Chipmaking in America
To survive, Intel must break itself apart
And it should do so before it is too late
Intel once set the pace of technological progress.
Gordon Moore, one of its founders, predicted in 1965 that chips would get faster and cheaper with metronomic consistency.
Over the decades Intel brought Moore’s Law to life, designing and building the processors that powered servers and, later, personal computers.
Today it makes headlines for its turmoil more than its technology.
On August 7th President Donald Trump demanded the resignation of Lip-Bu Tan, Intel’s boss, citing his links to China, only to praise Mr Tan four days later after meeting him.
Reports soon surfaced that the government was pursuing a 10% stake in the company, which would make it Intel’s largest shareholder.
On August 18th SoftBank, a Japanese tech conglomerate, announced that it would invest $2bn in the company.
The drama has refocused attention on Intel’s plight.
The company has missed nearly every big shift in its industry over the past two decades.
It failed to profit from the rise of smartphones, was slow to adopt advanced lithography tools and has largely sat out the boom in artificial intelligence (AI).
Between 2021 and 2024 revenue dropped by a third, from nearly $80bn to just over $50bn; last year it made a net loss of almost $20bn (see chart 1).
Over the past five years its market value has fallen by roughly half, to around $100bn.
TSMC, which has stolen Intel’s crown as the world’s leading chip manufacturer, is worth ten times as much.
Yet Intel still matters, as Mr Trump’s interest shows.
The most advanced chips, vital for smartphones and AI, are now made almost entirely by TSMC.
America’s tech giants depend on it.
Such reliance on a single supplier—particularly one based in Taiwan—is risky.
Intel is one of the few firms that could rival TSMC.
But it will need more than government subsidies to do so.
If it is to recover its chipmaking prowess, Intel will need to break itself apart.
Throughout its history Intel has designed and built its own chips.
That integration let it use its manufacturing prowess to deliver better products even when its designs lagged behind.
From the mid-2010s, however, repeated missteps in its manufacturing saw it fall behind TSMC.
Deprived of that advantage, Intel’s processors became uncompetitive with those from AMD, a long-term rival which gave up on manufacturing long ago.
In 2021 Intel, too, began outsourcing production of its most advanced chips to TSMC.
The erosion of Intel’s manufacturing leadership has coincided with fiercer competition in the market for designing processors.
As recently as 2019 Intel controlled 84% of the global market for PC chips and 94% for servers.
By 2024 those figures had fallen to 69% and 62%, respectively (see chart 2).
AMD, using the x86 architecture pioneered by Intel, has developed better chips.
Cloud giants such as Amazon, Google and Microsoft, which were once reliant on Intel, now design their own processors using outlines from Arm, a British company owned by SoftBank.
In December Amazon said that half the server capacity it added in the preceding two years used its own silicon.
Pat Gelsinger, Intel’s boss from 2021 to 2024, tried to reverse the slide.
He split design and manufacturing into two units, allowing the product arm to shop around for the best manufacturer while opening Intel’s chip factories, called “fabs”, to outsiders.
To build a contract-chipmaking business, known as a “foundry”, Mr Gelsinger then set about splurging $90bn on new fabs in four American states.
He tapped private equity and bagged nearly $8bn in subsidies under America’s CHIPS Act to fund his vision.
But the plan was thrown into disarray by a combination of technical problems at the foundry, which deterred external customers, and falling sales at the design arm.
Pat on his back
Mr Tan, who took over in March after Mr Gelsinger was sacked, seems to have different priorities.
He has rightly identified that the company is bloated; at the end of 2024 it employed 109,000 people, nearly as many as Nvidia, the leading designer of AI chips, and TSMC combined.
Mr Tan plans to cut Intel’s workforce by a quarter by the end of this year.
When it comes to AI, he believes that the firm should focus not on designing chips for training models, an area that Nvidia dominates, but on inference, the task of running them.
As for the foundry, last month Mr Tan scrapped projects in Germany and Poland, and pushed construction of Intel’s advanced fabs in Ohio back to the early 2030s.
He also hinted that the company might retreat from leading-edge manufacturing if it cannot secure external customers.
All that may help buy Intel time.
Yet it lacks the boldness needed to save the company from fading into irrelevance.
Evercore, an investment bank, reckons Intel’s design arm might be worth more than $100bn on its own.
But it faces a crowded field and its products are no longer distinctive.
Mr Tan could sell the division to another fabless chipmaker such as Broadcom while it still holds value and focus solely on the foundry, which is troubled but holds more long-term promise.
Its newest “18A” process incorporates transistors that are ahead of TSMC’s, as well as a novel way of feeding power through the back of the chip to save space and energy.
SemiAnalysis, a consultancy, reckons Intel will need to invest a bit over $50bn between 2025 and 2027 to make it competitive in leading-edge manufacturing.
A sale of the design division would more than cover that.
Parting with the design business would help in other ways, too.
Foundries must serve many customers using the same process.
To do so they provide “process design kits”—the blueprints chipmakers use to design their products.
TSMC’s kits are broad and easy to use. Intel still tunes its kits for its own products first.
One veteran designer who has used both says Intel “lacks the experience” of working with outsiders.
Ian Cutress, a semiconductor analyst, notes that Intel sought to buy that expertise with its attempted acquisition of Tower Semiconductor, an Israeli foundry, but the deal collapsed after Chinese regulators withheld approval.
By making its foundry truly independent, Intel may be better able to persuade other chip designers to work with it.
More customers would, in turn, make Intel a more compelling choice.
Foundries live or die by yield—the share of chips that function as intended.
New processes start buggy and improve only with volume.
Foundries typically need yields above 70% to break even; the current rate for Intel’s 18A process is reportedly closer to 10%.
America’s tech giants would certainly welcome another alternative to TSMC.
Samsung, the only other contender in leading-edge chipmaking, recently secured a $16.5bn contract from Tesla, a car company, to make AI chips at a new fab in Texas.
But the South Korean company has a reputation for being difficult with customers and has faced technical challenges of its own.
Indeed, if Intel’s shareholders would rather pocket the proceeds of a sale of the design arm, it is possible that a consortium of would-be foundry customers could be persuaded to invest instead.
SoftBank has also reportedly expressed interest in acquiring Intel’s manufacturing business.
A foundry-only business would certainly be a gamble.
But the longer it dithers, the lower the chance of success.
Intel’s greatness once lay in doing everything.
Its contribution in future may come from doing one thing well: making chips.
0 comments:
Publicar un comentario