Nvidia’s earnings may reveal that the company’s fastest-growing AI business is the infrastructure connecting its chips—not the chips themselves.

Wall Street has spent years obsessing over one number whenever Nvidia reports earnings: GPU sales.

That makes perfect sense. Graphics processors transformed Nvidia from a specialist semiconductor company into one of the most important businesses in the global artificial-intelligence economy. The company's accelerators became the engines behind enormous AI training clusters, and demand for them helped propel Nvidia's revenue and market value into territory few technology companies have ever reached.

But ahead of Nvidia's latest earnings report, another number is beginning to demand attention.

Networking.

Nvidia's networking business generated nearly $15 billion in the most recent quarter, up from roughly $3 billion per quarter two years earlier. Wall Street expects that figure to approach $17 billion in the latest quarter, according to Yahoo Finance's analysis.

The numbers matter because networking is growing considerably faster than Nvidia's already enormous computing business.

Last quarter, networking revenue was up nearly 200% year over year, compared with roughly 77% growth for compute. Analysts expect networking growth to moderate to around 134% in the quarter being reported, but that would still exceed expected compute growth of roughly 100%.

That changes the Nvidia story.

The company may no longer be simply the world's dominant seller of AI processors.

It may be becoming the company that sells the infrastructure required to make thousands of those processors function as a single, enormous machine.

The AI boom created a deceptively simple formula.

Buy more GPUs.

Build more data centers.

Train bigger models.

But once thousands of processors are packed into a single facility, another problem emerges.

The chips need to communicate.

Constantly.

An AI system can contain enormous numbers of accelerators operating in parallel. Each processor has to exchange data with other processors, storage systems and networking equipment. If communication between those components is too slow, the GPUs can sit idle.

That is an extremely expensive problem.

A company can spend millions of dollars on advanced processors only to discover that the surrounding infrastructure prevents those chips from being fully utilized.

Nvidia has increasingly positioned itself directly in the middle of that problem.

Its Spectrum-X products support high-speed Ethernet connections for AI workloads. Its InfiniBand technology is designed for extremely fast communication between systems, while NVLink provides high-bandwidth connections between processors.

The strategic idea is straightforward.

Nvidia does not just want to sell the engine.

It wants to sell the transmission system, the communications network and much of the machinery surrounding the engine.

One of the most interesting aspects of the networking opportunity is that Nvidia does not necessarily need every AI accelerator to carry its logo.

Alphabet, Amazon, Microsoft, Meta and other hyperscalers are investing heavily in custom processors designed for particular AI workloads.

At first glance, that appears to threaten Nvidia.

If large technology companies increasingly build their own chips, fewer workloads may require Nvidia GPUs.

But there is a twist.

Custom chips still need to communicate with each other.

They still need high-speed networking.

They still need systems designed to move massive amounts of information around a data center.

Nvidia's newest networking strategy increasingly acknowledges that reality.

The company has introduced NVLink Fusion, a platform that allows customers to integrate custom processors into a broader Nvidia-connected architecture involving high-speed communications, networking, power and cooling.

That means Nvidia can potentially earn revenue from AI systems even when the processor at the center of the workload is not exclusively an Nvidia GPU.

It is a subtle but potentially powerful change.

Nvidia's broader ambition is becoming easier to see.

The company increasingly wants to provide customers with complete AI infrastructure rather than individual components.

That includes computing.

Networking.

Interconnects.

Software.

Systems design.

And increasingly, the physical architecture around AI clusters.

Jessica Inskip of StockBrokers.com described Nvidia's strategy as investing in the bottlenecks created by AI infrastructure demand. She said Nvidia is effectively trying to become a “one-stop shop” for AI.

That approach could make the company more resilient to competition.

Suppose custom accelerators gain market share.

Suppose a hyperscaler develops a highly competitive chip.

Nvidia could still potentially supply the networking layer connecting that chip to the rest of the AI system.

That is considerably different from a company whose fate depends exclusively on maintaining GPU market share.

The size of Nvidia's networking business also matters.

Even with the extraordinary growth rate, compute remains much larger.

Networking is expected to represent roughly one-fifth of Nvidia's Data Center revenue after the latest quarter, according to the Yahoo Finance analysis.

That means Nvidia's core business is still the engine.

But networking is becoming an increasingly important second engine.

And second engines can dramatically change a company's long-term growth profile.

Investors have traditionally valued Nvidia according to the growth potential of accelerated computing.

Now they have another variable.

What happens if every new generation of AI data centers requires not just more GPUs, but significantly more networking equipment per cluster?

The answer could be that Nvidia captures a larger percentage of every dollar spent on AI infrastructure.

There is an important structural reason for this.

AI models have become dramatically larger and more computationally demanding.

That pushes data centers toward architectures involving more accelerators working together.

As cluster sizes increase, communication becomes more important.

A system containing a few processors can tolerate modest delays.

A giant AI cluster containing thousands of accelerators cannot.

At enormous scale, tiny inefficiencies can translate into substantial wasted computing capacity.

And because high-end AI accelerators are extremely expensive, customers have a strong incentive to make sure they remain fully utilized.

That creates demand for advanced interconnect technology.

In other words, the more ambitious AI becomes, the more important networking becomes.

This is one reason Nvidia's strategy could continue working even if the AI hardware ecosystem becomes more diverse.

Nvidia's networking growth may also help explain why its stock has behaved differently from much of the semiconductor sector.

According to Yahoo Finance, Nvidia has traded roughly sideways over the past two months while the iShares Semiconductor ETF fell around 20% after the broader semiconductor trade peaked on June 22.

That divergence is notable.

Nvidia has historically been treated as the leading semiconductor stock.

But investors increasingly appear to view the company as something broader.

It is becoming an AI infrastructure company.

That distinction could become more important as the industry matures.

Nvidia's latest earnings therefore have implications beyond whether revenue and profit beat expectations.

Investors will be listening closely for commentary about networking demand.

They will want to know whether the rapid growth rate is sustainable.

They will want details on hyperscaler spending.

They will also be watching management's comments about custom silicon and the role of NVLink Fusion.

Most importantly, investors will want to understand how much of the AI infrastructure dollar Nvidia can capture.

If networking approaches $17 billion and continues growing rapidly, the narrative around Nvidia could begin shifting.

The question would no longer be “How many GPUs can Nvidia sell?”

It would become:

How much of the AI data center can Nvidia own?

This could be Nvidia's most interesting strategic advantage.

The rise of custom AI chips is often described as a direct threat to Nvidia.

But Nvidia's networking strategy potentially turns part of that threat into an opportunity.

The company can allow customers to develop specialized chips while still remaining central to the architecture connecting those chips.

That creates a more flexible business model.

Nvidia does not necessarily have to win every semiconductor battle.

It needs to remain essential to the system.

That is a much broader competitive objective.

And it may prove more durable.

The early AI boom was about discovering what advanced models could do.

Then the market became obsessed with GPUs.

Now the industry is entering another phase.

The challenge is building systems large enough, fast enough and efficient enough to support the next generation of AI.

That means power.

Cooling.

Memory.

Networking.

Storage.

Software.

And enormous quantities of computing.

Nvidia has increasingly positioned itself across that entire stack.

The networking numbers suggest investors may have underestimated how quickly that strategy is working.

Nvidia's GPU business is not going away.

It remains the company's financial powerhouse and the center of its competitive advantage.

But the latest numbers suggest the most important story inside Nvidia's earnings report could be hiding beside it.

Networking is growing faster.

The business has multiplied several times over the past two years.

And Nvidia is using networking to make itself harder to dislodge from the center of AI infrastructure—even as technology giants develop alternatives to its processors.

That could be one of the company's smartest strategic moves yet.

Because in the next stage of the AI race, the winner may not simply be the company that makes the fastest chip.

It may be the company that makes thousands of chips work together most efficiently.

And Nvidia increasingly wants to be that company.

Source basis: Yahoo Finance's August 26, 2026 analysis of Nvidia's networking growth and broader AI infrastructure strategy.