Rene Haas, head of the intellectual property group at Arm Holdings, spent much of Monday trying to reassure customers that the SoftBank-owned chip design company was not about to turn them into second-class citizens.
His efforts followed news the previous day that SoftBank had agreed to sell the UK company to Nvidia for up to $40bn, in what could end up as the semiconductor world’s biggest-ever deal.
The deal was tantamount to dropping a bomb in the middle of the chip industry. Companies that license Arm’s designs — which are used in most smartphone processors and many other devices that require chips with lower power consumption — are worried they will be “disadvantaged” once the UK group falls under the control of one of their competitors, Mr Haas admitted.
Arm’s customers may find themselves at the back of the queue when trying to get the company’s newest designs, said Mark Lippett, chief executive of chip start-up Xmos. “You’ll find Nvidia will be the first out on to the market with the latest Arm architectures,” he said.
If it can head off those worries, however, buying Arm could set the stage for Nvidia’s next big act.
The first came when it repurposed its graphics processing units (GPUs), designed for video gaming, to handle data-intensive machine-learning tasks. That propelled Nvidia past Intel two months ago to become the world’s most valuable chipmaker. With ownership of Arm, it hopes to put its chip technology at the centre of the booming artificial intelligence market, ranging from cloud data centres to the many “smart” devices that are springing up.
The consternation caused by the deal springs from Nvidia’s attempt to mix two very different business models: selling chips and licensing intellectual property for other companies to make their own.
Mr Haas concedes that other than Qualcomm — the mobile chip company whose aggressive business tactics have often antagonised others that rely on its technology — no other major chip group has ever succeeded at both.
One difficulty is the differing requirements of each when it comes to developing, documenting and testing fundamental chip technology. This will make it hard for Nvidia to license its existing IP, said Woz Ahmed, head of strategy at Imagination Technologies, the UK chipmaker.
Mr Haas, who once worked at Nvidia, said this explains why a previous attempt by the company to license its GPU architecture to mobile chipmakers failed. As a result, it will be “some time” before Nvidia opens up any of its existing technology for licensing, and even then it will start with a narrow selection, he added.
A bigger problem arises from clashing incentives. Nvidia’s business revolves around selling chips for use in high-value devices such as gaming PCs or data centre servers. That gives it good reason to make use of Arm’s latest technology first, to get an edge over rivals that depend on access to the same technology.
Jensen Huang, Nvidia chief executive, says protecting the billions it spends on buying Arm gives the company a powerful economic reason to play fair. But there has been a resounding silence from big Arm customers including Apple, Qualcomm and Broadcom, with none publicly backing the deal.
Nevertheless, most customers are expected to go along with the status quo while they wait to see whether Mr Jensen lives up to his promises, said Mr Ahmed. One reason is that they have little choice. A rival, open-source chip architecture known as Risc-V has yet to advance beyond low-value chips.
Several industry experts argue, however, that Risc-V will now attract a wave of investment from companies seeking an alternative to Nvidia. The largest Arm customers will eventually switch to designing their own chip architectures to ensure control of their technology, Mr Lippett predicted.
But even losing an Apple or a Qualcomm may not matter much if Nvidia is able to use the deal to consolidate its position in a key market — the servers used in data centres — while making itself the technology axis for a booming industry of AI-powered devices.
As Mr Huang points out, Nvidia already did the hard work of rewriting all its software code to run on Arm-based processors when it backed a Fujitsu supercomputer last year that is now the world’s fastest. That will make it an easier step now to launch a server processor of its own, directly challenging Intel.
Nvidia’s second goal with the deal is to expand its technology to an ever wider range of devices. This will be driven by the growing need for AI “inferencing”, or applying pre-trained AI models to data gathered on the fly, as many day-to-day objects develop a basic level of intelligence.
In some cases, this will involve selling new chips of its own but often, particularly in cheaper devices, Nvidia plans to adopt Arm’s business model: packaging Arm’s processor designs with its own core technologies to create integrated blueprints for other chipmakers.
By owning Arm, Nvidia will be able to align all the companies’ technologies, such as their software libraries and developer tools, to make it easier for other companies looking to build on this foundation, said Chirag Dekate, an analyst at Gartner.
If Mr Huang is right, these chip designs could become the platform for the next wave of smart consumer devices. Those might include low-priced high-definition televisions and specialised tablet devices that come with the silicon smarts of advanced gaming computers, said chip analyst Patrick Moorhead.
Longer term, many everyday devices that lack screens will need a basic level of intelligence to understand verbal instructions, said James Wang, a former Nvidia employee and now analyst at Ark Invest.
This will mean a market of “hundreds of trillions” of computers, some day, that need his company’s technology, Mr Huang enthused this week.
But the first challenge will be to persuade the chip industry that Nvidia is more friend than foe.