What a year it’s been for Nvidia. Or rather, what another year, as once again the little chipmaker that could simply hasn’t stopped growing.

This year saw Jensen Huang and co become the first company in the world to achieve a market cap of above $5 trillion – that’s the equivalent of around two Canadas. Remember when the market was aghast that its market cap neared $1 trillion in 2023?

While the company’s “annual rhythm” of GPUs further elevates its already coveted high-end hardware to new heights in terms of demand, it’s hard not to think: where does this end?

The ever-looming spectre of the AI bubble grows larger by the day. But while AI has been the de facto driver of GPU demand, a segment of Nvidia’s business has slowly but surely become an integral part of its meteoric rise: networking.

Aside from the geopolitical underpinnings of its recent GTC event in Washington, D.C., the star of the show was in fact its networking components. From the apparent world-first way to connect a quantum system with classical GPU computers via NVQLink, to the long-awaited reveal of its BlueField-4 data processing unit (DPU), Nvidia came out guns-a-blazing with tantalizing interconnect offerings to further bolster its burgeoning stack.

Take its earnings report from August. Nvidia’s networking arm posted $46.7 billion in revenues, which was a 56% year-over-year increase and its highest-ever segment revenue. That rise demonstrates it’s not just shiny GPUs that are sought after, but solutions to bring those chips together.

Take its Ethernet-based AI fabric Spectrum-X. When it was showcased at Computex 2024, eyebrows were raised. Given Nvidia’s long-standing fondness for InfiniBand powering its interconnect systems, networking nerds everywhere were aghast that anything otherwise was on offer.

And yet, lo and behold, Nvidia’s latest earnings revealed that Spectrum-X delivered double-digit sequential and year-over-year growth, with annualized revenue exceeding $10 billion.

Throw in its traditional InfiniBand and NVLink offerings, and Nvidia’s array of networking offerings makes sense – it has a wide selection of protocols to fit the needs of its mammoth list of customers.

Nvidia’s networking expansion is akin to all those investors helping fuel its rise up the market cap ranks, diversifying its portfolio to borrow investment parlance in order to cast a wider net, instead of placing your bets in one pot.

We recently published a piece on how AI is making the network sexy again, and frankly, that concept is true.

Whereas the advent of 5G culminated in geopolitical strife, deployment malaise, and a genuine failure to build on initial enthusiasm, AI has done more for networks in two years than what next-gen telecom standards did in over half a decade.

You can have the shiniest, fastest GPUs in the world, but if they don’t all interconnect with one another, how do you expect to reach the level of inferencing we’re constantly told will usher in this next age of AI?

Of course, Nvidia is a part of the wider puzzle. It’s easy to get caught up in the glamor of GTC. Heck, the sheer importance of a technology like NVQLink alone is enough to make computing nerds like this writer giddy.

Having previously penned a gloomy view on the AI bubble and the potential for it to pop creeping ever closer, the evolution of networking technology will carry on. We’re in for more coherent optics and materials that’ll make speeds ever faster, using less energy. There’ll be more security considerations as SD-WAN technologies slowly become usurped by secure access service edge (SASE). And eventually, quantum computing and related networking will become the next big thing … just be a bit more patient.

From its re-engineered Ethernet products powering some of the world’s largest AI clusters, to breakthroughs that, in the company’s own words, have kicked off the “quantum-GPU computing era,” 2025 has been the year of networking for Nvidia.

Who’s to say it’ll slow down?