AMD's AI Accelerators and Market Growth

AMD's data center business doubled, expecting strong growth, with AI driving demand for their my 300 accelerator.

### AMD's Strong Q2 Performance and Future Roadmap in Data Center and AI
* 💻  AMD's data center business more than **doubled year-on-year** in the second quarter, with a **strong outlook for the second half of the year**.
* 🔢  Updated full-year guidance increased from **over $4 billion to over $4.5 billion**, driven by **hardware and software milestones**.
* 🧠  Lisa Su highlighted the importance of AI accelerators for both training and inference, with expectations that **inference demand will surpass training** in the future.
* ⚙️  AMD is addressing supply constraints by **expanding its supply chain and increasing overall industry capacity**.
* 📈  AMD is on an annual cadence for new accelerator releases, with the **MI 325 series** coming later this year, **MI 350** in 2025, and planning for the **MI 400 series** in 2026.
* 🖥️  The PC market is showing positive signs, with the launch of **AMD's Zen 5 family** expected to **boost performance** in the second half of the year.
* 🎮  Lisa Su discussed the gaming segment, acknowledging the **expected decline in revenue** in the second half of the console cycle.
* 🤝  AMD is making progress in the enterprise server market, with **one-third of new design wins** coming from customers who have **never used AMD before**.

What changed, whether it was supply or demand in shipping more than 300 family.
Well, good morning, Ed. It's great to be here with you today.
Look, we had a very strong second quarter, but very pleased with the
overall growth that we saw, really led by the data center business.
So our data center business more than doubled year on year in the second
quarter. And we see, you know, a strong second
half there as well as strength in PCs.
So now to your question about, you know, am I 300 and what are we seeing?
And I, you know, we're seeing an environment where everybody needs more
data center infrastructure. You know, computing is really the
foundation of what drives, you know, sort of all of these AI capabilities
going forward. You know, this growth has been well into
the qualifications at numerous of our top customers.
And you know, what we saw over the past

quarter, we passed some very significant
milestones in both. One, you know, ramping up our supply chain, but also
passing important, you know, data center customer milestones, where customers have
moved our product into production. So it's an exciting time right now.
You know, we did update our full-year guidance from greater than 4 billion a
quarter ago to greater than 4.5 billion. I think we're seeing great traction in
our hardware and software milestone. So overall, I'm feeling really good
about the setup going into the second half of the year.
Lisa, I'm really interested in kind of the design and performance differences
for all the air accelerators that are out there.
You and I have talked about the technology differences, but on the call,
you gave the example of Microsoft and its use in the quarter of the 300
generation. When you're speaking to a hyperscale,
when might they opt for using my 300

over Nvidia's H1 hundred, and I suppose
to a lesser extent Goudy. What is the specification or performance
driver that would make it a better solution in a certain use case?
Yeah, well, first of all, you know, I'd like to, you know, say that Microsoft is
a phenomenal partner. They're a great partner for us, you
know, across, you know, the data center as well as, you know, the client
businesses. And when you see, in particular, you
know, how complex these AI systems are, you know, you want to have, you know,
the best technology capability. You know, I can go into speeds and
feeds, but I won't. I think it's suffice it to say, you need this type of
computing for both training and inference.
In our case, you know, my 300 is the leading inference accelerator.
Today we have the highest memory, the highest memory, you know, bandwidth.
And that's really helpful.

You know, when you're asking questions to, you know, something like a transcript or llama model, and you want your answers to return faster, you want the fastest inference accelerator out there, and you know, am my 300 does very, very well in that framework. And then when you take a step back and just look at the overall A.I. market, I mean, the overall A.I. market is growing so fast, there is no one solution fits all. Actually, you know, you do have, you know, different workloads will be optimized with different hardware. So I think we have a tremendous opportunity. I think we're very excited about the overall market growth. And then as more customers get a chance to use our AMD instinct roadmap, I think we're getting very positive feedback. Lisa, you addressed head on on the call. I think that the investment cycle for AI has legs. It will continue. But just interesting what you just said. Can you kind of a portion of how much of

that spend or investment is going to
training versus inference at this stage? Yeah, absolutely.
So when we look at this cycle, you know, you and I have talked about this before.
I mean, we're forecasting that the overall market for accelerators, when
you look out and let's call it 27, 28 and beyond timeframe, you know, it will
be, you know, $400 billion plus, which is, you know, huge market overall. I
think in that you're going to need a lot of training, and you're going to need a
lot of inference. You know, today I would say that, you
know, most people would say that you're building foundation with training.
So training is a place where, you know, you want the largest models to really be
capable of, you know, of using all of the large data infrastructure.
But going forward, you would expect that, you know, inference demand will
actually surpass training.

Most people believe that. And, you know, I think this is, you know, we'll see how this plays out over the next couple of years. But you're going to need technology for both training and inference. And at the end of the day, you're also going to have a very diverse set of models and a very diverse set of needs. So you have the largest foundational models out there, but then you also have, you know, many enterprises are training on their own data and they'll use, you know, sort of a different type of infrastructure for that.
Lisa What's actually being done to address the supply constraints such as they are that impact your abilities with my 301. Well, I would say overall, when you look at the semiconductor supply chain, it's been amazing. I mean, we have substantially increased the overall capacity as an industry to address this very high air demand. We've also addressed from an AMD standpoint, our supply chain has been expanding every quarter.

You know, we announced in the second
quarter that we did over $1,000,000,000 of revenue in air accelerators, which is
a fantastic number. We see that increasing every quarter as
we go into the second half of the year. The supply chain will continue to be
tight as we go into 2025. And that just says a bit about just how
much demand there is out there. We have great partnerships with our
supply chain partners, and we're working on expanding that as fast as possible.
Lisa, the dominant player in the market for accelerators that go into data
centers, is committed to this cadence of essentially a new generation of chip
annually. What does AMD do to compete with that
and develop new products going forward? Well, Ed, what we see in the market, you
know, spending time with, you know, the industry leaders, is this market is
moving so fast that you have to be on an

annual cadence.
So, we announced last month that we have accelerated and expanded our roadmap.
You'll see us later this year. We're going to launch our May 325
series, which is the next next step in our roadmap.
Next year, in 2025, we will launch our Am I 350 series, which is a, you know, brand
new architecture. And we're already well on our way to our
more 400 series, which will launch in 2026.
So we are on an annual cadence. Not only that, it's not just about the
accelerator. You know, you know, CPUs continue to be
very, very important. We see that overall networking continues
to be very important. Overall system capability is also very
important. And we are investing in bringing all of
those capabilities together to really help our customers with the best
infrastructure, you know, for their AI workloads.
Lisa What I love about the Bloomberg

Technology audience is there's going to be a big body of people out there who think AMD and they think PC and gaming. You know, that's, that's kind of the reality of it. You painted this kind of positive picture for PC, at least for your company in the second half of the year. Is that a story of a market recovery and broadly the market doing well, or is AMD doing something specific to outperform in that environment? Yeah, we see a couple of things, Ed. First of all, the PC market is a good market. You know, the second half of the year is typically stronger than the first half anyways from a market standpoint. We see some positive signs in the market. I will say I'm very excited about APC's. I view APC's as an inflection point for what PCs can do in all of our daily lives. Whether you're talking about a corporate environment or a consumer home.

environment.
So that's a positive. I think we are also just launching a
whole new generation of chips. You know, we call it our Zen five
family, both notebooks and desktops. They are fantastic.
We're really happy with the reception for those.
And so we see that as an opportunity that is, you know, our product roadmap
launching usually gives us a lift into the second half of the year.
And then, you know, in the medium term, you know, going into 2025, you're going
to see a lot more APC's. You're also going to see, you know, we
believe in enterprise refresh cycles. So I think there are positive indicators
for the PC market. You know, going forward, would you be
able to give me some sort of forward-looking commentary on the gaming
business, in particular, when that might see some improvement?
Yeah. I mean, let me talk a little bit about
gaming.

You know, first of all, the gaming segment is a great segment. You know, we truly value the strong partnerships we have with the market leaders like Microsoft and Sony in the console cycle. Now, typically, these console cycles, you know, do have an ebb and flow. And so we're in the second half of the console console cycle, let's call it the fifth year. And so we do we have seen a decline in revenue. I think that was very much as expected. We would expect that, you know, it would take until the next refresh cycle for you to see a substantial change in that. But this is very much a market that we know we love and we believe has know great, you know, long term alignment, you know, with our roadmap and our IP. Lisa, I acknowledge you've had a very busy 2024. You spent a lot of time on airplanes. I think about Computex in particular while there you were on stage with a

number of partners.
And we talk just now about the sort of legacy gaming and PC business.
I just wondered if you have any sort of evidence or data you can give me that
those partners are putting AMD into sales channels where you might not have
any experience, for example, in like in the public sector or selling your
technology to governments rather than just the hyperscalers.
Yeah, Yeah, that's a great question. First of all, you know, we love our
partnerships. I think partnerships are, you know, what
makes this industry truly special, right?
Every company has their secret sauce, their great product.
But when you put it together and really bring the best of, you know, the the
industry together is when you get the, you know, the great payoff.
When you look at our business, first of all, you know, let me talk a little bit
about our, you know, enterprise server business.
Super pleased to see the progress that

we're making there.
When you think about, you know, the server CPU market, let's call it, you
know, outside of AI, there's a whole set of workloads that need the best, you
know, general-purpose computing. That's what we have with our, you know,
fourth-generation epic. And we just launched our fifth
generation, you know, epic or previewed our fifth-generation epic as well.
Very, very important, you know, foundation for, you know, these
enterprise data centers. We made substantial progress here in the
first half of the year. I would say one-third of our new design
wins are actually from customers that have never used AMD before.
So this that's exactly what you said is, you know, as we introduce AMD more to
the broader enterprise, you know, customers, we're seeing great traction
primarily because we have, you know, the best performance per dollar and the best
performance per watt, you know, out

there.
So we do see a tremendous opportunity in the enterprise market, both on the
traditional compute side as well as AI, as well as HPC.
So all of those are are great opportunity.
Lisa, I know you've got to go to your next meeting.
So very quickly, what's the talent story for AMD and getting the best people?
We are very, very happy with the talent story.
I mean, we have a tremendous engineering team.
They're doing great things on hardware and software, and we just announced our
intent to acquire a fantastic AI company, Sarlo AI, that that's based in
Europe. That is also, you know, adds great
talent to the company. So this is a war for talent.
And I would say we're very happy with, you know, the great engineering
capabilities that we have to move the roadmap forward.

AMD's CEO on Earnings, AI, Roadmap

Welcome to the Bloomberg Technology YouTube channel. Subscribe for original series and mini-docs on game-changing technologies, profiles on innovators, live conversations with experts, and breaking technology news & analysis from Bloomberg TV.

Bloomberg Technology

Strong Growth in Data Center Business and the My 300 Accelerator

The data center business more than doubled year on year in the second quarter, driven by the increasing need for data center infrastructure.

The company has passed significant milestones in ramping up the supply chain and in customer adoption of their products.

The My 300 accelerator is highlighted as the leading inference accelerator with the highest memory and bandwidth, making it a preferred solution for complex AI systems.

AMD's Positive Outlook on the Growing AI Market

AMD is optimistic about the rapid growth of the AI market, emphasizing the need for different hardware to optimize various AI workloads.

The company foresees a significant $400 billion plus market for accelerators, with a growing demand for both training and inference in the coming years.

AMD acknowledges the diverse set of AI models and needs, highlighting the expansion of the semiconductor supply chain to address the high demand.

AMD's Accelerated Revenue Growth and Product Development

AMD has exceeded $1 billion in revenue from air accelerators in the second quarter and anticipates continued growth throughout the year, attributing it to high demand and tight supply chains.

The company is strategically competing with Lisa, the dominant player in data center accelerators, by committing to an annual cadence of launching new products, including the upcoming May 325 series and the future Am I 350 series.

AMD is also focusing on the PC market, expecting a strong second half of the year and emphasizing the significance of APCs in transforming the role of PCs in both corporate and consumer environments.

AMD's Positive Outlook on Future Products and Partnerships

AMD is excited about the launch of its new generation of chips, expecting a boost in the second half of the year.

The gaming segment may see improvement after the next console refresh cycle, with AMD having a strong long-term alignment in this market.

The company values its partnerships and sees great potential in expanding sales channels, particularly in the enterprise server business and the public sector.

AMD's Fifth-Generation Epic and Expansion in the Enterprise Market

AMD has made significant progress with its fourth and fifth-generation Epic processors, gaining new design wins from customers who have never used AMD before.

The company's success in the enterprise market is attributed to its best performance per dollar and per watt, appealing to a broad range of customers for traditional computing, AI, and HPC workloads.

AMD is focused on talent acquisition, highlighting its impressive engineering team and the recent announcement of acquiring Sarlo AI, a company based in Europe, to strengthen its talent pool.