Nvidia, Broadcom, Alphabet, Arista: Four Monopolies of the AI Chip War

TL;DR The four AI-infrastructure megacaps each hold a different kind of monopoly — Nvidia in CUDA software, Broadcom in custom-chip design, Alphabet in its own TPUs and cloud, Arista in open Ethernet. With the pullback bringing them down to their 200-day lines, this is my entry zone.

Why I Start With the Megacaps

When I build a basket, I put the weight on the megacaps and scale down toward the small caps. The reason is simple: these companies sit in the seat that gets paid whoever wins AI. June dragged them down too, and I think the best companies grow the most from here.

1. Nvidia: The Real Moat Is Software, Not Chips

Nvidia's monopoly lives in software called CUDA, not in the chips. Almost everything in AI is written on Nvidia's platform. Around 6 million developers build on it, and over a billion of its chips are already out in the world. That lock-in is what no competitor can really break.

What makes it a monopoly in several fields at once is that Nvidia runs the same playbook everywhere. In the data center it does not sell a chip — it sells the entire rack as one computer. Its networking arm alone tripled in the past year. In robotics it builds the platform humanoids are trained in and run on. In quantum it makes the software those machines run, with 17 of the top quantum hardware makers and nine national labs already plugging in. Whoever wins AI, robots, or quantum, they all still pay Nvidia.

Then there is the number that stops me. Over three years revenue went up eight times, from $27 billion to over $215 billion — and the stock got cheaper as it climbed, its P/E falling from over 100 to about 30 because earnings grew faster than the price. The PEG, which weighs price against growth, sits near 0.5, and anything around one or lower says a stock is cheap for its growth. After this pullback, oversold on its 200-day line, that is exactly where I want to own the company the whole AI economy runs on.

2. Broadcom: The Partner Everyone Calls to Escape Nvidia

If Nvidia is the company everyone pays, Broadcom is the one they call when they want to stop paying it. When Google, Meta, OpenAI, or Anthropic wants to build its own AI chip to get out from under Nvidia, it co-designs that chip with Broadcom. And Broadcom sells the networking that wires those chips together, so it gets paid twice on the very same cluster.

That AI business is accelerating, not slowing. AI revenue grew more than 140% just last quarter, is on pace to nearly triple this year, and is guided to nearly double again the year after — all backed by a $73 billion backlog that is now nearly half the entire company. The VMware deal everyone thought would weigh it down did the opposite: VMware's software runs at a richer margin than Broadcom's own chips (around 77% versus 58%), so it lifted the blended margin. That is why the stock looks expensive at around 60 times earnings — that is VMware accounting hiding the real number. On next year's earnings it trades closer to 24 times with a PEG near 0.5, just as cheap for its growth as Nvidia. Back down on its 200-day line, that is my entry into the arms dealer of the whole AI chip war.

3. Alphabet: The Cheapest and Most Beaten-Down Megacap

Alphabet designs its own chips, TPUs, so it is not paying the Nvidia tax the rest of the industry pays, and it builds its own Gemini models on top. The thing to understand is that it is really four businesses moving at very different speeds.

The cash cow is Search, still by far the biggest piece and far from dying — its growth nearly doubled over the past year, from 10% to 19%, as AI mode crossed a billion users and started monetizing questions it never could before. The new rocket is Google Cloud, where revenue grew 63% and the operating margin roughly doubled in about a year and a half, from the high teens to 33%. A unit that used to lose money is now a serious profit engine, and customers have already signed for around $460 billion of future cloud work — a backlog bigger than a full year of Alphabet's revenue. Quietly underneath, YouTube has grown bigger than Netflix and become the most-watched service on American televisions. And Waymo rides along as a free call option, already running more than 400,000 paid robotaxi rides a week, about 10 times what it was two years ago.

One thing to flag: last quarter's profit looked bigger than it truly was, inflated by a one-time paper gain on an investment. Strip that out and the real earnings are a little lower. Even then, it is the cheapest megacap in the group, around 24 times forward earnings with a PEG near 1.3, and the most oversold name here after these pullbacks. It is slowly building momentum but still has a long way to go — and this is exactly where I like to see it.

4. Arista Networks: The Ethernet Winner That Wires the GPUs

Arista builds the high-speed switches that wire all those GPUs together inside an AI data center. Like Broadcom and Google, its real edge is software: every Arista switch, from a campus closet to a giant AI spine, runs the same single operating system, which is why the cloud giants standardize on it.

For years the standard way to connect GPUs was Nvidia's proprietary InfiniBand, and now the whole industry is moving that job onto open Ethernet — Arista's home turf. Meta proved it works by running a cluster of more than 24,000 GPUs on Arista Ethernet and matching InfiniBand. That is why Arista's AI networking revenue doubled last year and is guided to more than double again this year, and much of that demand is prepaid, so deferred revenue has roughly doubled to more than $6 billion sitting on the books. Underneath is a fortress: over six years revenue nearly quadrupled while operating margin widened from 30% to 43%, with zero debt and more than $12 billion of net cash.

The only real catch is the price. Near 44 times forward earnings, its PEG sits close to two — the richest in this group — and the market made that point when Arista beat and raised guidance and the stock still fell. So of all the megacaps, this is the one where price matters a little more, which is why I want it around that 50-day line where buyers tend to step back in.

Sorting It by Valuation

Stock	Valuation	PEG	Entry view
Nvidia	~30x forward	~0.5	Oversold on the 200-day
Broadcom	~24x next year	~0.5	200-day line
Alphabet	~24x forward	~1.3	Most oversold and cheapest
Arista	~44x forward	~2.0	Waiting for the 50-day

My conclusion: on price-for-growth, Nvidia and Broadcom are the most attractive; Alphabet is the cheapest and most compressed megacap, so it has the most room to snap back. Arista is the highest quality but the most expensive, which is the one name where I hold out for a better entry.