More details on Nvidia's acquisition of Arm revealed

Latest update time：2021-08-31 03:12

Reads：

Source: The content is compiled by Semiconductor Industry Observer (ID: icbank) from " seekingalpha ", thank you.

Nvidia recently announced its $40 billion acquisition of Arm, a UK-based company that provides chips that power nearly every smartphone, every new Mac, and are being heavily promoted into data centers, starting with Amazon's Graviton instances.

Although it is an extremely slow revolution, Arm-based chips are gradually eroding the x86 platform.

Phase 1: Intel and AMD missed the big smartphone opportunity, allowing Arm-based chips to go mainstream. This is where Arm is definitely ahead.

Phase 2: Arm-based chips replace x86 in PCs. This is already happening. Apple will have all Macs running on its own Arm-based chips within a few years.

Phase 3: Arm-based chips replace x86 in the data center. Again, this is already happening with Amazon’s Graviton2 instances.

More than 20 billion chips are now shipped each year using Arm IP. 70% of the world’s population owns a product based on Arm IP.

Transaction Details

Let's start with what NVIDIA earns, what it pays, and how it pays.

They didn't build an IoT division. Rumor has it that Softbank wanted to hold on, and they already did. That created more problems.

Arm screenshot. NVIDIA didn’t get the IoT stuff highlighted in red.

On top of that, NVIDIA gets the IP for instruction sets, core designs, and reference designs for smartphones, PCs, AI, and data centers. They also get the best of the UK tech company (one of the most important in the world) and all the talent.

They reportedly paid $40 billion, but that was not the case. Details of the deal show:

Nvidia only had to pay $2 billion in cash when signing, which was mostly a breakup penalty. As of the end of July, they had $11 billion in cash and investments, but that was offset by $2.4 billion in current liabilities.

They are expected to close another $10 billion deal in 18 months. That may be an optimistic timeline. But regardless, they have $5.6 billion in operating cash flow in TTM, so they have more than enough money to pay.

The bulk of the deal will be paid for with 44.3 million new NVIDIA shares at a stock price of approximately $485, for a total of $21.5 billion, which is 89 times TTM earnings.

In addition to the 44.3 million shares, Arm employees will receive another 3.1 million shares worth $1.5 million. Overall, this represents a 7.1% dilution for existing shareholders.

SoftBank has another $5 billion in cash coming in if Arm hits its performance targets. We don't know what those targets are.

So to sum it up, “$40 billion” actually means:

$12 billion in cash.

$23 billion in stock priced at a 16-year high earnings multiple.

$5 billion in cash that may or may not exist.

This looks like an NFL contract where they announce an unprecedented total value and the player only sees a portion of it.

Will it happen?

This will be one of the most closely watched deals ever, because there are so many stakeholders here. This deal is widely rumored across the industry, and it will cause trouble for anyone who makes chips, smartphones, PCs, or servers. In addition, there are national issues here for at least the United States, the United Kingdom, and China.

Bank of America's Vivek Arya put the issue clearly on the call:

“My question is about regulatory and ecosystem barriers. Given all the trade tensions, how do you see regulators responding, particularly in China? And what about Arm’s customers, several of whom are your competitors? And you know, they have unrestricted access to Arm’s technology. So what kind of barriers do you foresee from a regulatory and ecosystem perspective?”

NVIDIA CEO Jensen Huang did not respond to many questions about regulation, and instead focused his answers on Arm's customers, which include Apple, Qualcomm, Broadcom, Samsung, Intel, AMD, Huawei, ZTE, and pretty much anyone in the business of designing or manufacturing chips for phones, tablets, PCs, or servers. In this world of technology, there are a lot of things to lean on. Apple is simply betting the future of the Mac on the Arm instruction set. That fact has not escaped the control of the two companies.

The bottom line is that Arm has never been one to compete with its own customers, but that’s changing.

Huang Renxun gave the following answers to these questions:

“Simon [Segars, Arm CEO] was contacted by customers after hearing the rumors, and he spoke to many customers. And the fact that we announced the transaction also tells you something that we believe customers will be satisfied with our true intention to keep the platform open and neutral. We acquired the company and paid a very high price as you know because Arm has a huge network and ecosystem, and we like the business model…

We will continue to be open and fair in the future, providing all IP to all Arm customers, which I think will be very exciting. In terms of regulation, they are focused on professional competition, and we will bring more choice to customers. For the first time, we will have a credible plan to turn Arm CPU cores into comprehensive data center platforms...

That combination really allows us to focus and capture value and deliver it to the market, which is a very strong, very capable alternative platform. So regulators like to see strong competition in the market. So we are customers, we are competitors, and regulators should be very supportive of that.”

So he tried to frame it as x86 vs. Arm, which would provide more competition, not less. But a merger doesn’t really do that, it just changes the address of a company’s headquarters. Nothing he said would rule out competitor nightmares. For example, I didn’t hear him promise:

Keep the licensing terms unchanged for a period of time.

Do not force the licensee to license the entire IP suite, even some IP that is not needed.

License future developments under open, fair and non-discriminatory terms.

They talk a lot about AI. Will they be willing to license those developments on reasonable terms, not tied to other IP? How will they deal with AMD, their main GPU competitor? If they start developing their own data center chips, as they have indicated, how will they deal with Amazon or smaller companies like Ampere and Nuvia that are competing for chips?”

The phrase “open and neutral” is nice, but there’s a lot of wiggle room in those quotes. Just look at what Facebook has done with seemingly benign words.

Stacey Rasgon of Bernstein is also somewhat skeptical:

“I hear you talk about the potential value that customers will have. But at the same time, how does that really work in practice? I mean, it’s a practical consideration that you understand everything first. You’re going to control the direction of innovation in the company. You need to understand the roadmaps of all your customers first. So how do they really fit into this, especially in products like high-performance computing and servers? I wanted to ask if this might be within the scope of some of your previous efforts to drive the IP licensing business, which, as I recall, did not go well.”

Huang’s response to this question was basically “trust us.” He first described their current licensing approach, which allows customers to choose the desired stack layer. He then said:

“For Arm, it’s just a way of delivering technology, which means if customers want our technology already fixed and hardened in a chip, that’s fine. But if they want it to be flexible and malleable, that’s fine too. I think that business model, the modern business model, is not as strange as it seems. You know, we’re into the cloud, we have infrastructure as a service, platform as a service, software as a service. Platforms make it available to you wherever you want, just like you’re involved in it.

I think fundamentally, NVIDIA certainly continues to protect the most important thing to us, which is our corporate reputation, and we have to be a company that our customers and partners can trust…So we think this business model is part of what we want to bring to the company as it expands, and we think it’s good for the business, it’s good for the economy, it’s good for our strategy, and it will be the backbone of the expanded computing platform. If we believe in that, as I do, and we’re committed to openness, we’ll do it.”

“Trust us” is the same for Arm’s customers as it is for NVIDIA’s competitors. The difference between NVIDIA today and NVIDIA plus Arm is that they have the power to dictate the market that they didn’t have before. And with that comes a whole new set of incentives. Even if they can trust Huang and his team, what about his successors and their successors’ successors?

Again, Huang is really criticized on the regulatory issue. They have promised to keep Arm's headquarters in Cambridge, UK, and invest more in it, so this may take care of the UK regulators. But this does not mean that there will not be opposition. Hermann Hauser, one of the co-founders of Arm, has publicly stated his opposition to the merger. His three arguments:

1. NVIDIA will reduce employment in the UK. But NVIDIA seems to want to do the opposite.

2. Selling Arm to NVIDIA will undermine the foundation of Arm's business model. In the past, this model could be called the Switzerland of the semiconductor industry, where they traded on an equal footing with more than 500 licensees, most of whom were NVIDIA's competitors, including British companies.

3. His main argument is that the crowning jewel of British technology will be in the hands of an American company, subject to the control of the Treasury and Commerce Departments, particularly the Treasury Department's Office of Foreign Assets Control.

The US is going to be trickier because there will be a lot of US companies that are Arm customers who will be fighting for a range of unfavorable outcomes for regulators. Hermann Hauser’s second point is even more important in the US. If approved, NVIDIA will likely have to sign a consent decree or similar order that severely limits their authority. Unless their lawyers are not as competent as GM was in the Nikola case, I have to imagine they understand this. But even with a consent decree, Arm customers will still be subject to NVIDIA’s willingness to comply and the DOJ’s willingness to enforce.

Huang tried not to mention China in his response, and for good reason. Evercore’s CJ Muse brought him back to the China issue again, and Huang responded with the following non-answer:

“For China, it’s important to realise that ownership of IP is not ownership of a company. The origin of IP is the relevant factor in export controls. Arm’s IP started, was created, was developed in Cambridge over three decades. So the IP will essentially remain in the UK. Arm’s headquarters will be in the UK. We will continue to advance this technology in the UK.”

What does NVIDIA get from this?

This is a big question for me. Let’s start with their CEO Jensen Huang’s explanation on the conference call:

“We are at a moment when Moore’s Law is slowing, but the era of AI is driving a tremendous acceleration in the demand for computing. Because traditional architectures can’t keep up, a new approach to computing is needed – accelerators. NVIDIA’s accelerated computing platform has risen to the challenge and is leading the way forward. That’s why this transaction is such a compelling and complementary backdrop for both companies. Our combination has several exciting growth drivers. The first is Arm’s broad ecosystem. We can bring NVIDIA’s GPU and AI technology to large end markets including mobile devices and PCs.

Second, in the data center, NVIDIA will accelerate Arm's R&D efforts to meet cloud computing customers' demand for higher-speed Arm CPUs - Arm's server CPU roadmap, NVIDIA AI will provide excellent support for all data center CPUs, x86, power and Arm. Third, we can accelerate the Edge AI and IoT roadmaps and growth trajectories for the next wave of computing.

The combination of the two will boost our developer ecosystem, which is the cornerstone of our computing platform. Arm expands NVIDIA’s developer reach from 2 million to over 15 million.”

Then, later in the call, he tied in the acquisition of Mellanox and its new Bluefield data processing unit (DPU).

“CPUs are really great and do low-latency, single-threaded, predictable latency type of processing. GPUs are really great at high-throughput data processing. And DPUs are really good at networking and serial data movement, security processing. Those three types of processors are going to define the future of computing. That’s why we’re so excited about this acquisition, that with Arm, NVIDIA and Mellanox, we have three computing platforms, one for networking, one for high-throughput computing, accelerated computing and AI computing, and one for CPUs and single-thread computing.”

There's a lot there. But when we look at it, ask yourself: Can NVIDIA get this for less than $40 billion, or what will it end up costing?

The first point he made was to bring NVIDIA to Arm customers. This would give customers including AMD and Qualcomm ideas. For example, could they try to force Apple to license GPU technology that they don't need and the instruction sets it uses? This is exactly what Qualcomm is doing with Apple, but Apple is not happy about it and is spending billions of dollars to stop being a Qualcomm customer.

The second point is the focus on the data center, which will cause mixed reactions. On the one hand, companies that are working on Arm data center CPUs, such as Amazon, Ampere, and Nuvia, will be excited about the widespread attention paid to data center cores, but be wary of competitors NVIDIA getting one step ahead or even more.

Then there are Qualcomm, Huawei, Samsung, and every phone maker not named Apple, all shouting in unison, “What about us?” With the exception of Apple, which only licenses the instruction set, the entire smartphone industry is highly dependent on Arm to drive these cores forward. NVIDIA, through all of this, has said very little about smartphone SoCs.

His next point about the Internet of Things raises even more questions, so many that you’ll find them below, so bear with me for a moment.

He then trained Arm developers.

But the bone of contention is Moore's Law, which states that the number of CPU transistors doubles every two years. It no longer works for Intel and everyone else. A silicon atom is 0.2 nm wide. The smallest transistor in a commercial product was announced at Apple's September event, and its 5 nm means 25 atoms. That can only go so far. Multi-cores have long solved this problem, but single-core performance isn't as fast as it once was.

The core packing will continue. Early next year, Ampere will feature 256 Arm cores in a two-socket server design. But while the larger project remains the base of the stack, it’s moving beyond the CPU in the data center. On top of that, you take GPU accelerators for high-throughput tasks and add in DPUs that push data to the center at very high speeds. What ties everything together and makes the whole thing work is AI, which is already a core competency for NVIDIA.

But back to the question: Do they need to spend $40 billion or the final revenue? To almost all questions, the answer is "no". Imagine that instead of announcing this plan, they announced a plan to work hard on Arm development.

1. Continue to recruit and poach top talent like crazy.

2. Announce an annual conference specifically for Arm developers modeled after Apple’s annual Worldwide Developers Conference.

3. Show them the roadmap of how the CPU-GPU-DPU-AI stack works.

3. Make it a focus in company communications.

That might make things look a lot better.

They basically own all of Arm's core technology licenses as well as the architecture licenses for V8 and I'm sure V9 is coming out soon, so I guess the question that Jensen gets from people is, why acquire the entire company, right? What is your expectation for CPUs and GPUs in the data center to work closely together, or putting IP on client and Edge devices? Isn't that something that you couldn't already accomplish by just being an influential member of the Arm partnership and doing your own chip architecture licensing implementation?

Huang gives the same non-answer I already quoted because he can’t tell the real reason. What does NVIDIA get from this cheap strategic acquisition that it can’t? Control of the roadmap, and first access to it. Ultimately, that’s why it’s worth it to NVIDIA, and why the prospect has so many stakeholders so scared.

IoT and RISC-V risks

NVIDIA did not get the Arm IoT division that SoftBank retained. A very interesting tidbit from the deal is that Arm’s EBITDA margins went from 15% to 35% excluding the IoT division, so this is a huge drag on the combined company.

However, this divergence raises more questions about how this feature will work and what it means for the platform. For this, we need to talk about RISC-V.

RISC-V is a very powerful open source instruction set that is becoming a serious competitor to Arm in the IoT space. Since IoT SoCs are dedicated controllers that do not need to be purchased from a third party like smartphones, the general advantage of Arm disappears. By choosing RISC-V instead of Arm, companies can save millions of dollars in licensing fees. Just like the Linux space 25 years ago, companies like SiFive are filling the support gap in the open source ecosystem.

First, Arm responded by allowing smaller companies to start free trials of lower-level IP, with payment only made when production began. In addition, they made the low-level instruction set open and editable by customers, which is one of the advantages of RISC-V.

Therefore, three questions are raised here.

1. NVIDIA purchased the instruction set that the IoT division relies on. What are the details here? Does SoftBank get a free license? Will they fork the instruction set, or do they plan to fork the instruction set when the deal is completed? Doing so would undermine one of Arm's strengths, which is a common instruction set across multiple device types.

2. If the deal goes through, NVIDIA and SoftBank will become competitors in the IoT space. How point 1 will be combined to solve this problem is anyone’s guess.

3. But there is a low probability risk that Arm customers will be turned off by this deal and, fearing to rely on NVIDIA's credibility, turn to RISC-V. Again, the probability is low, zero in the short and medium term, but it still exists.

This part has been largely ignored, but has the potential to cause the most future problems for everyone involved if the Internet of Things really takes off.

opportunity cost

Opportunity cost is one of those things that people talk about a lot, but don't always factor it in. These are the costs of what you couldn't do because you did something else.

Any deal of this size comes with a huge opportunity cost. Not only is the money NVIDIA spends here, which at the end of the day isn't much, but the attention and focus will be drawn from other things they want to do. This may prevent NVIDIA from making bigger moves for the time being.

But the bigger risk here is that they spent 18 months trying to get this deal approved by regulators, and if that ultimately fails, the $2 billion is mostly a breakup fee, and they will have wasted 18 months and a significant portion of the leadership's attention on a deal that never happened.

Therefore, regardless of how this situation develops, the opportunity cost for NVIDIA in doing this will be significant.

★ Click [Read original text] at the end of the article to view the original link of this article .

*Disclaimer: This article is originally written by the author. The content of the article is the author's personal opinion. Semiconductor Industry Observer reprints it only to convey a different point of view. It does not mean that Semiconductor Industry Observer agrees or supports this point of view. If you have any objections, please contact Semiconductor Industry Observer.

Today is the 2440th issue of content shared by "Semiconductor Industry Observer" for you, welcome to follow.