To The Moon

For the last few years, the AI semiconductor story has been remarkably simple.

AI spending goes up.

Demand for GPUs explodes.

Nvidia wins.

But Broadcom’s upcoming earnings could test whether the next phase of the AI boom is becoming much more complicated.

Broadcom reports fiscal Q3 earnings on September 2, and expectations are enormous. Wall Street is looking for roughly $29.4 billion in revenue and $3.24 in adjusted EPS.

But those headline numbers are not what interests me most.

Broadcom has already guided for approximately $16 billion in AI semiconductor revenue, representing growth of more than 200% year over year.

That raises a much bigger question:

Does Nvidia actually need to lose for Broadcom to become one of the biggest winners of the AI infrastructure boom?

I don’t think it does.

And that may be the most important part of the Broadcom story.

📊 THE NUMBERS ARE ALREADY HUGE

Broadcom’s previous quarter was extraordinary.

Revenue reached $22.2 billion, up 48% year over year.

AI semiconductor revenue reached $10.8 billion, up 143%.

Adjusted EBITDA hit approximately $15.2 billion, representing 69% of revenue.

Free cash flow reached approximately $10.3 billion, or 46% of revenue.

Now Broadcom expects total Q3 revenue of approximately $29.4 billion, up 84%, while AI semiconductor revenue is expected to jump from $10.8 billion to roughly $16 billion.

That means Broadcom isn’t simply participating in the AI infrastructure boom anymore.

It is becoming one of its central players.

But the reason why is where things get interesting.

🧠 THE AI MARKET MAY BE MOVING BEYOND ONE CHIP

Nvidia’s greatest strength is obvious.

Its GPUs are extraordinarily powerful, CUDA has created a massive software ecosystem, and the company remains deeply embedded across AI training and accelerated computing.

But hyperscalers are spending extraordinary amounts of money building AI infrastructure.

At that scale, economics matter.

Power matters.

Efficiency matters.

Networking matters.

And eventually, designing chips specifically for certain workloads starts making more sense.

That’s where Broadcom enters the story.

Broadcom works with hyperscalers to develop custom AI accelerators, sometimes referred to as XPUs or ASICs, while also providing networking technology connecting enormous AI clusters.

Meta is a perfect example.

Meta and Broadcom are working together on multiple generations of Meta’s MTIA custom silicon, including a 2nm AI accelerator, with an initial deployment exceeding 1 gigawatt and plans for a multi-gigawatt rollout.

Broadcom isn’t only helping build the accelerator.

Its Ethernet technology also helps connect the infrastructure surrounding it.

Then there is OpenAI.

OpenAI recently unveiled its first custom AI chip developed with Broadcom, aimed primarily at AI inference, where trained models actually process requests.

Put those developments together and something interesting starts appearing.

The AI chip market may not be becoming less valuable. It may be becoming less uniform.

🟢 THE BULL CASE: BROADCOM DOESN’T NEED TO KILL NVIDIA

I think this is where the Broadcom thesis is sometimes misunderstood.

The bull case does not require Nvidia GPUs to disappear.

It doesn’t even require Nvidia to stop growing.

Instead, imagine an AI infrastructure market large enough to support several architectures simultaneously.

Nvidia remains dominant where flexibility, ecosystem and frontier computing matter.

Meanwhile, hyperscalers increasingly deploy their own custom accelerators for massive, predictable workloads where performance per watt and cost efficiency become critical.

Broadcom potentially benefits from both the custom silicon itself and the networking required to connect increasingly enormous AI clusters.

That creates an interesting possibility:

Nvidia could continue selling record amounts of GPUs while Broadcom’s AI business also continues exploding.

If that happens, this isn’t necessarily a battle over a fixed pie.

The pie itself is getting bigger.

🔴 THE BEAR CASE: $16 BILLION IS NOW THE STARTING LINE

There is one enormous problem with this story.

Expectations.

Broadcom has already told investors to expect approximately $16 billion of AI semiconductor revenue this quarter.

That represents growth above 200%.

At some point, investors stop asking whether AI is growing.

They start asking:

Can growth this extreme continue long enough to justify the valuation?

That is a much harder question.

Broadcom could report objectively excellent numbers and still disappoint Wall Street if investors were expecting something even better.

We’ve already seen this phenomenon across AI stocks.

When expectations become extreme, beating estimates isn’t always enough.

The company also faces concentration risk.

Only a relatively small group of hyperscalers can deploy custom AI accelerators at this scale. Those customers are enormous, technically sophisticated and capable of negotiating aggressively.

And custom silicon doesn’t automatically destroy Nvidia’s competitive advantages.

CUDA remains formidable.

Nvidia’s ecosystem remains enormous.

Its technology roadmap keeps moving.

So I think the simplistic argument that custom ASICs equal the death of Nvidia misses what is actually happening.

💰 THE $100 BILLION QUESTION

Broadcom has previously indicated that its AI semiconductor opportunity could exceed $100 billion in annual revenue during 2027.

Think about what needs to happen for Broadcom to reach that kind of scale.

Custom AI accelerators need to become permanent infrastructure rather than experiments.

Hyperscaler capital expenditure needs to remain enormous.

AI inference needs to keep expanding.

Networking requirements need to rise as clusters become larger.

Customers like Meta need to continue scaling their custom architectures.

And Broadcom needs to execute across several generations of increasingly complicated silicon.

If those things happen, Broadcom stops looking like merely another semiconductor company benefiting from AI.

It starts looking like one of the companies architecting the infrastructure underneath the AI economy.

🎯 MY TAKE

This is what I’ll be watching when Broadcom reports.

Not simply whether revenue beats $29.4 billion.

Not simply whether EPS beats expectations.

I want to know:

Does AI semiconductor revenue reach or exceed the $16 billion target?

Does management raise the forward AI trajectory again?

How much of the growth is coming from custom accelerators versus networking?

Is demand broadening across customers?

Can Broadcom maintain extraordinary profitability while its AI business scales?

And most importantly:

Does Broadcom give investors more evidence that custom silicon is becoming a permanent second pillar of AI compute?

Because I don’t think Broadcom needs to beat Nvidia.

I think Broadcom needs to prove that the AI market is becoming large enough for hyperscalers to want Nvidia AND their own silicon.

If Nvidia keeps growing while Broadcom’s custom AI business also explodes, that could tell us something much bigger than who wins a semiconductor rivalry.

It could tell us that the AI infrastructure opportunity itself is still expanding.

And if that’s true, investors may eventually stop asking:

“Who replaces Nvidia?”

And start asking:

“How many winners can this AI buildout actually create?”

👇 What do you think?

A) Nvidia remains overwhelmingly dominant

B) Broadcom and custom silicon take meaningful share

C) Both win because the AI market becomes much larger

D) Expectations across AI hardware are already too high

For me, C is the scenario I’m watching most closely.

$AVGO $NVDA $META $GOOGL $MSFT $AMZN

Personal market analysis only. Not financial advice. Always do your own research.