will get dibs on the next generation of the world’s most coveted chip-making machines, and reclaim its technology lead by 2025, says the company’s new CEO. He reckons the company could “triple, quadruple” in value. I’m quintuply intrigued, and one-quarter convinced.
This past week, Pat Gelsinger, head of Intel (ticker: INTC) since February, rose to the challenge of explaining his four-year plan for “nodes” to a guy who thought those were things doctors sometimes squeeze. It turns out they’re also chip manufacturing generations, and Gelsinger plans to race through a lot of them. “Intel was too arrogant,” he tells me. “We’re breaking that down very rapidly.”
This year, Intel will sell 85% of chips for so-called client computing, including laptops and such, predicts investment bank Raymond James. That would be a seven point drop in two years, and rival
Advanced Micro Devices
(AMD) has risen as quickly.
The trend in servers is similar. The review sites Tom’s Hardware and AnandTech say that Intel’s latest server chips are a big improvement, but that AMD still holds a lead in performance. Buyers for big organizations and data centers are risk-averse, prizing support and long experience, not just price-to-performance ratios, but that won’t slow Intel’s share losses forever. Its slippage in personal computers, meanwhile, has been offset by a Covid-19 surge in home-office buying, but that could change.
How did Intel fall behind? It made all-or-nothing technology bets that led to dead ends, while rivals turned out frequent, incremental improvements. It passed over a new manufacturing technique called extreme ultraviolet lithography, or EUV, which crams more circuits into silicon than traditional lithography.
And it might have been slow to react to a power shift toward foundries, like
Taiwan Semiconductor Manufacturing
(TSM). Taiwan Semiconductor is no mere order-taker. Its operating margins are double those of AMD. So, Intel has been waging a two-front battle on designs and manufacturing.
There have been other, longer shifts. Computing power has migrated to the cloud, so we make do with personal machines for longer. Advanced applications like artificial intelligence favor highly parallel processing, not unlike videogames;
((NVDA)) has parlayed its long success with shoot-’em-ups into data center riches.
The stock market’s judgment is stark. A decade ago, Intel was worth $118 billion, $40 billion more than Taiwan Semiconductor, Nvidia, and AMD combined. Now, Intel is up to nearly $220 billion, but the others combine for $1.1 trillion. After stock buybacks and dividends, Intel investors have made more than 220% over that period. But they could have done almost 100 points better with the
index—or 700 points better with the
PHLX Semiconductor Index.
A positive sign is that top engineers who left Intel in recent years are returning. “They feel the mojo coming back,” Gelsinger says. But it will take more than mojo.
The CEO says he will lean in part on outside foundries for now, while building a foundry operation that will serve other chip makers. Two new Arizona plants are being constructed for $20 billion, not counting equipment. The company has also reportedly held talks to buy GlobalFoundries for $30 billion.
Speaking generically, Gelsinger says, “There will be consolidation over time, and we will be a consolidator.”
Now, about those nodes: Intel has been naming them using ever-shrinking lengths, like “10 nanometer.” The numbers used to refer to a specific transistor part, but with modern architectures, chip makers have been throwing around measurements willy-nilly. So, from here, it’s just numbers: Intel 7 later this year, then 4, then 3. Then we get to Intel 20A and 18A, evoking “the angstrom era.” An angstrom is a tenth of a nanometer, so will those names be based on measurements? Nope: They’re just for marketing. I give the new naming scheme a four for clarity on a scale from orange to pi.
The new nodemap is more than a renaming, however. Proposed chip improvements will be rapid and steady. Intel will adopt EUV starting with next year’s batch. In 2024, it will make its first major architecture change in more than a decade—and says it will catch up with rivals on performance. The following year, it will pass the competition in a shift to EUV’s successor, called high-NA EUV. NA stands for numerical aperture, but it could stand for nougat and almonds so long as the performance gains are as big as promised.
Bulls and bears agree that the plan is aggressive. Bears say that it will cost too much, that results won’t be known for years, and that Intel will continue losing market share between now and then. Bulls say Intel will stabilize its share, and that the risks are reflected in the stock price of 11 times this year’s projected earnings, about half the broad market’s price. Barron’s has been bullish on Intel’s reinvention efforts. Investors who are undecided may want to wait until November, when Intel will hold an analyst meeting, and probably put a price on its plans.
Plenty will be spent on equipment. The EUV machines are made by
(ASML), which now wields vast power. “To the extent that ASML wants to decide market share in the foundry space, to whom it allocates those manufacturing slots is going to be pretty influential,” says Needham analyst Quinn Bolton, who is bullish on Intel.
Gelsinger says he has the EUV machines he needs for now. Of high-NA and his contractual relationship with ASML, he says, “We will be the first production users of those tools.”
ASML stock, as you might imagine, is priced an angstrom short of paradise at 48 times this year’s earnings forecast. Buyers of EUV machines need gear from other companies, too. Bolton’s favorite for stock investors is
(AMAT). It has multiplied five times in price in as many years, but still trades at a folksy 20 or so times earnings.