Nvidia's Real Growth Story Isn't Chips—Its Networking Business Is

The Side Hustle No One’s Pricing In

Nvidia reported nearly $15 billion in networking revenue last quarter. Two years ago, that figure sat around $3 billion per quarter. Wall Street’s earnings models, however, are still built for a chip company.

The discrepancy matters more than most investors realize.

Compute still dwarfs networking inside Nvidia’s Data Center segment. But compute grew 77 percent year over year in that same quarter. Networking grew nearly 200 percent. When Nvidia reports earnings Wednesday, analysts expect networking to approach $17 billion with growth still running around 134 percent versus compute’s roughly 100 percent. The growth gap is narrowing, yes—but it hasn’t closed.

What’s striking is that most analyst coverage, price targets, and valuation frameworks treat networking as a footnote. Nvidia is rated as a GPU company, not as an infrastructure monopoly. If the second quarter confirms that networking can sustain growth at these levels, the market may need to fundamentally rethink how it values the business.

This isn’t about semantics. It’s about what happens when a company’s fastest-growing segment is also its most defensible one.

Why Custom Chips Make Nvidia More Valuable, Not Less

The uncomfortable truth for Nvidia is that every major cloud provider is building its own AI processors. Alphabet, Amazon, Microsoft, and Meta have all invested billions in custom silicon designed to run specific workloads outside Nvidia’s GPU stack.

Google’s Tensor Processing Units, Amazon’s Trainium and Inferentia chips, Microsoft’s Maia—all of these are direct responses to Nvidia’s pricing power and supply constraints. On paper, they represent a long-term erosion of Nvidia’s addressable market.

If you read this as existential threat, you’re missing half the story.

Those chips still have to talk to each other.

An AI data center isn’t a warehouse of individual processors. It’s a single machine made of thousands of accelerators, and the bottleneck isn’t computation anymore—it’s movement. Getting data between those chips fast enough to keep them from sitting idle is arguably the harder engineering problem.

This is exactly where Nvidia has positioned itself. Its Spectrum-X products connect servers over high-speed Ethernet. InfiniBand and NVLink move enormous volumes of data between processors. NVLink Fusion, the newer play, lets companies plug their own custom processors into Nvidia’s ecosystem of networking, power, and cooling.

You’re no longer buying a chip. You’re buying a rack that behaves like one computer.

The Plumbing Premium

This shifts Nvidia from component supplier to system architect. The economics of that are meaningfully different.

When you sell GPUs, you compete on price, performance-per-watt, and availability. When you sell the infrastructure that connects everything, you gain something rarer: lock-in. A hyperscaler that has designed its custom chips around NVLink Fusion faces enormous switching costs if it ever wants to leave Nvidia’s stack. The custom chip race was supposed to reduce Nvidia’s dependency on hyperscaler spending. In practice, it may be deepening it.

Jessica Inskip at StockBrokers.com described this strategy in blunt terms on Yahoo Finance’s Opening Bid: Nvidia is “investing in the bottlenecks that are the AI infrastructure.”

Networking is one of those bottlenecks. Memory and storage are others. The point is that as AI gets bigger, the plumbing gets more expensive—and Nvidia owns more of it.

There’s also a psychological dimension that most coverage overlooks. Custom chip projects take years to mature. Even if a hyperscaler eventually achieves full self-sufficiency—which none have yet—they would need to rebuild their entire data center networking layer from scratch. That’s not a decision made lightly, and it’s not a timeline measured in quarters.

The Stock Market Already Knows This. Do You?

Nvidia’s stock has decoupled from the broader semiconductor trade for roughly two months. Since the iShares Semiconductor ETF (SOXX) peaked around June 22, it has fallen about 20 percent. Nvidia has gone essentially nowhere.

That’s not strength in a rising market. That’s resilience in a bear market. The chip sector is re-pricing for lower margins, cyclical weakness, and the reality that custom silicon is eating into addressable demand. Nvidia is being priced differently because it’s no longer just a chip company on anyone’s balance sheet—though the earnings models haven’t caught up.

The divergence between SOXX and NVDA is worth watching closely. When the semiconductor index corrects, Nvidia doesn’t move in sympathy. That tells you the market is already assigning a different risk profile, even if the financial models still label it as a pure-play semiconductor name.

What Wednesday’s Report Could Reveal

The networking story isn’t abstract. It’s quantifiable, and it’s about to become more so.

If Nvidia’s networking segment hits $17 billion as expected, that’s growth running almost 40 percentage points ahead of compute. The absolute dollar contribution of networking is approaching levels that would make it a standalone business larger than most semiconductor pure-plays. Combine that with the strategic shift toward infrastructure lock-in, and you get a company whose moat is widening even as the chip race intensifies.

The risk is real but different. If hyperscalers collectively decide to standardize on a third-party networking solution instead of NVLink Fusion, Nvidia loses its plumbing advantage. But there’s no sign of that happening. The trend is the opposite: every new custom chip project seems to deepen Nvidia’s integration.

Look at the capital expenditure commitments from the major cloud providers. Billions are being deployed into data centers that require massive networking investment. Every megawatt of AI compute deployed today is pulling in proportional networking demand. The ratio isn’t one-to-one—recent deployments suggest networking spend is actually outpacing compute spend at the margin.

The Valuation Question

Nvidia trades at a premium to the semiconductor index. Most of that premium is justified by compute dominance. But if networking continues compounding at these rates, the premium should broaden—or at least hold.

Investors who model Nvidia only on GPU shipments are leaving money on the table. The faster-growing, stickier, structurally advantageous part of the business is the one no one’s writing about. Wednesday’s earnings should make that impossible to ignore.

The question for investors isn’t whether Nvidia’s networking business is impressive. It’s whether the market will re-rate the company before or after the numbers confirm what the trajectory already suggests—that the plumbing business may soon rival the chip business in both absolute size and strategic importance.