AMD CEO Su Talks About Huge Chip Deal With OpenAI

Lisa Su
2 followers
Advanced Micro Devices has signed a deal with OpenAI to deploy 6 gigawatts of AI chips over multiple years, in a deal that could generate tens of billions of dollars in new revenue. AMD CEO Lisa Su and OpenAI President Greg Brockman join “Bloomberg Tech” to discuss the deal and the plans to finance it.
00:00 AMD chip deal
00:30 AMD CEO Su
01:00 OpenAI's Greg Brockman
02:00 AMD's commitment to OpenAI
03:15 the future of compute
04:05 What will AMD and OpenAI do together
06:30 Su: This is a win for everyone
#lisasu; #amd
Let's frame the opportunity, Lisa.
You know, the market reaction is very clear,
but for AMD and the AI industry at large,
what do you think this represents?
Well, look, this is a huge milestone for AMD.
We are so thrilled with the partnership
with the OpenAI team,
and it's also a huge moment for the AI industry
because when you get right down to it,
you need more AI compute.
I mean, that's where we are today.
Compute is the foundation
for all of the intelligence we can get from AI.
And we are a compute provider,
we have spent years on our roadmap,
we've spent years working with OpenAI and the team.
And, you know, together now we're embarking
on a massive build-out of six gigawatts of AI compute.
And it's a big deal for us,
for our shareholders, for our teams,
and for really, you know, the partnership
and the overall AI ecosystem.
Greg, I said at the top, the focus is inference.
I think that's really important to be specific
about what you will do with this capacity.
So literally explain that part.
And I'm conscious that, you know, in the first instance,
the first target is one gigawatt,
and then eventually six gigawatts.
But what will you use it for?
Well, I think that the world continues to underestimate
the amount of demand for AI compute, right?
That just, we've seen this explosion of demand
with things like ChatGBT, you know,
we're at 800 million weekly active users now.
This product didn't even exist three years ago.
And we're in a position where we cannot launch features,
we cannot launch new products
simply because of lack of computational power.
And we see these models continuing
to get exponentially better.
And I think we're just heading to a world
where so much of the economy is going to be lifted up
and driven by progress in AI.
And so we're very much heading to a world by default
that I think looks like a compute desert, right?
That there's just not enough compute to go around.
And so we're trying to build as much as possible,
as quickly as possible.
So we're starting with one gigawatt
simply because you got to start somewhere.
But honestly, we're building as fast as we possibly can
and trying to bring as much computational power
to bear for the economy and for the world.
Lisa, this is such a big commitment
to Instinct in particular.
As a customer, does it make OpenAI the largest
for that particular product?
Well, this is certainly the largest deployment
that we have announced by far.
I mean, you know, six gigawatts of compute.
As Greg said, we're going to start with the first gigawatt
in the second half of 2026
on our new next generation MI450 chip.
I think the thing to understand is,
these types of partnerships actually take years
to really get comfortable with the idea
that we're going to go all in together.
And this is an all in partnership
in terms of building out the AI compute
that OpenAI needs for everything
that they're offering to the world.
So yes, it's a huge deal.
And it also says a lot about how much needs to come together
for this entire ecosystem to operate.
So we are setting up,
certainly there's a lot of engineering work,
but our teams are working together on hardware, software,
we're ensuring the supply chain,
all of those elements are set up
and ready to deliver on this massive commitment.
Greg, talk us through a little bit about
the players that you need to also lean on.
This has been years in the making, as you say, with AMD,
but what other cloud providers were involved?
How are you thinking about this working
with an Oracle or others out there?
Yeah, we really think of this as an industry-wide effort.
And in general, we think that compute is something
that does require the entire supply chain to really wake up
and to really start building
much more than people were planning on.
I think this starts from energy
to try to get far more power to be built.
Things like nuclear, I think,
are going to be very important to come online.
The cloud providers are an important part of this as well.
So we're going to be deploying AMD in our own data centers.
We'll be deploying them together with cloud providers.
We have a deal with Oracle,
lots of other cloud providers out there.
You can really see that we're very much in the,
we just want compute, as much compute as possible.
We think this is important for the economy.
We think this is important for the nation.
We think this is important for humanity.
And so really we're working with everyone in this whole industry
in order to get as much compute power online
as quickly as we can.
Lisa, I'm sorry, specifics.
Where is this data center going to be?
Is it one single site?
Is it Oracle that will partner with you on this?
Well, actually, what this really is
is an announcement of what AMD and OpenAI are going to do together.
OpenAI has a lot of partners in terms of where they deploy.
I imagine a lot of it will be in cloud service providers.
It's really up to OpenAI and Greg and Sam and the team.
But the way to think about it is for this amount of compute,
it's going to have to be in a lot of different places.
I mean, it's a massive amount.
Multiple locations.
I would imagine multiple providers.
To really get this online as fast as possible.
Greg, there is a lot of focus on where OpenAI
is going to get the money from to fund all of this.
Sam Altman's big picture commitment is well documented, right?
And the numbers, to his mind, are in the trillions.
But have you specifically thought about debt financing
for this relationship with AMD?
Have you thought about doing a specific equity raise?
You are very committed across multiple projects.
The way that I would look at this is that AI revenue is growing faster
than I think almost any product in history.
And that ultimately, at the end of the day,
the reason this compute power is so important
and so worthwhile for everyone to build
is because the revenue ultimately will be there.
Now, as a company that is trying to move as fast as we can,
we look at everything, right?
We look at equity, debt.
We look at trying to find creative ways of financing all of this.
That's been actually a huge focus of us for the past couple of years
is thinking about how can we possibly build the amount of compute
that is required in order to really transform this whole economy
into an AI-powered economy.
And so I think you'll see lots of creative ideas.
But fundamentally, I think at the end of the day,
it is because we believe that the revenue is there.
Well, I'm sorry to jump in and interrupt.
And Caro, just forgive me on this one.
The condition of AMD issuing the stock to OpenAI
requires you to spend money, basically,
because you have to deliver that gigawatt of capacity first.
Lisa, I have to ask you if you have assurances that OpenAI is good for it.
Well, let me be clear.
I mean, this deal is a win for AMD.
It's a win for OpenAI.
And it's a win for our shareholders.
And that's kind of the way we put this together.
I have full confidence in OpenAI, Sam, Greg, Sarah.
I mean, this is a massive opportunity for us.
Right now, right here, it's about who has the most compute
and how fast can we get it online.
And we're committing to doing this together.
And the fact is, as OpenAI buys chips, that's great for AMD.
Our revenue goes up.
Our earnings go up.
We expect that it will also be very, very accretive
to our shareholders from day one.
And as we do that, we're very happy to have OpenAI as a deep partner.
And we win together.
So it's like a virtuous, positive cycle
in how we build out this big vision for having all this compute out there.
And yet, we still question,
as you were just talking about, Greg,
some of the other supply chain elements.
You're talking about the need for nuclear, for power.
What's really interesting is,
are you feeling confident enough
about the rest of the compute the supply chain is there?
Is this going to be US manufactured from your perspective?
Or are you looking at also building out internationally with AMD?
Yeah, we've been looking at really all options.
Our preference and really the core thing that we try to do
is build as much as possible in the US.
And you can see the commitments that we've made over the past year,
$500 billion of investment in the US.
And that's not stopping.
We're continuing to build.
I do think that international,
it is also going to be important for the world to have compute.
I think that compute is going to become this national security strategic resource.
And every country is going to need computational power.
And so we are really not limiting our sites in terms of where to build.
But we do think it is important that the US leads in this technology,
leads in computational power.
We're expanding the supply chain.
But you can see that we've really been working with partners across the globe
in order to actually meet the demand that we expect to be coming in upcoming years.
Lisa, the manufacturing of these chips,
will you look to Intel at all for it?
Do you think in the future?
Well, as you know,
the supply chain is something that we work on very, very meticulously.
I think we have a very strong supply chain.
We're certainly deeply partnered with TSMC across the supply chain.
Just to that earlier question,
we're absolutely prioritizing building in the United States,
because I think that's super important.
This is the US AI stack.
We want to have as much of it in the US as possible.
And we continue to really look at,
how do we ensure that there will be a strong supply chain going forward?
Greg, Sam posted on X that this deal with AMD
is incremental to what's already being done with NVIDIA.
But as Lisa knows, I've spent quite a lot of time
looking at the MI family and the newer generations of products to come.
Is there a very clear specific benefit to using AMD technology for inference
relative to the capabilities of NVIDIA?
Or do you just see it broadly as some sort of diversifying factor?
Well, I would look at it this way,
that there's a huge fixed cost to getting AI models running on any platform.
And so that when we look at what's out there,
that actually getting AI training to work is a huge amount of lift.
That's something we've really only done the work for NVIDIA.
But for inference, that's something that's much more,
that there's an easier barrier to entry there.
And one thing we found is that,
I think that the work that Lisa and team have been doing on the MI450 series,
it's looking like it's going to be a really incredible chip.
I think that the way that these things work
is there's niches for different balances of memory and computational power.
And so as we have a diversity of workloads,
we're finding that having a diversity of chips
also really accelerates what we're able to do.
Lisa, at the beginning of this conversation,
I said there were both operational and financial milestones to be met.
And Greg explained, you've got to start somewhere.
So in the first instance, one gigawatt.
But would you just sort of draw out the pathway to that first gigawatt?
It seems like you're prepared to move quickly here.
Yeah, absolutely.
And maybe, Ed, if I can just build on something that Greg said.
I think he's absolutely right.
We're a believer in there's a diversity of workloads,
and there will be a diversity of workloads across customers, models, use cases.
And from that standpoint, we feel really good about how we're positioned.
We love the work here because, frankly,
open AI is the ultimate power user of our chips and tests us in very good ways.
So I think that's what gives us confidence that the technology is there.
And then to your point about milestones, yes.
This is clearly a case where we are tied to each other.
The first gigawatt of deployment is super important.
We're going to start that second half of next year,
and we're going to build on from there.
And it really is not just the technology,
but commercial milestones, adoption milestones,
and just how we proliferate the capability going forward.
But I'm looking forward to building this as fast as possible.
We're already working with a number of cloud service providers
who are also very active on our technology.
And I think this is a great catalyst to get the industry to build faster.
Tied to each other is such an interesting turn of phrase.
And Greg, look, you are seeing more AI users and chip makers and designers
becoming more financially tied to each other.
Is this going to continue?
Is this the step forward for how you see this financing going forward?
Well, I really see the world transitioning to this AI-powered economy.
And the interesting thing is within OpenAI that we've really seen what it's like
when your progress is limited and accelerated,
as two sides of the coin, by computational power.
Like teams within OpenAI, that their ability to deliver
really is tied to the amount of compute that they get.
And I think we're heading to a world where that is how the whole economy will function.
And we're starting to see it, right?
That people having access to better AI tools.
If you're a coder, you're able to do far more
if you have access to better AI models.
And we're heading to a world where if you can have 10 times as much
AI power behind you, you will probably be 10 times more productive.
And so I think that we're moving to a world where the whole industry
is waking up to the fact that we have just not planned.
We have not planned for this moment where this explosion in AI demand is happening.
So it's happening all the way from the power to the silicon.
And I think this whole industry has to find a way
to actually rise to meet the occasion.
Lisa, you have given us a look into the future before
about how you see the total addressable market, the industry.
Now that the ink is dry with OpenAI and Greg,
are you rethinking either your bigger picture analysis
of the market for AI accelerators and GPUs?
Or do you see AMD now having an improved position in that market?
Relative, of course, to your friends at NVIDIA.
Well, again, I think, and I've told you before,
I believe that this is a huge market.
We have sized just the AI accelerator TAM
being over $500 billion in TAM over the next few years.
I think some might say maybe I was a little conservative in that TAM analysis.
But the way to think about it is there's so much need for compute.
I mean, you just heard it from Greg.
So this is a huge pie.
And you're going to see the need for more players coming into it.
And from my standpoint, this is a big validation of our technology and our capability.
As much as we love to work with OpenAI,
we're working with a lot of other customers as well.
There's a lot of excitement in the industry around MI450.
So we're ready for it.
