The new path for Arm is beginning to come into focus.
Weeks after Nvidia’s $40 billion bid to acquire Arm from SoftBank collapsed, the appointment of Rene Haas to replace longtime chief executive Simon Segars has set the business on a fresh trajectory. Haas appears determined to shake up the company, with plans to lay off as much as 15% of the staff ahead of plans to take the company public once again by the end of March next year.
It’s hard to argue that Arm isn’t thriving. Last week, the company reported adjusted profitability of $1 billion on revenue of $2.7 billion for fiscal 2021, the latter of which jumped a healthy 35%. Arm’s adjusted profit number leaves a lot to wonder about, as the company didn’t disclose what it had moved around to arrive at the $1 billion figure (the company did not respond to a request for further details). Arm counted $1.13 billion in licensing revenue, which comes from companies like Apple that pay to use Arm designs for their custom chips, and $1.54 billion in royalty sales. Throughout the year, the company said that 29.2 billion Arm-based chips shipped.
Protocol caught up with Haas in San Francisco three months after he took the top job and after nearly 9 years at the company, most recently running its intellectual property group. He discussed Arm’s future plans, its research plans and upstart RISC-V, which makes chip designs that compete with Arm. Haas has his work cut out for him: In a now-deleted blog post from its corporate site, his predecessor described going public as an existential threat to the company. An Arm spokesperson also declined to comment about the deleted post.
This interview has been edited and condensed for brevity.
Your predecessor Simon said last year, in no uncertain terms, that if Arm went public, it would die. Now Arm plans to IPO, and I know you’re shedding some of the workforce, but it doesn’t solve the fundamental problem that Simon articulated — that the pressure to achieve short-term revenue growth and profitability would suffocate the company’s ability to invest, expand, move fast and innovate. So how are you going to make it work?
It’s a fair question, and I’ll answer it. Talking about the results for a moment, because that does segue into it — we’ve never done over $2 billion of revenue in our history, so now we did $2.6 [billion]. Our non-royalty number was over a billion. We’re never done over a billion, ever. It was up 60% year-on-year, which is kind of the indication for what the demand for products looks like going forward. And then our royalties were $1.5 billion, also a record.
I took over the [intellectual property] business in 2017, and we did a kind of fundamental pivot away from the general purpose processors and moved more to market-specific products. At the same time, we started shedding products: display IP, video IP, that weren’t highly differentiated and commoditized. We really doubled down what I would call the compute platform — the CPU, the GPU, and the business around that.
The restructuring was really not about trying to fit into an EBITA envelope. It was a combination of looking at whether we have the right profile of [expenses] versus research and development. We need to invest more in R&D, so part of it was creating space to invest more, and that gives me a high degree of confidence — IPO or non-IPO — that we have a very healthy business.
This year, we’ve started to transition to v9 [architecture] royalties, but not in a big way, that will start next year. V9 royalty rates are better than v8, so that gives us confidence. And at the same time, these new markets with hyperscalers such as Amazon and Microsoft, automotive and such, I think we can manage that balance.
It’s a tricky balance, and I’m not going to take issue with what Simon said because those are his words, not mine. But I’m confident that we’ve got a strong business.
You’re thinking of investing more in research and development, in expending more resources there: How does Arm go about doing that? How do you plan to allocate R&D dollars? Some companies have made specific performance targets, such as an 1,000x increase.
We’re very focused on performance efficiency. Because I think one of the things that’s going to drive our growth, we’ve already seen: more and more chips are using more and more cores.
In a multicore system, you need a high degree of efficiency around performance per watt. And if you think about the data center, or the EV, or the base station, those all need a heck of a lot of performance, but they really need performance per watt. So if Intel is saying 1,000x better performance, we’re certainly going to be competitive on performance, but we’re going to be relentless on performance per watt.
Looking at the data center — if a new data center is going to be built in Ireland, let’s say, for them to get the land and an energy contract, only so much square footage is going to be allowed, [and] there’s going to be only so many megawatts they’re allowed. If they’re going to have no compromise on performance, it's really going to be around performance per square foot, and performance per square watt. This is why when [Amazon CEO] Andy Jassy stood on stage and said, “Why Graviton2?” He talked about 40% better performance in the same power envelope.
More broadly speaking, for us, it’s all about the compute platform, in terms of the GPU, the CPU and machine learning. Arm did have a lot of other products that were around the periphery, which took a lot of engineering work, and I don’t think we were highly differentiated or added a lot of value. Whereas on the flip side, the software investment required for these ecosystems is kind of insatiable. And that's also gonna be a high degree of focus for us.
Arm used to develop cores and then let other people build their own chips: How is that changing over time? Does Arm develop more of the final design for customers now than it used to?
Increasingly, what we're finding is that throwing that piece of IP over the wall isn't going to be enough to ensure a world-class product. There's a lot of things around system design that are increasingly important: the interconnect, the physical design, the memory subsystem. We are being much more prescriptive about how to build a [system-on-chip] using Arm [designs].
As opposed to just a straight IP licensing model — here's the recipe, go forth and build — we're now doing things around subsystems that have essentially allowed people to build better SoCs. That will also give us better performance and performance per watt, because we can now guarantee a certain threshold of performance. We understand that you’re building in TSMC, you’re building in Samsung, you’re building in GlobalFoundries — we do work with the library, and we’re doing more of that ourselves.
When you think of high-performance systems, they’re really hard to build. Our world is disaggregated: there’s Cadence tools, there are Synopsis tools, there’s TSMC, there’s a substrate guy. If you leave it up to everybody to figure out what their secret sauce is, you’re going to have some compromises. Since the performance engine is most tied to the CPU, we feel we have an obligation to be a bit more prescriptive in terms of how those products are built.
Is exerting more control over exactly how products are built more lucrative?
It could be — I can’t say, it’s too forward.
With only two dominant chip designs, who do you see as Arm’s biggest competition at this point?
If you just think about the CPU business, the way to think about competition is the instruction set architecture. There are really only three that are in vogue. There’s x86, which only two companies in the world [Intel and AMD] can build. So it’s a competitor, but not in the classic sense, because it’s not open.
There’s only one other open ISA, and that’s RISC-V. And RISC-V is absolutely a competitor to what we do, and they’re unique because they’re an open-source implementation. They’re a very interesting competitor because one of their greatest strengths could be their greatest weakness.
They’re open source, so anyone can build anything they want with it. It can be modified, it can be extended, everywhere in the world could have their own RISC-V flavor. But the more you differentiate, and the more you create different extensions not adopted by everyone in the ecosystem, you get a high degree of potential for software fragmentation.
For RISC-V, the jury is still out — it’s early. There’s a lot of momentum around RISC-V, no doubt about it. Where we see they’re getting traction is in what I would call the deeply embedded space, where the software externally doesn’t really matter. It’s controlled, in a black box. There’s no developer, there’s no software kits in that world. For the hyperscalers, there’s a lot of work that needs to be done around the software optimizations and workloads. And then you get into the situation of if everyone would be happy with their own kind of risk-five implementation. There has to be a standard.
I think the RISC-V thing will be a very interesting story to see play out.
How do you think about your relationship with Intel? On the one hand, the company is a competitor, but on the other hand, it has opened its factory doors and is willing to make anyone’s chips.
People would be surprised that Intel is a big customer of ours: one of our largest ones, believe it or not. I would love to see Intel Foundry Services be wildly successful, because when you take Arm, which is the most ubiquitous processor in the world, there’s going to be a marriage there somewhere. They cannot be successful without having a strong offering and being very competitive with Arm. On the flip side, for IFS to be successful, we need to work really closely with them. And we do.