When people talk about NVIDIA today, the first association is often very simple:
AI = GPU = NVIDIA.
However, by August 2026, continuing to view NVDA simply as:
“A company that sells GPUs”
would significantly underestimate what the company is building.
NVIDIA’s strategic position has expanded far beyond individual AI chips.
The company is increasingly moving toward:
GPU + CPU + Networking + Storage + Software + Complete AI Factory Infrastructure.
In other words:
In the past, NVIDIA sold:
AI computing chips.
Today, it is increasingly aiming to provide:
Complete AI factories.
Therefore, the most important point when analyzing NVDA on August 25 was not only:
how many Blackwell systems the company could sell,
but also whether NVIDIA was successfully entering its next major platform cycle:
Vera Rubin.
First, Look at the Latest Available Earnings Report: $81.6 Billion
The FY2027 Q1 earnings report released on May 20 was extraordinary.
NVIDIA reported quarterly revenue of:
$81.615 billion.
Compared with:
$44.062 billion in the same period last year.
This represented:
85% year-over-year growth.
And:
20% quarter-over-quarter growth.
GAAP net income reached:
$58.321 billion.
GAAP diluted EPS was:
$2.39.
Non-GAAP diluted EPS was:
$1.87.
GAAP gross margin remained at:
74.9%.
The truly remarkable aspect of these figures was not simply that NVIDIA’s revenue had reached an enormous scale.
It was that:
The company continued maintaining nearly 85% year-over-year growth despite already operating at a quarterly revenue scale above $80 billion.
Many companies can grow from:
$1 billion to $2 billion.
But for a company already generating more than:
$80 billion in quarterly revenue,
maintaining this level of growth indicates something beyond a normal semiconductor upgrade cycle.
It suggests:
Global AI infrastructure construction remains in a large-scale expansion phase.
The Real Growth Engine Remains Data Center
In FY2027 Q1, NVIDIA Data Center revenue reached:
$75.2 billion.
Representing:
92% year-over-year growth.
And:
21% quarter-over-quarter growth.
Within this segment:
Data Center Compute revenue reached:
$60.4 billion
representing:
77% year-over-year growth.
Data Center Networking revenue reached:
$14.8 billion
representing:
199% year-over-year growth.
There is an important shift that is easy to overlook.
Historically, investors analyzing NVIDIA focused mainly on:
How many GPUs were sold.
However, networking is becoming an increasingly important part of the business.
Because large-scale AI data centers are not simply:
“Buy 10,000 GPUs and plug them in.”
A large AI cluster requires thousands or even hundreds of thousands of GPUs to exchange data efficiently.
This is where technologies such as:
NVLink, InfiniBand, Spectrum-X, and networking systems
become increasingly important.
Therefore, what NVIDIA is truly providing to customers is no longer only:
Computing power.
It is also:
The ability to connect and coordinate that computing infrastructure.
Why Is This Development So Important?
Because GPUs will eventually face competition.
AMD develops AI GPUs.
Google has TPU.
Amazon has Trainium.
Microsoft and other large cloud providers are also developing their own AI chips.
Large technology companies do not want to rely permanently on a single supplier.
If NVIDIA’s competitive advantage were only:
“Our GPU is faster than others.”
then competitors could gradually narrow the gap over time.
However, what NVIDIA is building today is much more complex.
The company is combining:
GPU
CPU
NVLink
Networking
DPU
Storage architecture
CUDA
AI inference software
into one integrated platform.
This changes the customer decision from:
“Which chip is cheaper?”
to:
“Which platform can bring an entire AI data center online faster while achieving the lowest cost per token?”
This is how NVIDIA’s competitive advantage is evolving.
After Blackwell, the Next Major Focus Is Already Vera Rubin
Over the past two years, the market has focused heavily on:
Blackwell.
However, by August 2026, NVIDIA had already begun shifting attention toward its next-generation platform:
Vera Rubin.
Vera Rubin is not simply a replacement GPU.
It represents an entirely new AI computing platform.
The platform includes:
Rubin GPU
Vera CPU
NVLink
BlueField-4
as well as supporting networking and AI infrastructure.
In the Q1 earnings report, NVIDIA officially introduced the Vera Rubin platform and positioned Vera CPU as a processor designed for:
Agentic AI workloads.
This shows that NVIDIA’s product evolution is increasingly becoming:
Each generation is not simply a chip upgrade, but a complete AI factory upgrade.
In July, Japan Provided a Specific Real-World Application Scenario for Vera Rubin
On July 16, NVIDIA announced a collaboration with Japan-based Noetra to build a:
Vera Rubin AI Factory.
The project includes:
13,750 Vera CPUs
and
27,500 Rubin GPUs.
The overall data center capacity is expected to reach:
140MW.
Supported by Japan’s Ministry of Economy, Trade and Industry, the project will serve industries including:
Manufacturing,
Logistics,
Healthcare,
Communications,
and Physical AI applications.
This announcement was highly important.
Because it showed that demand for Vera Rubin was no longer only:
“Customers may purchase this platform in the future.”
Large-scale infrastructure projects were already beginning to plan around the next-generation platform.
A Larger Partnership Emerged in Late July
On July 24, NVIDIA and SK Group announced an expanded collaboration.
The total planned AI partnership scale exceeded:
$500 billion.
Within the collaboration, SK Telecom plans to develop a NVIDIA Vera Rubin DSX AI Factory with capacity reaching:
Up to 2GW.
At the same time, NVIDIA and SK hynix established a long-term AI memory partnership involving next-generation:
HBM.
The most important point is not the headline figure of:
“$500 billion.”
Because the figure includes long-term plans, intentions, and multiple projects.
It does not mean that $500 billion will directly become NVIDIA revenue.
The more important development is:
AI infrastructure planning is beginning to move from megawatt scale toward gigawatt scale.
What does this mean?
Simply put:
Previously, customers were building:
An AI server room.
Now, they are increasingly building:
An AI industrial campus.
On July 27, Safe Superintelligence Also Selected Vera Rubin
Safe Superintelligence (SSI), founded by Ilya Sutskever, announced a long-term strategic partnership with NVIDIA on July 27.
SSI will gain access to:
NVIDIA Vera Rubin systems
to increase its computing capacity by:
An order of magnitude.
NVIDIA also made an investment in SSI.
The importance of this type of partnership is that NVIDIA is not only supporting today’s largest AI companies.
It is also positioning itself early with:
The next generation of AI laboratories that may grow into major platforms in the future.
Providing computing infrastructure to an AI startup today may create long-term ecosystem value if that company eventually becomes a large-scale AI platform.
This is why CUDA and NVIDIA’s hardware ecosystem create a form of:
Developer and customer path dependency.
On August 24, One Day Before the Entry Date, Another Highly Important Announcement Emerged
NVIDIA announced:
SpaceXAI would adopt NVIDIA Vera CPU.
SpaceXAI plans to use the Vera Rubin platform to expand Grok’s AI infrastructure and scale toward:
Gigawatt-level computing capacity.
More importantly, this collaboration is not limited to traditional ground-based data centers.
SpaceXAI plans to extend NVIDIA accelerated computing into space.
Its first-generation:
Starmind AI Satellite
is planned to use an optimized:
Vera Rubin NVL72 system.
Why was this announcement particularly important for August 25?
Because it was publicly released only one day before the analysis date.
Therefore, as of August 25, the market could already see:
Vera Rubin was not simply a distant next-generation product.
Customers had already begun planning infrastructure around it, including:
Japan’s national-scale AI infrastructure.
SK Group’s gigawatt-scale AI Factory.
SSI’s next-generation superintelligence training capabilities.
SpaceXAI’s Grok computing infrastructure.
And even AI computing applications in space.
NVIDIA’s Real Strength Is That Customers Are Already Spending on the “Next Generation” Before the Current Generation Is Fully Complete
The biggest challenge in the semiconductor industry is:
A product generation performs well today, but customers do not purchase the next generation.
Because technology evolves extremely quickly.
Blackwell may be highly competitive today.
But it does not automatically guarantee the same position several years later.
Therefore, NVIDIA’s most important capability is not simply:
“Blackwell is selling strongly.”
It is:
“While Blackwell is still being deployed at scale, customers are already planning data centers around Vera Rubin.”
This allows NVIDIA to create a powerful upgrade cycle:
Hopper
↓
Blackwell
↓
Blackwell Ultra
↓
Vera Rubin
↓
Future platforms
Customers continue expanding AI infrastructure with every generation.
If this cycle continues, NVIDIA would not simply benefit from a single AI investment wave.
It could create:
A recurring AI infrastructure upgrade cycle.
Another Often Underestimated Advantage Is Software
GPUs can be purchased.
Networking equipment can be purchased.
Servers can also be purchased.
But one of NVIDIA’s most difficult advantages to replicate is:
The CUDA ecosystem.
A large number of AI models, development tools, scientific computing applications, and enterprise software systems have already been built around CUDA.
As the industry moves toward Agentic AI, NVIDIA continues building additional software layers.
In Q1, NVIDIA introduced:
Dynamo 1.0
into production.
The company stated that it can improve generative AI and Agentic AI inference performance on Blackwell GPUs by up to:
7 times.
This explains why NVIDIA continues emphasizing:
Cost per Token.
Customers are not only asking:
“How much does this GPU cost?”
They are asking:
“How many tokens can I generate for every dollar spent?”
If a more expensive NVIDIA system ultimately delivers lower cost per token, customers may still choose that platform.
Therefore, the Shift Toward AI Inference May Not Be Negative for NVIDIA
The early stage of AI development was primarily focused on:
Training.
Training large AI models requires massive amounts of GPU computing power.
However, after models are trained, the activity that happens continuously every day is:
Inference.
A user asking ChatGPT a question.
AI generating code.
AI creating a video.
An enterprise AI Agent completing a task.
All of these activities require inference computing.
Training may happen periodically over months.
But inference can occur:
Billions or even hundreds of billions of times every day.
Therefore, as AI enters large-scale commercialization, computing demand may not decline.
Instead, it may expand from:
“Training models”
toward:
“Running models continuously in everyday applications.”
This is why NVIDIA continues emphasizing:
Agentic AI.
Another Important Number on August 25: $91 Billion
As of August 25, the market did not yet know NVIDIA’s final Q2 results.
The earnings report had not been officially released.
However, the company had already provided Q2 revenue guidance in May:
$91 billion, plus or minus 2%.
The company also expected:
GAAP gross margin of 74.9%
and:
Non-GAAP gross margin of 75.0%.
More importantly:
NVIDIA explicitly stated that this $91 billion guidance:
Did not assume any Data Center Compute revenue from China.
Therefore, from the perspective of August 25, the key market question was:
Even without including China data center computing revenue, could NVIDIA continue maintaining extremely strong growth?
This was one of the most important factors investors were waiting to evaluate.
The analysis should not use actual results released after August 26 to justify what was known on August 25.
China Exposure Was Also an Important Risk Factor at That Time
One of NVIDIA’s biggest risks was not that AI demand would disappear.
Instead:
Geopolitical factors and export restrictions.
Advanced AI chips have become closely connected with national security and technology competition.
If U.S. export restrictions on advanced AI chips become stricter, NVIDIA could lose access to certain market opportunities.
The fact that NVIDIA excluded China Data Center Compute revenue assumptions from its Q2 guidance showed that this issue had already become part of financial planning.
Therefore, analyzing NVDA requires separating two different questions:
Global AI demand remains strong.
and:
Whether NVIDIA can freely sell products into every market.
These are not the same issue.
The Second Risk Is Customers Developing Their Own Chips
Google has:
TPU.
Amazon has:
Trainium.
Microsoft and other major technology companies are also developing their own AI accelerators.
Why?
Because their annual spending on NVIDIA chips has become extremely significant.
If they can move some workloads to internally developed ASICs, they may reduce costs and decrease dependence on a single supplier.
Therefore, NVIDIA must continue proving:
Even when customers develop their own chips, NVIDIA remains the most valuable AI computing platform.
This is why the company increasingly emphasizes:
GPU + CPU + Networking + Software
rather than focusing only on:
GPU.
The Third Risk Comes From NVIDIA’s Own Success: Its Scale Is Already Extremely Large
Once quarterly revenue exceeds:
$80 billion,
maintaining 80% or 90% growth becomes increasingly difficult.
For a company generating:
$1 billion in quarterly revenue,
adding another $1 billion represents doubling the business.
But for NVIDIA, maintaining high growth requires adding:
Tens of billions of dollars in additional quarterly revenue.
Therefore, future market expectations for NVDA may become increasingly demanding.
The question is no longer:
“Is the company growing?”
Instead:
“Is growth continuing to exceed already extremely high expectations?”
This is why NVIDIA’s stock price may still experience volatility even after delivering very strong earnings.
Strong results may not always be enough.
The market may already be expecting:
Even stronger results.
Valuation Also Cannot Be Ignored
By 2026, NVIDIA was no longer a small company waiting for the market to discover its potential.
The market already recognized:
AI growth,
Blackwell demand,
Vera Rubin upgrades,
Agentic AI,
and sovereign AI opportunities.
Therefore, when analyzing NVDA on August 25, the question should not only be:
“Is NVIDIA the leader of AI?”
The more important question is:
“Can future growth continue exceeding the expectations already reflected in the current valuation?”
A great company does not automatically mean every price represents an attractive investment opportunity.
This point is especially important for NVIDIA, which has already become one of the world’s largest technology companies.
The Most Important Change on August 25 Was That NVIDIA’s Business Model Had Upgraded Again
Previously:
Selling GPUs.
Later:
Selling GPUs + CUDA.
Then:
Selling GPUs + CPUs + Networking.
Now:
Selling complete AI factories.
Vera Rubin is expanding this vision even further.
Japan’s project requires:
27,500 Rubin GPUs.
SK Telecom’s plan includes:
Up to 2GW AI Factory capacity.
SSI aims to increase computing capability by:
An order of magnitude.
SpaceXAI announced on August 24:
Using Vera Rubin to expand Grok infrastructure and extending NVIDIA computing capabilities toward Starmind AI satellites.
When these developments are viewed together, the trend becomes increasingly clear:
AI infrastructure is moving from “buying GPUs” toward “building AI factories.”
Therefore, focusing on NVDA on August 25 was not simply a bet on:
“Whether tomorrow’s earnings report would beat expectations.”
At that point, the more important question was:
Can the strong growth created by Blackwell transition smoothly into the next Vera Rubin cycle?
If:
Blackwell demand remains strong,
Vera Rubin enters large-scale deployment successfully,
Agentic AI and inference continue expanding computing demand,
and NVIDIA continues integrating GPUs, CPUs, networking, storage, and software into a complete platform,
then what NVIDIA is truly selling is no longer just a single chip.
It is:
The complete production infrastructure of the AI era.
Throughout history:
The most important companies of the Industrial Revolution sold machines.
The most important companies of the Internet era provided computing and networks.
NVIDIA’s ambition today is even more direct:
As companies, governments, and technology organizations around the world begin building AI factories, NVIDIA wants as many of the core hardware and software components inside those factories as possible to come from NVIDIA.
This is the core logic worth continuing to monitor for NVDA as of:
August 25, 2026.