Every few months the market holds its breath for one of those earnings calls that actually moves the entire semiconductor sector.
Today is one of those days.
Broadcom steps up to the plate after the close, and honestly, the stakes feel higher than usual. The stock has already doubled in the past twelve months and sits near all-time highs, yet the street’s average price target barely budges above current levels. That disconnect alone makes this print fascinating.
But forget the valuation debate for a second. What really matters tonight is whether the AI frenzy still has legs, or if we’re starting to see the first real cracks in the narrative.
The One Number Everyone Is Watching
Let’s cut through the noise.
Wall Street expects Broadcom to guide something around $20-21 billion of AI-related revenue for fiscal 2025. That’s roughly 60% growth from where we sit today. And several top-tier analysts think the company could comfortably push toward $50 billion in AI revenue by fiscal 2026.
Yes, you read that right. Fifty billion. With a B.
In my experience covering this space, those kind of numbers usually get laughed off as overly optimistic, except nobody is laughing right now. The order patterns from hyperscalers look that solid.
Custom ASICs: From Sideshow to Center Stage
A few years ago, Broadcom’s custom chip business (the application-specific integrated circuits or ASICs that hyperscalers design for their own data centers) was interesting but hardly the main event.
Fast forward to 2025 and it’s arguably the most important growth driver the company has.
Google’s TPU program has been the anchor tenant for years, but the newest Ironwood generation is apparently ramping faster than anyone expected. Add in ByteDance, Meta, and now very credible chatter around OpenAI and potentially a fifth major player, and you start understand why analysts keep nudging numbers higher.
“We return to AVGO as our top pick as we see the larger upside to estimates as ASICs hit an inflection point.”
Jefferies analyst team, early November note
That inflection point comment keeps popping up lately. And when Jefferies calls something their top pick in semis, people listen.
The Mystery Customer Everyone Thinks They’ve Solved
Last quarter management dropped a bombshell: a single customer committed to roughly $10 billion of XPU (their branding for these massive custom accelerators) over multiple years, built at rack-scale from the ground up.
They refused to name the partner. The street immediately started playing detective.
Most fingers now point squarely at OpenAI. The timeline matches reports of their first in-house AI training chip launching in 2026, and the dollar figure lines up almost too perfectly. Management has a chance tonight either confirm or keep us guessing, but the smart money already treats it as fact.
Google’s TPU Ramp Looks Almost Scary Strong
Perhaps the most under-appreciated part of the story is how dominant Google’s TPU has become in certain workloads.
Remember when everyone assumed Nvidia had the inference market completely locked up? Turns out Google quietly built a monster competitor, manufactured exclusively by Broadcom, and now there’s talk of Google actually selling TPU capacity to third-party clouds.
If that merchant model takes off, the addressable market for Broadcom suddenly gets a lot bigger than even the bulls modeled twelve months ago.
- Ironwood (current gen) still in early ramp
- Next-gen TPUv7 already taping out on 3nm
- Potential merchant sales starting late 2025 or early 2026
That combo could drive meaningful upside to even the most aggressive forecasts.
Networking: The Quiet Multi-Bagger Hiding in Plain Sight
Lost in all the custom ASIC excitement, Broadcom’s networking franchise keeps printing money.
The Tomahawk 5 switch chip is sold out for quarters, and Tomahawk 6 samples are apparently blowing away customer expectations on power and bandwidth. Every time a hyperscaler adds another 800G port, Broadcom collects a very healthy check.
JPMorgan believes networking alone could contribute mid-teens growth in 2025, then accelerate again in 2026 as the next-gen platforms hit volume. That piece often gets overlooked when people focus purely on accelerators.
What About the Non-AI Businesses?
Here’s where things get interesting in a different way.
Broadcom’s legacy semiconductor segments (wireless, broadband, storage, industrial) have been in various stages of cyclical downturn for almost two years now. Most analysts expect those markets finally bottomed in calendar 2024 and should start recovering through 2025.
Add in the VMware integration synergies that still have several quarters of cost saves left to drop to the bottom line, and you have a setup where earnings power could surprise dramatically to the upside even if AI demand merely stays flat.
Deutsche Bank recently highlighted a $110 billion billion total backlog, the highest in company history. That kind of visibility in a cyclical recovery is almost unfair.
Valuation Feels Expensive Until You Run the 2026 Numbers
Current-year multiples look rich, no argument there. But start pushing earnings estimates out to fiscal 2026 and suddenly the stock screens reasonably priced for the growth profile.
Several firms already model north of $25 in non-GAAP EPS for 2026. At today’s price that’s barely 16x forward earnings for a company growing revenue 30%+ with expanding margins.
I’ve found that when the street’s consensus target sits below the current price yet the fundamental story keeps improving, those situations often resolve in favor of higher targets rather than lower prices.
Risks That Actually Matter
Of course nothing this good comes without real risks.
- Customer concentration remains sky-high – three or four hyperscalers drive the vast majority of AI growth
- Any meaningful delay in TPUv7 or the OpenAI program would dent 2026 numbers
- Gross margin pressure from newer rack-scale programs is real (though management claims they priced accordingly)
- Geopolitical tension around Taiwan (TSMC manufactures almost everything leading-edge)
Those are legitimate concerns, but they’re also largely priced in at current levels, in my view.
Bottom Line Heading Into Tonight
Look, I’m not here pound the table and tell you Broadcom is cheap, because it’s not.
But the growth trajectory looks stronger today than it did six months ago, and dramatically stronger than twelve months ago. When a company consistently raises the bar while the stock rerates higher anyway, that’s usually the sign of a genuine multi-year winner.
Tonight we find out if management sees the same runway extending even further than the street currently models. My gut says they do.
Either way, this earnings report will tell us a lot about where the entire AI infrastructure build-out is headed in 2025 and beyond. Strap in, it should be a wild ride after the close.
