Networking Has Become a Larger Part of Nvidia’s Computer Than It Was at Sun

As major server-component suppliers report their financial results, the gap between demand and available supply is becoming clearer. Nvidia’s results for the second quarter of fiscal 2027 also provided indications about potential demand and supply levels in fiscal 2028.

When Intel reported its first-quarter results in April, it said the supply shortfall across its CPU business was measured in billions of dollars. That suggests the Xeon server CPU shortage was at least 20 percent and possibly closer to 25 percent. AMD may be facing an even larger imbalance, perhaps around 30 percent and increasing as agentic AI drives renewed demand for CPU capacity in AI datacenters. Several original equipment manufacturers have also indicated that demand is 25 percent to 30 percent above supply.

Nvidia already has much of its fiscal 2027 component supply allocated to OEMs and ODMs, and much of the expected fiscal 2028 supply may also be committed. During the company’s fiscal 2027 second-quarter call, chief financial officer Collette Kress said Nvidia had enough component supply to grow its datacenter business by slightly more than 70 percent in fiscal 2028.

Later in the call, chief executive officer and co-founder Jensen Huang discussed what Nvidia’s revenue might look like without supply constraints. He said Nvidia was on track to double revenue in fiscal 2027 and suggested that it might have been able to repeat that performance in fiscal 2028 if not for shortages of DRAM, HBM, flash, and other components.

If Nvidia could have doubled sales again in fiscal 2028, demand would have been just under 40 percent greater than available supply. Nvidia’s fiscal 2027 demand therefore appears broadly consistent with the shortages reported by Intel and server manufacturers, while AMD has described its situation less specifically.

“The unconstrained would be a lot higher,” Huang said when asked about the gap. “We will grow 100 percent year over year this year. The unconstrained, you know, is significant, and so we are just going to have to go work hard to get more capacity. And you know we have a large supply chain, we have a really gigantic supply chain, and so we have incredible partners, and we have secured a lot of supply. But we just need a lot more.”

Nvidia’s purchase commitments now total $279 billion, according to Kress’s commentary and the fiscal 2027 second-quarter results. That is 2.3 times the $119 billion secured through the end of the first quarter of fiscal 2027. The commitments cover the rest of fiscal 2027 through fiscal 2029, with a small amount extending into fiscal 2030 and later.

Those commitments could eventually grow to support approximately $1 trillion in sales from Grace-Hopper NVL72 and Vera-Rubin NVL72 systems across fiscal 2027 and fiscal 2028. A possible allocation would be $365 billion in fiscal 2027 and $625 billion in fiscal 2028. If supply constraints ease, or if Nvidia reduces the amount of DRAM and HBM used in its GPUs, CPUs, and DPUs, the resulting sales could be higher.

Nvidia’s fiscal second-quarter results

Nvidia nearly reached $100 billion in total quarterly sales in July. Revenue rose 105.9 percent year over year to $96.22 billion, largely because of the datacenter business. Operating income increased 124.1 percent to $63.73 billion, while net income rose 125.9 percent to $59.96 billion.

Net income represented 62 percent of revenue. That was slightly below the percentages recorded in the fourth quarter of fiscal 2026 and the first quarter of fiscal 2027, which had set consecutive company records. Even so, the figure is comparable to the gross margins many public companies target but rarely achieve.

Nvidia spent $7.05 billion on research and development during the quarter, more than twice its spending two years earlier. That investment is supporting current Grace-Blackwell sales and the initial Vera-Rubin shipments.

Compute and Networking dominates Nvidia’s business

Nvidia’s Compute & Networking group, which includes its datacenter products and some edge products, generated $88.3 billion in revenue, an increase of 113.6 percent from the same period a year earlier.

The Graphics group, which includes PC graphics cards and workstation GPUs, generated $7.92 billion, up 46.4 percent year over year. These businesses helped establish Nvidia’s datacenter position and remain important, but they now represent a much smaller share of the company’s financial results.

Nvidia has reorganized these activities. Automotive electronics, graphics products, and its developing workstation business are now grouped under a division called Edge Computing.

The Edge Computing division remains small compared with Datacenter. Nvidia has divided its datacenter business into hyperscalers, which include the major cloud infrastructure providers, and ACIE, a collective category covering AI Clouds, Industrial, and Enterprise.

The hyperscale part of the datacenter business generated $48.71 billion in fiscal 2027 second-quarter revenue, up 104 percent year over year. ACIE generated $40.31 billion, up 134.2 percent. Neoclouds, some sovereign operators, and many HPC and AI centers, including national laboratories that previously focused primarily on HPC, likely account for much of the ACIE revenue.

Networking’s growing role

Nvidia no longer reports compute and networking revenue separately in the same detail it once did. However, Kress said networking revenue was approximately $15 billion and compute revenue approximately $60 billion in the first quarter of fiscal 2027. Those figures, combined with sequential and annual growth rates, allow the second-quarter results to be estimated.

One estimate places datacenter compute revenue at $71.57 billion in the second quarter of fiscal 2027, up 111.5 percent year over year. Networking revenue is estimated at $17.45 billion, up 140.7 percent.

Within networking, Quantum-X InfiniBand revenue is estimated at slightly less than $7.5 billion, up 58.9 percent. That would leave slightly less than $10 billion for Ethernet and NVSwitch components. Ethernet may have contributed approximately $5.46 billion, with NVSwitch accounting for roughly $4.5 billion.

Networking represented an estimated 19.7 percent of datacenter sales in the first quarter of fiscal 2027 and 19.6 percent in the second quarter. That is about twice the networking proportion typically associated with cloud and hyperscale systems.

The difference comes from rack-scale systems. These systems use NUMA links between GPUs or other XPUs, so networking is not used only to connect separate server nodes. It also connects many processors into a single large virtual accelerator. In effect, the system attempts to make one giant virtual GPU from dozens of smaller physical GPUs, increasing the networking requirement within the rack.