Good morning, AI tollbooth watchers.
Everyone looked at last week’s NVIDIA’s earnings and saw another massive quarter.
The bigger story is that Jensen Huang is turning NVIDIA from the chip supplier of the AI boom into the toll road every AI company needs to keep scaling.
NVIDIA just reported another unusually strong quarter.
Revenue hit $81.6B, up 85% year-over-year.
Its Data Center business, which is basically the part selling chips and systems for AI, made $75.2B (up 92%).
And the company guided for $91B next quarter while assuming zero China revenue, which is wild because China used to be a major buyer.
Source: Fiscal
But the most important number was buried deeper: Networking revenue hit $14.8B in one quarter, up 199% year-over-year.
Networking sounds boring, so here’s the simple version.
When companies build giant AI systems, they do not just buy one powerful chip and call it a day.
They buy thousands of GPUs, then need all those chips to communicate with each other extremely fast, almost like one giant computer.
That requires switches, cables, connectors, network cards, and software that helps everything move together (it’s very complex).
Source: Nvidia
GPUs = the brain / Networking = nervous system
Without it, you just have thousands of expensive chips sitting in separate rooms, unable to work together properly.
That is why the 199% growth in networking revenue tells us the AI buildout is getting bigger, harder to manage, and more dependent on NVIDIA’s full system.
On May 14, we published AI Is Eating All The Memory and said AI was starting to squeeze the memory market.
Since then, Micron is +31%.
And DRAM ETF, the clean memory play we sent to readers who replied, is +23%.
Think this was early?
The next setup could be even bigger.
Reply “SKY” and we’ll send you the special edition before the crowd catches on.
This is where NVIDIA’s story gets bigger than GPUs.
For most of this AI buildout, NVIDIA looked like the company selling the most important tool everyone needed (chips).
Now the business is starting to look more like the company selling the entire operation around that tool.
The chip is still the centerpiece, but the real power comes from everything around it:
CPUs that coordinates the work
Networking that connects the chips
Software (CUDA) that keeps developers inside NVIDIA’s world
As AI gets bigger, the bottleneck changes.
At first, everyone needed more GPUs.
But once companies start building massive AI factories, the harder problem becomes making 10,000 or 100,000 chips behave like one giant machine.
That is where NVIDIA’s full stack starts to matter.
This also addresses how we should think about the biggest threat to NVIDIA.
A cheaper custom chip from Google (TPUs) or Amazon (ASICs) may scare some NVIDIA holders.
But if the whole system is already built around NVIDIA’s networking, CUDA software, and data center architecture, switching becomes painful.
NVIDIA makes the AI factory easier to build.
And in a world where every AI company is racing to scale, the easiest and faster system to build around may become the hardest one to leave.
Quick question before you go:
If NVIDIA’s next big edge is the full AI factory, which layer do you think matters most?
Reply with one letter:
A — GPUs, the brain still wins
B — Networking, bigger AI needs faster connections
C — CPUs, agents need coordination
D — CUDA/software, switching is painful
I’ll read every reply and share the most interesting takes in a future issue.