NVIDIA’s pending acquisition of Arm melted tech social media last night and today. Anytime anyone throws down $40B, it tends to get attention. Add in the allure of creating an AI behemoth that can tackle everything from the edge to the enterprise data center, and it’s fascinating. It’s also a courageous move that could alter intellectual property (IP) licensing – if they can get past potential channel conflict.
Customers hate it when their supplier sets up shop competing with them. Combining channels and business lines via acquisition should open more opportunities in the long run. But in the short term, it can create competitive situations that didn’t exist before. If anything resembling conflict develops, customers reevaluate their relationship with that supplier.
Arm’s success is due to more than outstanding technical know-how and strong execution. Those strengths set Arm apart from other processor IP suppliers. Semiconductor IP is a tough business, though. Customers not only have to use your stuff, they must be able to afford success, and see a compelling roadmap.
Arm architecture took off when customers figured out the royalty-based licensing model was a win-win. Upfront investments for the IP started low, and once a product sold with royalty costs baked in, licensing was easy for customers. Even as higher performance IP brought higher costs, customization benefits got stronger. Later, Arm created the architectural license for Apple and others, with hefty sums for broad IP and tool access.
Which brings me back to the $40B. Either NVIDIA sees a big ongoing revenue stream from Arm IP licensing, or it sees a gigantic AI opportunity and paid for IP and expertise. There are big brains at NVIDIA, right? They must have evaluated channel conflict risk versus licensing.
Totally amazing that none of this would have occurred to the company that has been an ARM licensee for 15+ years.
— Steven Sinofsky (@stevesi) September 14, 2020
It wouldn’t be the first time a big tech company underestimated how its customer base would react to a big acquisition. Google made a splash in buying Motorola, then had to unload them like a hot potato when its mobile device Android customers pushed back en masse. Palm suffered after bringing Handspring back into the fold when it tried to sell both devices and operating system software. More recently, Qualcomm failed to clear Chinese concerns in buying NXP and abandoned the deal.
This isn’t about the Apples, Qualcomms, and Samsungs of the world. I see this having no impact on mobile business, where Arm rules and smartphone chip development costs are in the billions. NVIDIA deemphasized mobile after Tegra, and there is zero chance they displace Apple designs in Arm-based Macs. One wonders how much bad blood still exists after the 2006 PortalPlayer incident where NVIDIA got punched in the face by Samsung over the iPod. No matter, none of that is NVIDIA’s target now.
Where this does matter is where NVIDIA is focused. Self-driving vehicles making heavy use of AI. Edge computing where IoT data is being pre-processed before the cloud. Data center applications for AI. These segments are spawning startups, and many of them are or might be Arm customers. There may be merchant chip companies, or custom silicon solution designers.
Channel conflict depends on how NVIDIA chooses to run the combined business. One choice is firewalling, where NVIDIA leaves Arm operating as an independent subsidiary. It worked with Arm as part of SoftBank because most customers saw SoftBank as an investment house, but NVIDIA won’t be viewed that way. Another choice is for NVIDIA to avoid branded system-on-chips – oooops, the Orin SoC is out and there are certainly more chips coming.
There’s also the RISC-V dynamic. Before RISC-V, there were few serious alternatives to Arm for processor IP. ARC, MIPS, and Tensilica are still out there, and they all have a niche. RISC-V has the open cachet many designers look for today, and they are building momentum. They’ve adopted neutrality by moving their incorporation to Switzerland, more than symbolic.
Arm still dominates processor IP, and NVIDIA is assuming inertia is on their side. It’s a huge undertaking to switch out of the Arm ecosystem. Besides dealing with chip-level stuff, you need to look at EDA tools, software, and foundry partners. Most Arm customers aren’t ready for that risk; the few that are already stuck a foot in the RISC-V pool before this acquisition. I’m also hearing there’s some legalese in Arm contracts that makes defection a bit less likely.
And there’s the possibility that NVIDIA doesn’t care. Not because they dislike existing Arm customers, but because they see a bigger opportunity in Intel’s data center fortress. If they manage to switch say a third of a $15B market (conservative estimate for 2025) that’s a big number and would offset losses of AI startups. Big if, but Intel is emitting signs of vulnerability we haven’t seen before.
Acquisitions can only be evaluated a few years down the road, when the assumptions are tested and the side effects are understood. I’m rooting for this – I have a lot of acquaintances (and former customers, in the interest of disclosure) at Arm and want to see these folks succeed. It could change the landscape and shift how IP licensing is done, again, and make my channel conflict concerns unfounded.
I’ll admit, I didn’t see anything like this coming when I finished “Mobile Unleashed” nearly five years ago now. I saw Arm after the IoT, and the SoftBank acquisition was a bankroll with deeper pockets to fund development. Another semiconductor giant stepping in the second time Arm went on the block is a sign of change.
We should check back in two years from now and see how this is going. Until then, keep an eye on if channel conflict develops and gets significant, or if the bet on AI grows results quickly.