Comment

FTC puts Arm lock on Nvidia

The GPU maker’s purchase of the UK chip designer looks further off than ever

The Federal Trade Commission may have driven the decisive stake through the heart of the plan by Nvidia to bolster its ambitions to become a full-spectrum computer designer. On Thursday, the US antitrust organisation said it would sue to prevent the merger going through, following a series of investigations being opened around the world to scrutinise the deal.

In the summer, Citi analysts saw the UK regulator clearing the deal as three sizeable customers in the shape of Broadcom, Marvell and MediaTek said they were OK with it going ahead. It is worth noting that Nvidia competes head-on with Marvell in the relatively new market of data processing units (DPUs), which are designed to speed up network switches. Geopolitically, the friction between the US and China seemed to tilt things in favour of Nvidia. Chinese companies fear having access to Arm being cut off after the acquisition, which is one thing that might make the US government give the action the nod. Why not seal control of a major processor architecture within the US? 

But the action by the FTC has caused analysts to slash the odds significantly. Citi now puts just a 5 per cent chance of Nvidia succeeding, according to Ian King writing for Bloomberg. 

There are two ways to look at the potentially anti-competitive nature of the deal. One is to look upon Arm purely as a vendor of processor cores. A few years ago, the case against a merger of a chipmaker and computer company with Arm would have been significantly stronger. The FTC argues Nvidia’s control of Arm would allow it to tip the direction of the company in its own favour and make it harder for other companies in driving-assistance systems, DPUs and cloud-computer makers to compete if they continue to use the Arm architecture. 

Arm still has the pre-eminent position in mobile devices, which curiously the FTC does not mention directly as being harmed by the acquisition. That may be a clue as to who has lobbied hardest to get the FTC’s attention. But Arm is getting stronger in server and communications switches. At the same time, the bigger trend is towards accelerator-heavy architectures as these potentially offer better energy efficiency. Those designs may use Arm at their heart but the RISC-V instruction set architecture (ISA) offers a viable alternative, particularly for the kinds of companies that are most likely to have complained directly to the FTC. 

The big advantage of RISC-V is that it is offered as open source and has fewer commercial restrictions than Arm. The flipside to that is that the control that Arm exercises prevents balkanisation and undesirable incompatibilities between different implementations. Arm is consolidating that with programmes such as ServerReady that make it easier for the customers of its customers to get Linux up and running on different processors and boards. Were nVidia to succeed in completing its purchase of Arm, a bigger shift to RISC-V seems all but inevitable. From an antitrust perspective, however, it means the option is there.

Potentially a stronger argument and one that matters very much to chipmakers is the FTC’s secondary choice: “The complaint also alleges that the acquisition will harm competition by giving Nvidia access to the competitively sensitive information of Arm’s licensees, some of whom are Nvidia’s rivals, and that it is likely to decrease the incentive for Arm to pursue innovations that are perceived to conflict with Nvidia’s business interests.”

Having your requests for tweaks to Arm processors winding up in the Nvidia archives is a major disincentive to use the architecture. And if you do not feel you have a choice but to use Arm, a major concern. However, chipmakers have learned to live with this kind of problem and may well have to get used to it in other areas.

Back in ye olden days – namely about thirty years ago – if a company without access to its own fab wanted to make chips, its management had to go cap-in-hand to a competitor that did own one and take the risk that it would reveal valuable intellectual property to that competitor’s design engineers in the process. This risk provided the tailwind behind Morris Chang’s decision to build an independent foundry supplier that would guarantee confidentiality to its customers. Your IP might be shared with TSMC but the foundry has taken great pains to ensure that it does not leak to other customers of the shared fabs.

Today, TSMC dominates the foundry market, and if you want an alternative source you do not much much in the way of choice. GlobalFoundries has opted to save money by not pursuing the bleeding edge of Moore’s Law. That leaves South Korean chaebol Samsung, which arguably competes with more companies than Nvidia in key Arm-based markets. 

Competitors may just have to learn to deal with each other in another way. Several of the markets named by the FTC are where the trend is towards multi-chip modules rather than pulling as much as possible into single chips. To save development costs, the chips that go into these modules are going to have to come from different sources who may well compete in other areas. The big problem with making and testing these modules is that you need access to a lot more information that chipmakers treat as confidential. This information is at a lower and more tactical level than what Arm picks up from its customer base but it is sensitive information nonetheless. The FTC’s action may be popular among certain chipmakers looking to compete with Nvidia but it may reflect an older landscape.

If the actions by the FTC and others block the deal, it’s bad news for SoftBank. Since the announcement of the acquisition, Nvidia’s stock price has soared for various reasons. Some of those have made Nvidia a more dangerous competitor to some companies, which has probably helped drive this action relatively late in the day. At its technology conference last month, Nvidia was touting its transformation into a much more vertically integrated computer maker focused on machine learning. The current fashion for all things AI means the deal is worth almost twice its value the day the plan to acquire was announced. It is hard to imagine anyone else being willing to front close to $75bn to buy Arm. And anyone who is would probably wind up being blocked as well. That’s a major headache for SoftBank and potentially for Arm’s management as well.

Though the original plan by SoftBank was to treat Arm as a strategic component, the attitude towards the chip designer has clearly changed in the space of just five years. If it continues to try to sell Arm, can SoftBank ever get a fair price for it? If you can’t sell something to just anybody, the purchase price tends to drop. The most likely option seems to be an initial public offering (IPO), seeing Arm return to the stock markets, though probably in New York rather than London. At that point, the business model will come under a far more intense spotlight, which may suppress the price that SoftBank can realistically charge. That may in turn force SoftBank to change direction and bring Arm back into the fold. But for a sector very focused on trade secrets, is SoftBank the best option for them long-term? 

The conglomerate has taken many forms since its formation. The diversity of its portfolio seemed to provide protection for Arm’s “Switzerland of the chip market” status, but is that guaranteed as SoftBank makes further purchases and divestments? The chipmakers may simply have to find ways to work around the trade-secrets problem with Arm much as they will have to with the way their chips are made and packaged.

Sign up to the E&T News e-mail to get great stories like this delivered to your inbox every day.

Recent articles