There was a fierce study this week in contrasts at the top of the chip industry. Nvidia, which recently acquired Intel to become the world’s most valuable chipmaker, reported blowout revenue. Jensen Huang, its founder and CEO, used the moment to explain his vision for what comes next.
As he spoke, Intel sought to renew its flagship support on Wall Street by accelerating the repurchase of $ 10 billion worth of shares. As a measure of confidence in its ability to hold off on cash, it was certainly powerful: the purchase is commensurate with the total free cash flow that Nvidia has generated together over the past three years. But it could not hide the fact that Intel’s manufacturing process technology hit a wall.
The failure has left a big question as to where Intel’s future lies, and whether it even has a place at the leading edge of chip manufacturing. But if it flexes its muscles and explores the landscape before it is, what will Nvidia, the new king of the hip, become?
For the past 21 years, the California company has taken its graphics processing units, such as GPUs, from their original brand in gaming PCs to data centers, where their parallel processing capabilities have made them the main engines for the data-intensive task of training AI systems. With its attempt to buy Arm, the SoftBank-owned chip design company, it is now trying to consolidate that position while also getting its first ropes in some of the largest current and future markets for silicon.
Data centers are, according to Mr. Huang, the new “computer unit”. Instead of writing programs that run on a single processor or server, coders will one day write software designed to run on an entire data center: what happens behind the scenes, because data shuttles between machines and large-scale Internet services are packaged in the most efficient way, will be the concern of a company like Nvidia, not the programmer. Nvidia has all the GPUs and network technology to pursue much of this: what it lacks is a base in CPUs, like central processing units, the core of Intel’s business.
© Tyrone Siu / Reuters
Years of design work by Arm to match Intel in data center CPUs have finally paid off. One sign of this: Amazon, the largest provider of cloud computing infrastructure, is now designing its own Arm-based server chips, under the aptly weighted name Graviton.
Possessing all the intellectual property for the silicon that powers gigantic data centers, Nvidia would give a neat position that even Intel – which makes more than half of its profits from this market – could not match. But it’s not clear that an acquisition is necessary: Nvidia may have IP (of all) license from Arm. And there is a good chance that striving for an acquisition, which has yet to be confirmed, will antagonize others in the sector and provoke unpredictable competitive responses.
A deal would put Nvidia in a position to not only sell its own silicon, but also provide fundamental IP licensing that other companies – some of Nvidia’s competitors – would have to create their own chips.
The chip industry already has a company like this. Qualcomm, which invented mobile communications, has long come under fire from some of its biggest customers and rivals over its techniques for generating maximum revenue from the sale of both IP licenses and its own chips. A U.S. appeals court defeated a government anti-trust case against Qualcomm last week, ruling that its corporate practices, while “hyper-competitive,” did not break the law.
The pursuit of Arm can be motivated primarily by Nvidia’s ambitions in data centers, but it also opens up a broader landscape. Nvidia is already active in other AI markets, such as robotics and self-driving cars. And Arm would give Nvidia its first entry on the ‘Internet of Things’ and – by far Arm’s largest current market – smartphones.
Removing the neutral ownership of SoftBank would have unpredictable consequences. Apple – which fought a long and valuable battle against Qualcomm through the courts before reaching a settlement last year – is one giant Arm customer with which Nvidia would need common ground. Qualcomm even uses Arm technology in the CPUs embedded in its all-in-one chipsets that run high-end smartphones.
These companies are not currently competing directly with Nvidia, so forcing them to apply for a technology license could not cause any immediate problems. But if Nvidia ends up paying more than $ 32 billion for Arm – the price discussed – it will have strong incentives to maximize its licensing revenue to get its bill back.
And in the background, there will always be a lingering question about where a new embedded Nvidia will turn next if it is looking for new silicon worlds to conquer.
