Bloomberg Businessweek (May 6, 2024)

Год выпуска: May 6, 2024

Автор: Bloomberg Businessweek

Жанр: Бизнес

Издательство: «Bloomberg Businessweek»

Формат: PDF (журнал на английском языке)

Качество: OCR

Количество страниц: 80

THE TECH ISSUE

A FEW MONTHS AGO, A TUGBOAT SET off from New Orleans, pushing a barge up the Mississippi River. It moved northward through the Deep South and then cut east, past Indiana farmland along the Ohio River. The barge was carrying a cold box for a gigantic air separation unit, a piece of industrial equipment that pulls nitrogen from the atmosphere for use in manufacturing. The cargo was so big that transporting it through Cincinnati, where container ships usually unload in southern Ohio, was impractical. At 270 feet long, 19 feet wide and 24 feet tall, it wouldn’t fit under the highway overpasses near the port. So the boat floated 75 miles or so upriver before stopping at an abandoned dock just outside the tiny village of Manchester. For weeks now, it’s been sitting by the bank, waiting for the final leg of its journey, set to begin later this spring.

Then a team of workers will load the cold box onto a massive semitrailer designed for so-called superloads. While police cars clear the roads between the river and the tower’s destination, cherry-pickers will push power lines and traffic lights out of the way. This convoy-slash-roadblock will proceed at about 8 mph to New Albany, Ohio, stretching out a trip that most cars can make in two hours to about two weeks. Over the next few months, a similar operation will take place 20 more times, delivering air separation and crane parts to a construction site that many in the region are betting will be the future of the semiconductor industry.

This is Ohio One, a future “mega fab”—or factory—being built by Intel Corp. The company has committed $28 billion to build a fab that, if all goes well, will be able to credibly compete for contracts to manufacture state-of-the-art chips used in smartphones and in the training of next-generation artificial intelligence algorithms. Ohio One, says Chief Executive Officer Patrick Gelsinger, could eventually be the biggest chip factory on Earth. Not to be grandiose or anything, he adds, but it’s part of a strategy that will also stave off America’s economic decline, because chips are essential to almost every manufactured good. “I just view this as existential to our economy, to our national security,” Gelsinger says at Intel’s headquarters in Santa Clara, California. “I don’t view failure as an option.”

The fab isn’t scheduled to open until at least 2027, but politicians are already taking credit for even the tiniest bits of progress, posting videos of the superload trips on social media and using every opportunity to talk about the potential job creation. These officials include state Republicans as well as members of the Biden administration, which has made the Intel plant a showpiece for its economic policy. In his State of the Union address two years ago, President Joe Biden called the Ohio One site a “field of dreams,” describing it as “the ground on which America’s future will be built.” This past March the US announced it had agreed to provide Intel with $19.5 billion in loans and grants to help finance the project in Ohio, along with similar ones in Arizona and Oregon, as part of the Chips and Science Act. “Folks, it’s about time,” Biden said, announcing the Intel subsidy package. “Even though we invented the most advanced chips, we make 0% of them today.”

The subtext wasn’t very sub. From war-making to basic communication, the world runs on semiconductors, and most of the best ones are made in Taiwan. In the event of another pandemic or if Beijing decides to invade Taiwan, US supplies of all kinds of hardware could be crippled. In the weeks that followed the announcement of the Chips Act award, there were more subsidies for chip factories in the US: $12 billion for one being built in Arizona by Taiwan Semiconductor Manufacturing Co., better known as TSMC, and $6 billion for one in Texas being built by South Korea’s Samsung Electronics Co. These domestic facilities can’t produce everything the US needs. But while they don’t represent chip independence, they’re at least an insurance policy and, maybe, a start.

Today, Biden’s field of dreams is mostly a muddy expanse on the edge of a pig farm. Still, things are happening. Suppliers are renting real estate, community colleges are spinning up programs to train future semiconductor technicians, and 1,400 workers are digging trenches, laying foundations and pouring concrete for the building itself. That’s the easy part. Semiconductor fab production is arguably the most challenging and expensive act in all of business, a mess of attempts to turn experimental science into mass production.

If the stakes are high for the US, they may be even higher for Intel. The company that turned the phrase “Silicon Valley” into a metonym for the tech industry, and was one of the world’s largest by market value two decades ago, has been upstaged by TSMC and a raft of chipmakers that outsource the actual making of chips. Nvidia Corp., once a bit player best known to gamers who coveted its graphics cards, is now worth more than $2 trillion, or more than 16 Intels. Sales of Intel’s once-dominant processors are stagnant, its stock has fallen more than 30% this year, and its manufacturing, once the crown jewel, is bleeding money. In April, Intel disclosed that the manufacturing operation lost $7 billion last year, with more losses to come.

“Huge numbers, atrocious numbers,” says Gelsinger, a wiry, intense engineer who, at 63, has spent most of his professional life at Intel. But Gelsinger blames those losses on the previous regime. He says that when he took over three years ago, Intel was months away from being sold for parts. Now, he says, the company has embarked on a yearslong plan to restore its dominance, just as it did in the 1990s when it overcame what seemed like insurmountable competition from Japanese chipmakers. “This is gonna be hard, multiyear, heavy, expensive,” he says. On the other hand, he adds, “we’ve done it before.”

AT BOTTOM, THE MODERN TECHNOLOGY industry requires somebody to figure out a way to arrange tiny electric circuits, known as transistors, onto polished silicon discs. Robert Noyce and Gordon Moore pioneered the process at Fairchild Semiconductor, a company they co-founded in 1957, and industrialized it at Intel, which they started 11 years later. Intel created the first commercial microprocessor, the 4004. The 4004 included about 2,300 transistors, with circuits that were about one-tenth the thickness of a human hair, intricately arranged in a thumbnail-size package with the processing power of a computer that would have once occupied an entire closet. This was done using photolithography, a printing technique in which the silicon disc—a “wafer”—is covered with a light-sensitive material and hit with light, burning an intricate pattern. The process has only gotten more complex. In 1976 the top-of-the-line chip had 29,000 transistors. The first Pentium, released by Intel in 1993, had 3.1 million. The company’s Xeon chips, which enabled the rise of Gmail, Netflix and other cloud computing services, have billions.

Intel has defined itself, and the rest of the industry, largely in terms of this sort of escalating complexity. In 1965, Moore predicted that the number of transistors on a state-of-the-art chip would double every year. Moore himself regarded this prediction as more of a guess—decades later he described it as “wild extrapolation”—but Intel took the idea, which has since become known as Moore’s Law, and helped turn it into something like a Silicon Valley religion. Elon Musk has applied it to batteries for electric cars, Masayoshi Son has employed it as a justification for SoftBank Group Corp.’s wild investments in robotics, and Ray Kurzweil has made it central to his theory that someday we’ll all upload our brains to computers and live forever.

While Moore’s Law has exploded in popularity as a metaphor, its predictive power has weakened. In 2006, Gelsinger, then an Intel senior vice president widely seen as a potential CEO, wrote an essay for an engineering newsletter that declared “the genius lives on,” noting “we see no end to Moore’s Law in the coming decade.” But about this time, developers found they could no longer make a chip twice as efficient by simply doubling its number of transistors. As breakthroughs grew tougher and fab costs rose, the market drove more manufacturers out of the business. This included Intel’s main US competitor, Advanced Micro Devices Inc. (AMD), which eventually outsourced its production of high-end edge chips to TSMC instead. In 2009, Intel announced a reorganization that sidelined Gelsinger, and he left to become chief operating officer at EMC Corp., which makes memory systems for servers.

Gelsinger, who’d joined Intel as an 18-year-old trade school graduate from rural Pennsylvania, hasn’t forgotten how it felt to be pushed aside. He’d been at the company for 30 years and was a mentee of legendary CEO Andrew Grove. “It took me a long time to get over it,” he says, before invoking the five-note jingle that still plays at the end of Intel’s commercials. “I did the Intel Bong as I went to sleep each night, and I didn’t stop doing it for three or four years after I was gone. I wanted to run this place.”

His departure coincided with a series of increasingly dire missteps at Intel. The company passed up the chance to manufacture chips for Apple Inc.’s iPhone and struggled to find a way to break into that market in the years that followed. Rivals that took the job—Samsung and, later, TSMC—rode the smartphone boom and used the cash to chase Moore’s Law. There, too, Intel shrank from a challenge.

Intel had worked closely with a Dutch lithography company, ASML Holding NV, to develop an etching technique called extreme ultraviolet lithography. EUV promised to help squeeze the size of the features on a cutting-edge chip down from 14 nanometers to 10nm and eventually to even smaller sizes. (In chips, of course, smaller is better.) But the technique was unproven and expensive, and then-CEO Brian Krzanich attempted a workaround with less advanced machines. Gelsinger now calls this “one of the stupidest decisions that Intel made.” He blames it for delays that left the company a full two generations of chips and several years behind TSMC by 2020. (Krzanich didn’t respond to requests for comment.) These technical setbacks would have lasting damage, as AMD’s Taiwan-made server chips gained market share and as companies like Apple, Google and Amazon cut their purchases of Intel’s chips and designed their own instead, paying TSMC to manufacture them. In 2020, Krzanich’s successor, Bob Swan, suggested Intel might have to give up manufacturing, too.

Gelsinger watched this with a mix of pain and frustration. Intel, he believed, was perhaps six months from a total collapse where “nobody could have saved it,” he says. “We would’ve been exiting technology development, and we would’ve been selling the manufacturing assets at dimes on the dollar.”

At the time, he was running another company, VMware, but he wrote a memo to Intel’s board, urging a radically different strategy. Gelsinger would bet it all on manufacturing, investing billions of dollars on fabs while frantically trying to catch up to TSMC’s most sophisticated manufacturing processes. Intel would enter TSMC’s main business of making chips for other companies—known as the “foundry” model— which would lead to higher volumes and allow it to continue to pursue next-generation research. In February 2021, Intel’s board announced the CEO job was his. “The old Intel is the new Intel,” Gelsinger told a group of Wall Street analysts the following month. “The world needs more semiconductors, and we’re going to step into that gap.”

The comments were meant to evoke nostalgia for the Intel of yore and speak to a new political reality. The supply chain disruptions of the Covid-19 era had crystallized the risks of outsourcing. Prominent Republicans and some Democrats were loudly warning about China’s influence on the tech industry and threat to Taiwan. And Biden had just won the White House promising massive government spending to help rebuild the US industrial base. Gelsinger, adopting a pose that was part patriotic technologist and part cold-hearted capitalist, suggested to lawmakers that the US would need to step up its subsidies or lose a national champion. “We either get the Chips Act done,” he recalls saying, “or I should build all my factories in Asia.”

There was a logic to this, but it sidestepped the role Intel itself had played in the decline of US chipmaking capacity when it ceded technological leadership to TSMC. It arguably understated the challenges of mounting a comeback. Intel’s share price has plummeted because investors worry the company can no longer sell enough chips on its own to build lots of new state-of-the-art factories, and its ability as a foundry is unproven. Gelsinger has disputed this, but the fear is that big foundry customers might wait years before giving Intel a substantial contract, says Matthew Ramsay, a semiconductor analyst at TD Cowen. “Intel is still one of the most, if not the most, strategically important manufacturing companies in the United States,” he says. “But when you look at their track record over the last seven or eight years juxtaposed against the amount of wood-chopping they need to do, it’s going to be hard.”

THE FIRST THING YOU NOTICE WHEN you drive onto the grounds of any fab is the infrastructure: power stations, water treatment plants, storage tanks and a vast thicket of pipes used to bring in the enormous quantities of purified air and water the chipmaking process needs.

Each Intel fab starts with the same basic design, a flat-roofed structure with the main factory on the third floor. This is the clean room, the heart of the operation. It sits above two floors of utilities, ducts and equipment, and one floor below what’s got to be the world’s most powerful HVAC system, an array of fans and filters blowing in ultra-pure air. The floor tiles are perforated, and a huge vacuum system underneath them ensures that the air turns over six times per minute. The air filtration is expensive but necessary. A single particle—a tiny flake of dead skin, a droplet of water from a sneeze—can destroy a $10,000 chip. Bright light applied at the wrong time is bad, too, which is why the rooms are lit a dull yellow.

Intel executives like to boast that their clean rooms are a thousand times cleaner than a hospital operating room. The low light and all the talk of purity can make a visit feel a little cultish, as can the secrecy. Until recently, Intel almost never let outsiders in, and even executives would need special permission. But the promise of almost $20 billion in taxpayer money, and the need to persuade other chipmakers to trust it with their designs, has changed the equation.

In mid-April, a gang of Intel executives and their publicists led two Bloomberg Businessweek reporters through what one guide called a “hierarchy of cleanliness” at D1X, the company’s research and development fab just outside Portland, Oregon. We put on glove liners, hairnets and protective shoe covers. We sanitized our phones with wipes, then turned them off because radio signals can disrupt the machines. That got us into an inner room where we put on hoods with face masks, followed by Gore-Tex coveralls, a second pair of shoe covers, a second pair of gloves and, finally, safety glasses. We were given latex notepads (regular paper sheds particles when written on) and were warned that bumping into anything could risk shutting down production. Then we proceeded, very carefully, through a heavy set of glass doors...

Intel Stakes Its Future on Ohio

• The chipmaker is spending billions to cast itself as a national champion

The UAW Takes Notes on a Swedish Tesla Strike

• The monthslong stalemate may embolden unions fighting Musk in the US

Crypto’s Hot Again. What Could Go Wrong?

• Regulators need to protect investors from an even colder winter

GM Kicks Apple Out of Its Cars

• The carmaker bets it can win CarPlay users over to its own digital dashboard

Cruel Intentions

• How a massive hack by a young Finnish cybercriminal left a nation scarred

IN BRIEF    

• G-7 coal phaseout
• California grows
• DEA weed shift

OPINION

• The right way to help struggling communities

AGENDA

• Columbia commencement
• UK growth
• US inflation

REMARKS

• The geopolitics behind the battle over TikTok

BUSINESS

• AI may automate the dangerous job of butchering meat
• Paramount or no Paramount, Legendary is dreaming big

TECHNOLOGY

• Musk and Bezos shoot for the moon
• China looks offshore for an edge in the space race
• Travel agent to the stars (or at least low Earth orbit)

FINANCE

• Money may have failed Bill Hwang, but his faith is unbowed
• Regional banks embrace their lack of excitement

ECONOMICS

• Rising prices at the pump are plaguing Biden
• Modi’s long-shot pledge for a developed India

PURSUITS / SUMMER TRAVEL

• Finding freedom in the English countryside
• Hot hotels for when you want to splurge or scrimp
• Set sail for savings with European cruises
• The perfect straw bag for summer adventures
• Escape the heat—and crowds—on the Danish coast
• Going beyond the croissant in Paris pastry shops
• The thrilling urban art of Belgium’s Liege
• Romania emerges as an offbeat luxury gem
• Where to experience Italy without the crush of tourists

LAST THING

• Keep an eye on these folks for big things in tech

скачать журнал: Bloomberg Businessweek (May 6, 2024)