The shadow of the relentless Andy Grove, who ran Intel for more than a decade at the end of the last century, still looms large at the US chipmaker. He was a hard taskmaster, instilling a culture of operational excellence. The title of one of his books — "Only the paranoid survive" — became a watchword for how to avoid the tech industry's frequent upheavals.
The days since midsummer haven't been kind to Grove's legacy. First Apple dropped Intel as the supplier of processors for its Mac computers, switching to an in-house chip based on technology from ARM. Low-power chip designs from the SoftBank-owned company already froze Intel out of the smartphone market.
Then, this week, Intel lost the title of most valuable American chipmaker. The crown was seized by Nvidia, which was thought of until recently as little more than a specialised producer of chips for high-end video gaming systems. Its shares have risen twenty-fold over the past five years as it has succeeded in adapting its graphics chips, or GPUs, to other data-intensive tasks, most notably training the machine-learning algorithms at the heart of today's AI systems.
Other rivals have also made headway. TSMC, the manufacturer, or "foundry", that turns out chips for other companies (including Nvidia), overtook Intel in market cap terms some time ago. AMD, which has long struggled to make a dent in the market for x86 chip architecture dominated by Intel, has bounced back with designs that are winning a strong following from massive customers like Google.
Half a decade ago, Intel was worth as much as these three companies combined, and its market value has since doubled to nearly US$250 billion ($381.7b). But the basket of competitors is now worth more than US$620b.
There are two factors behind this massive shift in relative value. The most significant has been Intel's failure to maintain its flawless execution. Its move to the latest generation of chips — with features that are only 10 nanometres wide — has been plagued with manufacturing problems. For the first time, TSMC has leapt ahead.
Their different business models mean the two are not direct competitors. But TSMC has served as a manufacturing platform for a new raft of new competitors, opening the way to fabless chipmakers that do not own their own plants, including Nvidia and AMD.
The other factor in Intel's recent eclipse has been the data explosion and rise of machine learning. The most valuable part of the market, for training machine learning algorithms, is one where Nvidia is well ahead.
Intel is not alone in failing to understand the role that GPUs would come to play in AI. But it has failed in its own alternative plan — to develop specialised chips known as accelerators to handle the most data-intensive part of machine learning calculations. It failed to develop a competitive chip based on the acquisition of start-up Nervana in 2016, forcing it late last year to mount a second acquisition — the US$2b purchase of Habana — to secure the technology.
It would be easy to write off Intel's stumbles as part of the long death throes of a former tech monopolist: that it held on too long to a vertically integrated business model and has been outflanked by a new industry structure that has unleashed fleet-footed, fabless competitors. Or that its myopic focus on its own technology blinded it to the future — which arrived first in the shape of ARM in smartphones, and now GPUs for machine learning.
But it would be wrong to read too much into the recent slip — provided Intel can recover its operational prowess. Apple, given its reliance on smartphones, had good reasons to switch its Macs to ARM. But makers of PCs — Intel's stronghold — are unlikely to follow suit, given the weight of software tied to the existing technology.
In a capital-intensive business, Intel's US$21b in profits last year — 40 per cent more than TSMC, Nvidia and AMD combined — make it a fearsome competitor. Even as it invests to iron out the kinks in its 10nm technology, it is now ploughing money into the next generation of even smaller, 7nm chips with the aim of getting back on a level with TSMC by the end of next year.
If Intel's next 10nm chips live up to expectations, next year should see the company back on top in its main markets, according to chip analyst Patrick Moorhead, himself a former AMD hand. For the first time, it also has bets in a wide range of markets well beyond its old PC core — not just the data centre market that has been its main engine of growth, but also in memory, networking, autonomous cars and GPUs.
It is 20 years since Intel's shares peaked at the end of the dotcom boom. They came close to topping that level before the coronavirus crisis hit earlier this year, and the company's earnings power and relatively low market valuation could finally see them reach a new record. Sometimes, the long cycles of tech history can be surprisingly durable.
Written by: Richard Waters
© Financial Times