On Tuesday, 17 March 2026, NVIDIA (NASDAQ:NVDA) presented at the
Conference GTC Financial Analyst Q&A, highlighting its strategic
advancements in AI and financial outlook. The company showcased strong
growth prospects, driven by innovative AI systems and robust demand for
its architectures. However, challenges in maintaining margins and
competition were also discussed.
Key Takeaways
- NVIDIA anticipates over $1 trillion in demand for its Blackwell and Rubin architectures by the end of 2027.
- The company plans to allocate 50% of its free cash flow to stock repurchases and dividends.
- NVIDIA is expanding its AI solutions, integrating CPUs and storage to enhance its AI factory platform.
- The
shift towards token-based models is reshaping the IT industry, with
NVIDIA positioning itself as a leader in this transition.
- NVIDIA’s customer base is diversifying beyond hyperscalers to include regional clouds and enterprises.
Financial Results
- NVIDIA projects strong visibility in demand, with over $1 trillion expected for Blackwell and Rubin architectures through 2027.
- Excluding other products like Feynman and Groq, the company sees potential for additional business growth.
- Approximately
50% of free cash flow will be directed towards stock repurchases and
dividends, with potential increases based on future growth.
Operational Updates
- The rise of agentic systems marks a new inflection point in AI, enhancing NVIDIA’s growth trajectory.
- NVIDIA’s AI coverage has expanded, adding platforms like Anthropic and Meta AI.
- The company is focusing on optimizing AI inference with architectures like Groq, catering to diverse performance needs.
Future Outlook
- NVIDIA plans to ship Vera Rubin before Groq, enhancing throughput with NVLink 72 and Vera Rubin Ultra NVLink 144.
- The company aims to increase throughput and pricing, potentially boosting demand to $1.25 trillion if Groq is fully integrated.
- Jensen envisions 99% of global compute focusing on inference, positioning NVIDIA at the forefront of AI innovation.
Q&A Highlights
- Non-public AI companies are reportedly seeing revenue increases of $1 to $2 billion per week.
- The IT industry, valued at $2 trillion, is expected to grow to $8 trillion, integrating OpenAI and Anthropic models.
- Jensen emphasized NVIDIA’s value proposition, highlighting the importance of integrated hardware and software solutions.
For a detailed understanding, readers are encouraged to refer to the full transcript below.
Full transcript - Conference GTC Financial Analyst Q&A:
: Good morning, everyone. As we quiet the back room, I
have a very important job. As a reminder, the content of this
presentation may contain forward-looking statements, and investors are
advised to read our reports filed with the SEC for information related
to risks and uncertainties facing our business. With that, I will turn
it over to Jensen and Colette.
Colette: All right. Good morning, everybody. I hope you
enjoyed the presentation yesterday. Went a little bit longer, but I
think it was an absolutely great summary for us. We’re gonna take this
time to focus on your needs and some of the additional kind of questions
you are. We’re gonna start with a couple, maybe the first slide or so,
and then we’ll open it up for questions, and I’m gonna turn this over to
Jensen with that.
Jensen: As I was saying
yesterday, there were three inflection points in recent AI. The first
one was generative AI, the second was reasoning, and we’re at the third
inflection point now, and each one builds on the others. There’s a lot
of technical reasons why each one of them built on the others, but here
we are with the third inflection point, which is agentic systems.
Agentic systems that are able to operate autonomously, that’s why they
call them agentic, because they have agency. You can give them goals,
and instead of just answering questions, they can now perform tasks.
Tasks could be anything. Of course, one of the most popular applications
of agentic systems is write software.
You know, engineers in your company, I’m sure, and engineers in my
company for sure are using agentic systems all day long. What used to be
a thing for engineers is, you know, when you come to work, they give
you a laptop. Now when you come to work, they give you a laptop and
tokens. Token budget is now a real thing. Every engineer is gonna have a
token budget. You know, the idea that you would hire a $300,000
engineer and they spend no tokens in doing their job, you gotta ask the
question, what are they doing?
It is very, very clear now that every engineer will have a lot of
tokens that they will have to consume, and those tokens are going to be
produced. Now, I just said something a second ago. If you just connected
the dots, we used to be, when an engineer comes to work, software
programmer, somebody comes to work, give them a laptop. That’s a tool.
Today, we give them a laptop and tokens. Those tokens have to be
manufactured. A computer used to be just a tool. A computer of the
future is a manufacturing equipment. These computers, as you see,
they’re no different than ASML manufacturing equipment of the future.
They’re producing something that is sold. Just, it’s no different than a
dynamo machine a long time ago that produced electricity.
These
are manufacturing systems, and the energy efficiency of it, the
production efficiency of it matters everything because it drives your
revenues, okay? The third inflection point is here. As you know,
OpenClaw. Oh, many of these things when they first drop, these open
source projects, they seem like toys. You take a step back and just
analyze what is OpenClaw on first principles, and I explained that
yesterday. OpenClaw on first principles is really a computer, the
operating system of an AI computer, a personal AI computer. It has all
of the properties of a computing system. It has all the properties of an
operating system of this new computer. You know, it manages resources,
it schedules, it does IO, and, you know, it networks.
All of the properties of a fundamental computer it has, okay? You
could see the red line is actually not the y-axis. The red line is its
growth. That’s just the extraordinary thing. Every company in the world
will now need to have it. What is your OpenClaw strategy? Every single
software company, every single company needs to have an OpenClaw
strategy. Just as we all had our Linux strategy, just as we all had to
have an internet strategy, just what is your mobile cloud strategy? Now,
the question is what’s your OpenClaw strategy? Okay? This is a very big
deal. I wanted to answer the questions about what I said here a little
bit more.
First of all, a year ago, I said that we had strong visibility of our
Blackwell and Rubin shipments of $500 billion through 2026. I was
standing in 2025, right? GTC 2025 was around, what was it? Was it March?
April?
Colette: October.
Jensen: It was October?
Colette: October.
Jensen: Okay. October. I was standing there. You sure? It was October?
Colette: I’m sure.
Jensen: GTC DC or GTC. I said it twice, though. The first time I said it was GTC here, right?
Colette: I think you’ve been saying it twice.
Jensen: Yeah.
Colette: I don’t think all the way back.
Jensen: I see. Yeah. Okay. Anyways. Anyhow, in 2025,
one of those months, I said that we have strong visibility of Blackwell
plus Rubin demand, purchase orders and demand, okay? Very firm demand of
$500 billion. There were a lot of questions from many of you that, you
know, so where are we now? You wanted an update on where we are now, and
I thought I’d give you guys an update. Where we’re standing right now
and what month are we? Just for the record. March. Here we are in March.
The end of 2027, as you know, is many more months away. I just wanna
first let you guys know that.
However, because we’re building infrastructure and factories and the
lead times for everyone is long, they wanna make sure they give us firm
demand or give us purchase orders and firm demand as early as they can
to secure their supply. Okay? We have strong confidence and visibility
of $1 trillion plus. You know, it’s not a floating point number, you
guys. Okay? It is also not 94 digits of accuracy, okay? We’re not
counting cents. You can keep your cents. However, we have strong
visibility of $1 trillion plus of Blackwell plus Rubin. The reason why
it’s only Blackwell plus Rubin and not all of the other things that we
sell is because I referenced it from the last year when I was only
talking about Blackwell and Rubin. Does that make sense?
Last year
we didn’t have Groq. Last year, we weren’t selling standalone CPUs.
Last year, we didn’t have many of the things that we have to sell now,
and it wouldn’t have made sense for me to include those today and not
because we didn’t have those things yesterday. Does that make sense?
Somebody nod, then I can continue. Okay? A couple of things. It’s only
Blackwell and Rubin. It’s not Feynman. It’s not Rubin you know, Rubin
Ultra. It’s not Vera standalone. It’s not Groq. Blackwell plus Rubin, we
have high confidence, strong visibility, demand, forecast, purchase
orders of $1 trillion+. We close businesses that we ship oftentimes, and
we expect to close and ship more business between now and the end of
2027.
We expect to close, book, and ship more business on top of this
between now and 2027. The reason for that is because we expect to be
coming to work between now and the end of 2027. Now, unlike other
businesses, because we build and complete systems of this quality, we
can actually win, book, ship new business in the same quarter. Of
course, you can’t do that if you have to build an ASIC or you know,
obviously. If you don’t see it now, you’re not shipping it by the end of
2027. That’s not true for us. We build inventory. We have a pipeline of
supply that.
We have to take care of customers who come out of the blue because
they’re desperate for more compute. Does that make sense? When they’re
desperate for more compute, and all of a sudden in the last day, they
say, "You know, goodness gracious, I could use more," I would like to be
able to say, and we are always in a position to say, "We’d be more than
happy to help you." We’re also working on new customers, new markets,
new regions that we haven’t put in here yet because we still have, well,
about 21 months to go. Okay? I want you guys to understand what that $1
trillion is. It’s, by definition, going to keep growing. By definition,
because what I compared it against, it will keep growing, and it’ll be
larger than that.
A couple of things that I wanted to say, also,
that last year was a really good year because 2025 was our year of
inference, and I think we helped everybody understand that the price of
the computer and the cost of the token, the price of the computer and
the cost of the token are only marginally related. The price of the
computer and the cost of the token. Remember, people are buying these
computers to produce tokens. The effectiveness of the production of
those tokens matter greatly. They’re not reselling the computer. If you
bought a computer and it’s expensive, if you resold it, and that’s it,
then it’s expensive. But you bought a computer, and it’s expensive
because the technology’s incredible, but it produces tokens at such
incredible rates.
You had simultaneously have purchased the most expensive computer and
produced the lowest cost tokens. Does that make sense? This is what we
do every day. This is our job. It is the reason why we deliver the value
that we deliver. The value discrepancy that we deliver here, the two
numbers that I just described, is how we’re able to secure our gross
margins. We have to deliver, and we consistently deliver so much more
value, which is tokens per second, which is tokens per second per watt.
We deliver so much more value every single generation that customers
would prefer to buy our next generation product at a higher price than
our current generation product at a lower price. They prefer
instantaneously to convert. The moment that Vera Rubin comes, it is
smarter to install Vera Rubins than to continue to buy Grace Blackwells.
Are
you guys following me? Somebody nod. Okay. The value is better even
though the price is higher. I’m comparing these two systems because
these are the two de facto systems in the world. Until you can beat
these two systems, there’s no point buying something else. These two
systems are incredibly hard to beat because Moore’s Law doesn’t give you
35x. Moore’s Law alone won’t do it. Building a faster chip won’t do it.
You’re gonna have to build a faster lots of chips. 2025 was year in
inference, and I think we demonstrated our inference leadership.
Training to post-training to now inference. Then some of the other
things that we did last year that was really great is we expanded the
reach.
We expanded the number of AIs that now support our platform. Last
year, 2025, we added Anthropic to our platform, which is net new. We
added Meta AI, which is net new. We’re still working with Meta on all of
the other stuff. Meta AI is a net new entity, and they have net new
computing requirements. We can all acknowledge that last year, open
source software, open source models really took off to the point where
API inference service providers now see that open models probably
represent, approximately represent the second most popular AI model.
Meaning the first one, of course, is OpenAI into total number of tokens
generated. In aggregate, open models represent number two. As you know,
NVIDIA is the best platform for open models in the world. We are the
standard for open models everywhere. Number one, OpenAI.
Number
two, all the open models. Number three, Anthropic. Number four, xAI.
Just take your list, keep working. I think NVIDIA’s coverage of models
last year increased substantially, which explains our accelerating
growth at a very large number. We are already a very large company, as
you know, and we’re now accelerating. Our rate of growth is actually
accelerating. Anyways, that’s I think about it. Oh, one last point. We
love our hyperscaler partners, and we work very, very closely with them,
but it’s important to understand that our relationship with
hyperscalers is we’re not just selling to them. We attract customers for
them. Having CUDA in their cloud brings all of the CUDA developers, all
the AI natives, all the large companies that we work with.
Whenever we accelerate those large or small companies, we bring them,
we train them, we have them hosted in the world’s CSPs. We are one of
the best sales forces of the world’s CSPs. It is the reason why if you
go down to the show floor, they have all of the largest booths. AWS has
the largest booth here. Google Cloud has the largest booth here. Azure
has the largest booth here. Oracle, giant booth here. CoreWeave, big
booth here. Does that make sense? Because we bring customers to them.
Why are they here? To sell to my developers. All of our developers only
know how to program one thing. They only know how to program CUDA, and
they only use CUDA-X libraries.
when we win and when we help those
developers integrate NVIDIA, they land on one of our CSP partners. We
are one of the CSP’s best sales forces. All right. However, we’re also
seeing tremendous customer diversity outside of the CSPs. Regional
clouds, industrial, enterprise on-prem, when Dell and Lenovo and HP,
they’re all growing so fast and all the ODMs are growing so fast. A lot
of that business go towards the right-hand side of that chart, the 40%.
Most people see our business in the left 60%. The right 40% without
NVIDIA’s full stack, the fact that we can build you the entire AI
factory and the fact that all of the world’s open platforms run on top
of NVIDIA, you have no hope addressing the 40%.
The net of this chart is this, a big part of that 60% is NVIDIA
developers landing in the cloud. 100% of the 40% is impossible without
full stack, without end-to-end. Was I successful in communicating that?
It’s important to understand our business. We aggregate that whole thing
into what is called accelerated computing, and it’s probably a
disservice to you. Next year, we’re gonna separate it out a little
differently. Well, in the future, we’re gonna separate it out a little
differently, and it’s gonna look probably like this chart. You’ll see
something like hyperscalers or something like that in 60% of it. Even
when you see that, remember, a lot of those customers, we brought to the
clouds.
On the right-hand side, that 40% is completely impossible
if you just build a chip because they don’t buy chips. They buy
platforms. Three messages all in one slide, which probably made your
brain blew up, and therefore, I did it again. Was that helpful? You know
what I should have done? I should have made three panels or three
slides. It would’ve been a seven-hour keynote. But it would’ve been
worth it. Okay. That’s it. Thank you. Questions?
: We’re opening it up for questions now.
Ben Reitzes, Melius Research: Hi, it’s Ben Reitzes,
Melius Research. Thanks for having us here at this event. It’s amazing
access that you guys provide. Congrats to you and the team for that.
This is great. Jensen, last night, when we took a picture. By the way,
you all can still like that picture. I need to beat last year’s record.
Jensen: What, what picture?
Ben Reitzes, Melius Research: Well, we took a quick picture, and I posted it.
Jensen: Uh-huh.
Ben Reitzes, Melius Research: I’m trying to beat last year’s likes.
Jensen: Oh, okay. All right.
Ben Reitzes, Melius Research: Yeah. Anyway.
Jensen: Was I in some vulnerable position or anything?
Ben Reitzes, Melius Research: Let’s put it this way. The camera added 10 pounds to me, but not to you. I don’t know how that works.
Jensen: Thank you.
Ben Reitzes, Melius Research:
You look great. So, I promised I’d ask you an inference question, and
this is related, is, you know. This is great. Like, I think a lot of
people here get this. I think the main pushback we get is the juice
worth the squeeze, and will the hyperscalers have upside to their
revenues for API and cloud that justify all the spend, and what is
Jensen seeing? Because, you know, I have estimates for the hyperscalers,
and I’ve said there’s upside to the revenues. For now, the CapEx is 20%
above their cloud/API revenue, and I’m wondering what you’re seeing.
You’ve said in the past that there’s this massive upside to these cash
flows and from your customers, particularly hyperscalers and those that
are serving Anthropic and OpenAI. When do we adjust those higher?
I know this is a tough question for you because you got to guide for
five other companies. If we see that upside, I think your stock will
behave a lot better because then we’ll realize this build can keep
going. When is this inflection? I mean, we’re seeing the inflection, but
when is it, you know, what is the upside to their revenues, and how do
we feel better about it?
Jensen: I wish those companies were public, and the
reason for that is because then you’ll see what I see. No companies in
history has ever grown as a startup company, non-public company. As a
startup company, increased revenues by $1 billion or $2 billion a week.
That’s what they’re experiencing right now. Now, remember, I just said a
week. The entire IT software industry is, call it $2 trillion. That $2
trillion industry, I don’t believe is gonna be disrupted. I think it’s
going to be transformed.
I believe that $2 trillion IT
industry is going to integrate a combination of OpenAI, Anthropic, and
open models and connect it with an open source software called OpenClaw
that we turned into an enterprise-ready version called NemoClaw, and you
have instantly an agent. 1.5 million people downloaded OpenClaw and
built themselves an agent. It’s one line of code. Then you tell the
agent to finish building itself. Oh, you don’t know this thing? Go learn
it. It goes off and learn it, you know. In the future, those agents
will be integrated into the IT industry. This IT industry is $2 trillion
of software licenses today.
It’s probably going to be, let me just pick a random number, $8
trillion that also resells an enormous amount of tokens. 100% of the
world’s IT industry will become resellers of OpenAI and Anthropic. Are
you guys following me? No?
Ben Reitzes, Melius Research: Take your estimates up for OpenAI and Anthropic.
Jensen: I believe that Anthropic and OpenAI and of
course all of the IT company will also modify and customize their own
software, their own models with open models, and that’s what Nemotron’s
for, and that’s what Nemo’s for and we’ve created all the tools, and
that’s why we’re working with all of them. They’re all going to create
agents that integrate these three components. I believe they’re gonna
grow incredibly. The time is gonna come, it’s gonna come soon. The
reason for that is you could see it in Anthropic’s numbers, you could
see it in OpenAI’s numbers. They’re growing an entire IT company in a
month. The revenues of these AI companies, their AI will be used by
enterprise directly, but it’s also going to be resold through IT
companies, integrated into IT companies. Does that make sense?
Ben Reitzes, Melius Research: Yep.
Jensen: Because just think of it, that AI is just
software. Their software is gonna be offered directly to enterprises,
but it’s also going to be integrated and become domain-specific and
specialized, governed, secured, easily provisioned, connected to their
system of records, so on and so forth. There’s gonna be a whole, you
know, and that agentic system will be rented to customers, but they
still would have to consume tokens through factories. If it comes down
through OpenAI, that’s terrific. It comes down through Anthropic, that’s
terrific. If it comes down through open models, that’s terrific. They
all have to have tokens generated. The net-net is IT companies of the
past license software. IT companies of the future will rent tokens, will
generate tokens. Are you guys following me? Their business models will
change. The companies will become bigger. Their gross margins will
change.
Gross margin profile will change because they have COGS in their
business model now. They offer greater, much, much more value. This is
exciting for them. Super exciting for them.
Ben Reitzes, Melius Research: Okay, great. Passing this $8 trillion microphone.
C.J. Muse, Cantor Fitzgerald: Thank you. Good morning. C.J. Muse, Cantor Fitzgerald. Thank you for hosting this event. Really appreciate it.
Jensen: Yes, C.J.
C.J. Muse, Cantor Fitzgerald: Wanted to, I guess, maybe
follow up on Ben’s question, and think about the evolution of this
chart of 60/40. You know, you talked about NemoClaw, and then you
announced yesterday the Vera Rubin DGX AI factory reference design,
essentially, providing the blueprint for your non-hyperscale customers
to compete with the hyperscalers. I’m curious, you know, as you put it
all together, you see, you know, massive spike in token generation, how
you’re expecting, you know, pretty much this chart to evolve over time,
and how we should be thinking about the different players inside there,
as to their relative kinda growth vectors....
