why tech giants now produce their chips (Apple, Google, Facebook, Tesla, Amazon…)

Last March, Apple announced the launch of its M1 Ultra chip, the latest in its line of in-house computer chips specially designed to equip its computers. This novelty, the most powerful in the range, is intended for the most powerful Macs, cut out for graphics and video professionals. This launch took place two years after a strategic decision made by Apple. In the spring of 2020, the apple firm ended its fifteen-year partnership with Intel. The goal ? To manufacture the microprocessors of its Macs itself, using an ARM architecture. A real leap into the unknown, not without risk, for Apple. A strategy that is being emulated by American tech giants, from Tesla to Facebook. They want to reinternalize this strategic component and emancipate themselves from third-party companies.

Tim Cook’s bet seems to have paid off: review after reviewer claiming that computers with M1 chips are faster and less overheating than previous models with Intel chips. And Macbook sales, which were faltering a few years ago, are rebounding spectacularly.

The first computer equipped with an M1 chip, a Macbook Air, was launched in November 2020. In April 2021, Tim Cook proudly announced that the new chip had contributed 70% to the growth in revenue from sales of Mac. In the second quarter of 2021, these generated $9.1 billion in revenue, compared to $5.35 billion a year earlier during the same period. This momentum continued in the third quarter of 2021, where Mac sales saw a further 16% year-over-year increase.

Apple made this decision to design a chip tailored to meet the exact quality requirements the company wanted to provide to its customers, in terms of computing power and energy efficiency, explains Russ Shaw, founder of Global Tech Advocates, a international network of new technologies players. As the company knows perfectly the different components of its ecosystem, from its various devices to its Apple Store, it can design a perfectly adapted chip so that this ecosystem works optimally and without a hitch.

Apple plans to evolve this ecosystem in the future, an important dimension as bringing a new chip to market takes time: those found in devices sold today were designed at least three years earlier. early continues the semiconductor industry veteran.

All the tech giants are getting in on it

In turn, Google plans to market Chromebooks and tablets using the Chrome OS operating system equipped with its own chips by 2023, while Amazon is developing in-house chips intended to increase the performance of its cloud data centers. . Facebook and the Chinese giant Baidu are also working on the design of microprocessors cut for artificial intelligence.

In this area, Tesla hit the headlines last August by announcing the development of the Dojo chip, specially designed to drive Tesla’s autopilot algorithm in the company’s data centers. If the manufacturer has already been designing the chips on board its vehicles since 2019 which allow the autopilot to make decisions in real time, the microprocessors necessary for driving the algorithms are latest generation chips, of unparalleled complexity. with these.

Like Apple, Tesla thus intends to offer itself a tailor-made chip, perfectly adapted to its needs (in this case, the improvement of its autopilot algorithm), according to Antoine Chkaiban, analyst at News Street Research, a consulting firm. market intelligence specializing in new technologies. ” Compared to using a generic GPU, offered by Nvidia, they lose in flexibility what they gain in performance for their own use case.

Because Tesla has the financial resources, they can afford to shell out hundreds of millions of dollars for a specially designed chip to drive their autopilot algorithm, and thus gain an extra edge over their competitors who use generic chips. It is also consistent with the strategic ambitions of the company: now, Elon Musk speaks of Tesla as a company specializing in artificial intelligence. »

Tesla: autopilot in the sights of the American authorities

A strategy that remains the prerogative of technological giants

For Mike Demler, an independent analyst specializing in the semiconductor industry, these companies have not followed Apple, their strategies are simply part of the same trend.

Each of these companies has their own reasons for making their custom chips, and none sell consumer products like Apple. However, they have in common that they seek to develop highly specialized microprocessors to run their machine learning algorithms in their data centers.

Some, like Google and Facebook, are at the forefront of AI research, and when you develop your own algorithms, you know exactly what they need from a computational perspective, and can design chips that can to answer it. »

For all of these companies, the fact of developing their own chips is also a way of keeping the competition at bay: by each developing their own piece of silicon designed to fit harmoniously into their technological ecosystem, they offer themselves a additional protection against competitors who would attempt to copy their products or services.

However, this strategy is only accessible to technological giants, as it is too costly for smaller players. Using generic chips already available on the market remains much more economical, especially since they already offer a wide range of possible customizations, notes Antoine Chkaiban. “ A company like Nvidia knows how to be very flexible in its offer, their architecture makes it possible, for example, to target different machine learning models, from vision to protein simulation via transport optimization… Their GPUs can To do. Only the wealthiest companies using huge volumes of microprocessors can therefore afford to design their own chips.

A dynamic made possible by the evolution of the market

In any case, this trend would not have been possible without the evolution that the semiconductor market has taken over the past two decades, as explained by Wayne Lam, analyst at CCS Insight, an intelligence firm in market. ” Over this period, the market has seen both the development of smelters specializing solely in production (such as TSMC) and the rise in popularity of ISA (Instructional Set Architecture) such as ARM.

If companies like Amazon and Google can afford to invest in custom silicon, it’s because the components of the semiconductor value chain have broken down, and we’ve moved from a monolithic business model to a more distributed manufacturing model. Thus, any company juggling sufficient volumes can design its own chips using ARM’s architecture, and work with a foundry like TSMC to mass produce. No more owning your own foundry or owning your own intellectual property for chip architecture, which makes developing a chip much less expensive. »

An explosive geopolitical context

Finally, the recomposition of geopolitical relations also plays a role in the strategy of these companies, according to Russ Shaw. ” The United States has dominated the chip design market since the beginning, but China has been investing heavily in this sector for several years with the intention of taking the lead.

They are still far from being able to compete with American players, but we can see that the question worries the American authorities, who are currently putting in place measures to support their industry and rebuild production capacities on national soil, through Intel, in particular . Faced with China’s claims on Taiwan, TSMC is also under pressure to geographically diversify its production, they are in the process of building their first factory outside of Taiwan in Phoenix, Arizona… The tech giants are aware that the world is moving and want to ensure more control over their value chain in order to limit the risks. »