillion, or more than 30 times Intel’s market cap of just over $90 billion. Intel said in August that it’s cutting 15,000 jobs, or more than 15% of its workforce.

But Gelsinger is still calling the shots and, for now, he says Intel is pushing forward as an independent company with no plans to spin off the foundry. In a memo to employees on Monday, he said the two halves are “better together,” though the company is setting up a separate internal unit for the foundry, with its own board of directors and governance structure and the potential to raise outside capital.

Intel CEO Pat Gelsinger speaks while showing silicon wafers during an event called AI Everywhere in New York, Thursday, Dec. 14, 2023.

Seth Wenig | AP

For the company that put the silicon in Silicon Valley, the road to revival isn’t getting any smoother. By forging ahead as one company, Intel has to two clear two gigantic hurdles at once: Spend more than $100 billion through 2029 to build chip factories in four different states, while simultaneously gaining a foothold in the AI boom that’s defining the future of technology.

Intel expects to spend roughly $25 billion this year and $21.5 billion next year on its foundries in hopes that becoming a domestic manufacturer will convince U.S. chipmakers to onshore their production rather than relying on Taiwan Semiconductor Manufacturing Company (TSMC) and Samsung.

That prospect would be more palatable to Wall Street if Intel’s core business was at the top of its game. But while Intel still makes the majority of processors at the heart of PCs, laptops, and servers, it’s losing market share to Advanced Micro Devices and reporting revenue declines that threaten its cash flow.

‘Next phase of this foundry journey’

With challenges mounting, the board met last weekend to discuss the company’s strategy.

Monday’s announcement on the new governance structure for the foundry business served as an opening salvo meant to convince investor that serious changes are underway as the company prepares to launch its manufacturing process, called 18A, next year. Intel said it has seven products in development and that it landed a giant customer, announcing that Amazon would use its foundry to produce a networking chip.

“It was very important to say we’re moving to the next phase of this foundry journey,” Gelsinger told CNBC’s Jon Fortt in an interview. “As we move to this next phase, it’s much more about building efficiency into that and making sure that we have good shareholder return for those significant investments.”

Still, Gelsinger’s foundry bet will take years to pay off. Intel said in the memo that it didn’t expect meaningful sales from external customers until 2027. And the company will also pause its fabrication efforts in Poland and Germany “by approximately two years based on anticipated market demand,” while pulling back on its plans for its Malaysian factory.

TSMC is the giant in the chip fab world, manufacturing for companies including Nvidia, Apple and Qualcomm. Its technology allows fabless companies — those that outsource manufacturing — to make more powerful and efficient chips than what’s currently possible at volume inside Intel’s factories. Even Intel uses TSMC for some of its high-end PC processors.

Intel hasn’t announced a significant traditional American semiconductor customer for its foundry, but Gelsinger said to stay tuned.

“Some customers are reluctant to give their names because of the competitive dynamics,” Gelsinger told Fortt. “But we’ve seen a large uptick in the amount of customer pipeline activity we have underway.”

Prior to the Amazon announcement, Microsoft said earlier this year it would use Intel Foundry to produce custom chips for its cloud services, an agreement that could be worth $15 billion to Intel. Microsoft CEO Satya Nadella said in February that it would use Intel to produce a chip, but didn’t provide details. Intel has also signed up MediaTek, which primarily makes lower-end chips for mobile phones.

U.S. President Joe Biden listens to Intel CEO Pat Gelsinger as he attends the groundbreaking of the new Intel semiconductor manufacturing facility in New Albany, Ohio, U.S., September 9, 2022.

Joshua Roberts | Reuters

Backed by the government

Intel’s biggest champion at the moment is the U.S. government, whish is pushing hard to secure U.S.-based chip supply and limit the country’s reliance on Taiwan.

Intel said this week that it received $3 billion to build chips for the military and intelligence agencies in a specialized facility called a “secure enclave.” The program is classified, so Intel didn’t share specifics. Gelsinger also recently met with Commerce Secretary Gina Raimondo, who is loudly promoting Intel’s future role in chip production.

Earlier this year, Intel was awarded up to $8.5 billion in CHIPS Act funding from the Biden administration and could receive an additional $11 billion in loans from the legislation, which was passed in 2022. None of the funds have been distributed yet.

“At the end of the day, I think what policymakers want is for there to be a thriving American semiconductor industry in America,” said Anthony Rapa, a partner at law firm Blank Rome who focuses on international trade.

For now, Intel’s biggest foundry customer is itself. The company started reporting the division’s finances this year. For the latest quarter, which ended in June, it had an operating loss of $2.8 billion on revenue of $4.3 billion. Only $77 million in revenue came from external customers.

Intel has a goal of $15 billion in external foundry revenue by 2030.

While this week’s announcement was viewed by some analysts as the first step to a sale or spinoff, Gelsinger said that it was partially intended to help win new customers that may be concerned about their intellectual property leaking out of the foundry and into Intel’s other business.

“Intel believes that this will provide external foundry customers/suppliers with clearer separation,” JPMorgan Chase analysts, who have the equivalent of a sell rating on the stock, wrote in a report. “We believe this could ultimately lead to a spin out of the business over the next few years.”

No matter what happens on that side of the house, Intel has to find a fix for its main business of Core PC chips and Xeon server chips.

Intel’s client computing group — the PC chip division — reported about a 25% drop in revenue from its peak in 2020 to last year. The data center division is down 40% over that stretch. Server chip volume decreased 37% in 2023, while the cost to produce a server product rose.

Intel has added AI bits to its processors as part of a push for new PC sales. But it still lacks a strong AI chip competitor to Nvidia’s GPUs, which are dominating the data center market. The Futurum Group’s Daniel Newman estimates that Intel’s Gaudi 3 AI accelerator only contributed about $500 million to the company’s sales over the last year, compared with Nvidia’s $47.5 billion in data center sales in its latest fiscal year.

Newman is asking the same question as many Intel investors about where the company goes from here.

“If you pull these two things apart, you go, ‘Well, what are they best at anymore? Do they have the best process? Do they have the best design?'” he said. “I think part of what made them strong was that they did it all.”

— CNBC’s Rohan Goswami contributed to this report

WATCH: CNBC’s full interview with Intel CEO Pat Gelsinger

Technology

How Elon Musk hopes his new supercomputers will boost his businesses

Published

7 hours ago

September 21, 2024

admin

How Elon Musk hopes his new supercomputers will boost his businesses

Elon Musk is on a mission to build new supercomputers. As the CEO of Tesla and his new artificial intelligence startup xAI, the tech titan has big plans for how artificial intelligence can help to supercharge his businesses.

In January, he wrote on X that Tesla should be viewed as an AI/robotics company rather than a car company. Tesla’s custom-built supercomputer named Dojo is key to this transformation. Tesla has said it plans to spend $500 million to build the supercomputer in Buffalo, New York. Tesla is also building another supercomputer cluster, called Cortex, at the company’s headquarters in Austin, Texas.

Dojo will process and train AI models using the large amounts of video and data captured by Tesla cars. The goal is to improve Tesla’s suite of driver assistance features, which the company calls Autopilot, and its more robust Full Self-Driving or FSD system. Subscriptions to Tesla’s FSD features cost $99 a month and include automatic lane changes, automatic parking and automatic stopping for traffic lights and stop signs.

“They’ve sold what is it, 5 million plus cars. Each one of those cars typically has eight cameras plus in it. And if you think then that those cars are driving around, let’s just say 10,000 miles a year on average, they’re streaming all of that video back to Tesla,” says Steven Dickens, chief technology advisor at the Futurum Group. “So what can they do with that training set? Obviously they can develop Full Self-Driving and they’re getting close to that.”

Despite their names, neither Autopilot nor FSD make Tesla vehicles autonomous and require active driver supervision, as Tesla states on its website. In the past, the company has garnered scrutiny from regulators who say that Tesla falsely advertised the capabilities of its Autopilot and FSD systems. But reaching full autonomy is critical for Tesla, whose sky-high valuation is largely dependent on bringing robotaxis to market, some analysts say.

The company reported lackluster results in its latest earnings report and has fallen behind other automakers working on autonomous vehicle technology. These include Alphabet-owned Waymo, which is already commercially operating fully autonomous taxis in several U.S. cities, GM’s Cruise and Amazon’s Zoox. In China, competitors include Didi and Baidu.

Tesla hopes Dojo, which Musk says has been running tasks for Tesla since 2023, will change that. A Tesla robotaxi event originally scheduled for August is now expected to occur in early October.

Dojo can also be useful for training Tesla’s humanoid robot, Optimus, which the company plans to use in its factories starting next year. Musk has said that Tesla plans to spend $10 billion this year on AI.

Musk is also betting on supercomputers to run his new AI venture xAI. Musk launched xAI in 2023 to develop large language models and AI products, like its chatbot Grok, as an alternative to AI tools created by OpenAI, Microsoft and Google.

Despite being one of its founders, Elon Musk left OpenAI in 2018 and has since become one of the company’s harshest critics. In June, it was announced that xAI would build a supercomputer in Memphis, Tennessee to train Grok. In early September, Musk revealed that a portion of the Memphis supercomputer, called Colossus, was already online.

To learn more about Elon Musk’s supercomputer plans, watch the video.

Technology

SEC says Elon Musk should be sanctioned if he keeps dodging Twitter depositions

Published

21 hours ago

September 20, 2024

admin

SEC says Elon Musk should be sanctioned if he keeps dodging Twitter depositions

Elon Musk, Chief Executive Officer of SpaceX and Tesla and owner of X looks on during the Milken Conference 2024 Global Conference Sessions at The Beverly Hilton in Beverly Hills, California, U.S., May 6, 2024.

David Swanson | Reuters

The Securities and Exchange Commission has asked a federal judge to sanction Elon Musk if he continues to violate the court’s order to appear for a deposition in a probe of his 2022 Twitter acquisition.

The SEC has been investigating whether Musk or anyone else working with him committed securities fraud in 2022 as the Tesla CEO sold shares in his automaker and shored up a stake in Twitter, ahead of his leveraged buyout of the company now known as X.

In May, the court ordered Musk to appear for a deposition by the financial regulators regarding the Twitter deal.

“Musk has now failed to appear before the SEC twice: first in September 2023, in defiance of a lawful administrative subpoena, and last week, in defiance of a clear court order,” SEC attorney Robin Andrews said in the Friday filing.

Andrews asked the judge to consider sanctions should Musk delay further, according to the filing.

“The Court must make clear that Musk’s gamesmanship and delay tactics must cease,” Andrews wrote.

The filing also revealed, in a footnote, that the SEC intends to ask the court to hold Musk in “civil contempt” for canceling a deposition on Sept. 10, giving the agency only a few hours notice that he would not appear. Musk’s cancellation cost the SEC time and money after it sent personnel to Los Angeles to depose him and he didn’t appear for the investigative interview, the agency said.

Musk’s deposition in the probe has been rescheduled for a date in early October at an SEC office, the filing said.

“Without further action by the Court, nothing deters Musk” from “simply failing to show up for that date,” Andrews wrote.

Musk’s attorney, Alex Spiro, a partner at Quinn Emanuel in New York, wrote in a response that “such drastic action would be inappropriate,” adding that the SEC and Musk had agreed rescheduling would be permissible in light of an emergency.

Additionally, Musk and his companies have “cooperated and are cooperating with the SEC in multiple other ongoing investigations,” Spiro wrote.

In a separate, civil lawsuit concerning the same Twitter deal, the Oklahoma Firefighters Pension and Retirement System has sued Musk in a federal court in New York accusing him of deliberately concealing his progressive investments in Twitter and intent to buy out the company.

The pension fund’s attorneys argue that Musk, by failing to clearly disclose his investments in and intentions to buy Twitter, had influenced other shareholders’ decisions and put them at a disadvantage.

Discovery from that case in New York yielded correspondence between an unnamed person at Morgan Stanley, and the executive who manages Musk’s money, Jared Birchall. In the messages, the Morgan Stanley contact wrote in February 2022 that Musk’s Twitter stock-buying strategy was closely held.

“No one knows what is going on and why but you and me,” the person at Morgan Stanley wrote. “Not compliance, not anyone.”

Read the court filing below: