Nvidia's purchase commitments to suppliers have reached $279 billion, up 2.3 times from the $119 billion secured through the end of Q1 fiscal 2027. That single figure, disclosed alongside second-quarter results, says more about the state of the AI supply chain than any analyst model currently on the street.

The company posted $96.22 billion in revenue for the July quarter, up 105.9 percent year on year. Operating income rose 124.1 percent to $63.73 billion. Net income reached $59.96 billion, up 125.9 percent — 62 percent of revenue, a ratio most public companies aim for in gross margin and rarely reach in profit.

The constraints are real and they are shared across the industry. Back in April, Intel said the supply shortfall for its overall CPU business was a number that started with a "B," implying a Xeon server CPU shortage of at least 20 percent, more likely 25. AMD's Epyc over-demand may run to 30 percent, given its price and performance position and a CPU resurgence in AI datacenters as agentic AI takes hold. Various OEMs have hinted their demand exceeds supply by 25 to 30 percent in recent quarters.

Nvidia sits at the center of that squeeze. Fiscal 2027 components were long ago allocated to OEMs and ODMs, and the majority of fiscal 2028 supply is likely spoken for as well.

What did management actually say?

Chief financial officer Collette Kress did not mince words on the call: Nvidia has enough component supply to grow the datacenter business by slightly more than 70 percent in fiscal 2028.

Co-founder and chief executive officer Jensen Huang went further when asked what revenue would look like in an unconstrained supply environment. "The unconstrained would be a lot higher," Huang said. "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."

The arithmetic is sobering. Nvidia is on track to double revenue in fiscal 2027, and it might have doubled again in fiscal 2028 were it not for shortages in DRAM, HBM, flash, and other components. Had that second doubling happened, demand would sit just under 40 percent above supply.

The $279 billion in commitments spans the remainder of 2027 through fiscal 2029, with a sliver allocated to fiscal 2030 and beyond. It is reasonable to expect those commitments to keep growing to cover the roughly $1 trillion in Grace-Hopper NVL72 and Vera-Rubin NVL72 systems penciled into the order books across fiscal 2027 — perhaps $365 billion — and fiscal 2028, perhaps $625 billion. If Nvidia ever scales back DRAM and HBM content per GPU, CPU, and DPU to ship more units, those figures could climb higher still.

How the segments shake out

The Compute & Networking group, which sells the datacenter portfolio plus some edge products, brought in $88.3 billion, up 113.6 percent. The Graphics group, covering PC graphics cards and workstation GPUs, contributed $7.92 billion, up 46.4 percent. The businesses that once made Nvidia's name are now statistical noise.

Accordingly, Nvidia has reorganized. Automotive electronics, graphics, and the nascent workstation business now sit in a group called Edge Computing. The Datacenter division splits into hyperscalers on one side and an ACIE cluster — AI Clouds (neoclouds and sovereigns), Industrial, and Enterprise — on the other.

• Hyperscaler revenue: $48.71 billion, up 104 percent year on year.
• ACIE revenue: $40.31 billion, up 134.2 percent.
• Datacenter compute (model estimate): $71.57 billion, up 111.5 percent.
• Datacenter networking (model estimate): $17.45 billion, up 140.7 percent.

The ACIE figure deserves attention. Neoclouds, sovereign buyers, and formerly HPC-only national labs likely account for the bulk of that growth, and it is growing considerably faster than the hyperscaler base.

Networking deserves its own paragraph because it has stopped being an accessory. Within the networking estimate, Quantum-X InfiniBand sales rose 58.9 percent to just under $7.5 billion, leaving roughly $10 billion for Ethernet — around $5.46 billion by this model — and NVSwitch at about $4.5 billion. Networking represented 19.7 percent of datacenter sales in Q1 and 19.6 percent in Q2.

That is roughly double the networking share clouds and hyperscalers prefer. The reason is architectural: in a rack-scale system with NUMA links between GPUs, you are not scaling out individual nodes. You are building one giant virtual GPU out of dozens of real, smaller ones, and the interconnect bill arrives accordingly.

Why the Sun comparison earns its keep

The title of this piece refers to Sun Microsystems and to a phrase coined in 1984 by Sun computer scientist John Gage, employee number five after founders Scott McNealy, Andy Bechtolsheim, Bill Joy, and Vinod Khosla: "The Network is the Computer." Gage said it a decade before the Internet went commercial, which was prophetic. At Nvidia, nearly one datacenter dollar in five now goes to the network, and the ratio is structural rather than cyclical.

One more number for the file. Nvidia spent $7.05 billion on R&D in the quarter, more than twice what it spent two years ago, and that spending drives perhaps 35 to 40 times its cost in future revenue. The money machine is not short of demand. It is short of DRAM, HBM, and flash — and the $279 billion in purchase commitments is what cornering that supply looks like on a balance sheet.