Nvidia posted $96.22 billion in Q2 fiscal 2027 revenue, up 105.9% year on year, with operating income rising 124.1% to $63.73 billion and net income reaching $59.96 billion (up 125.9%). Purchase commitments hit $279 billion, up 2.3X from the $119 billion secured through Q1.

The July quarter confirmed what component suppliers have been signaling for months: server-side demand continues to outrun allocated supply by a wide margin — roughly 40% in Nvidia's case.

How large is the supply-demand gap?

"The unconstrained would be a lot higher," co-founder and chief executive officer Jensen Huang told analysts on the call. "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."

Chief financial officer Collette Kress was equally direct. She said Nvidia had secured enough components to grow the datacenter business "by slightly more than 70 percent" in fiscal 2028. Huang implied that Nvidia could have doubled revenues again in F2028 absent shortages in DRAM, HBM, flash, and other parts.

That arithmetic — doubling in F2027 and roughly 70% growth in F2028 versus a potential second consecutive doubling — leaves demand about 40% above what constrained supply can deliver. The shortfall mirrors figures from elsewhere in the supply chain. Intel described its overall CPU gap as a number "starting with a B" in April, implying at least 20% and likely closer to 25%. AMD's Epyc line appears to face 30%-plus over-demand. Several OEMs have reported 25%–30% gaps.

What drove the quarter?

The Compute & Networking segment — Nvidia's datacenter core — delivered $88.3 billion, up 113.6%. Within it:

• Hyperscale customers: $48.71 billion, up 104%
• ACIE (AI Clouds, Industrial, Enterprise): $40.31 billion, up 134.2%

The Graphics business, now folded with automotive and workstation products into a renamed Edge Computing division, contributed $7.92 billion, up 46.4%.

Why does networking now account for 19.6% of datacenter revenue?

Networking hit an estimated $17.45 billion in Q2, up 140.7%. Compute reached $71.57 billion, up 111.5%. Networking's share of datacenter sales held at 19.6% — about double the typical hyperscaler ratio.

The split:

• Quantum-X InfiniBand: ~$7.5 billion, up 58.9%
• Ethernet: ~$5.46 billion
• NVSwitch: ~$4.5 billion

In rackscale systems like Grace-Hopper NVL72 and the upcoming Vera-Rubin NVL72, NUMA links between GPUs force roughly twice the networking spend of conventional hyperscaler clusters. The architecture converts dozens of smaller accelerators into one large virtual GPU.

What does the forward supply picture look like?

The $279 billion in purchase commitments covers late F2027 through F2029, with a small slice allocated to F2030 and beyond. Nvidia has penciled in roughly $1 trillion in sales for Grace-Hopper and Vera-Rubin NVL72 systems across F2027 (~$365 billion) and F2028 (~$625 billion). If memory supply loosens, those figures could climb.

R&D outlays underline the runway. Nvidia spent $7.05 billion in Q2 — more than double the figure from two years ago — driving the Grace-Blackwell and Vera-Rubin roadmaps.

Sun's prophecy, Nvidia's realization

John Gage, computer scientist and employee number five at Sun Microsystems, coined "The network is the computer" in 1984. The line came two years after Sun's founding and a decade before the commercial Internet. Four decades later, with networking holding a structural 19.6% of datacenter sales at Nvidia, that vision has finally materialized at scale — though the computer now belongs to Nvidia more than it ever did to Sun.