To The Moon

BROADCOM'S 3-DAY NIGHTMARE: When AI Growth Becomes an F-Word (And It's Not the One You Think)

A Humorous Deep-Dive into the Most Paradoxical Tech Sell-Off Since DeepSeek Made Us Question Everything

Imagine this: You're at a dinner party and someone asks how your restaurant chain is doing. You respond with unbridled enthusiasm: "Sales are up 28%! Revenue just hit an all-time record at $64 billion! We just secured $73 billion in advance orders!"

Everyone applauds. You take a bow. Then—inexplicably—the host asks you to leave.

Welcome to Broadcom's week.

THE SETUP: What Happened in Broadcom's Tragic 3-Day Arc

On December 10th, 2025, Broadcom was trading at $412.97—basically the AI darling of the semiconductor world, up 55% year-to-date, riding the wave of hyperscaler demand like a boss. Then came Thursday night's earnings call.

CEO Hock Tan walks into the earnings presentation like a man who's about to deliver the best news imaginable. And he does: Q4 revenue of $64 billion (up 28% YoY), earnings crushing expectations, $73 billion AI backlog locked in over 18 months. It's the kind of print that should've had investors doing backflips in their office chairs.

Instead, what followed was a 12.75% three-day collapse—one of the worst stretches since April 2025. The stock cratered from $412.97 to $359, wiping out roughly $213 billion in market capitalization in 72 hours. Translation: The market collectively decided that record AI revenue + record profit + monster backlog = time to panic-sell like it's March 2020.​

This is what I call "The Expectation Guillotine."

THE PARADOX: Why the Market Hated the Best News

Here's where it gets deliciously stupid.

Broadcom didn't miss. Broadcom didn't disappoint on execution. Broadcom delivered on exactly what Broadcom does—namely, winning the custom silicon race against Nvidia while securing the world's best customer list (Google, Amazon, Microsoft, Meta, Anthropic).

So why the bloodbath?

The culprit: Margin dilution from AI chips. You see, the AI accelerators and custom chips that are now fueling Broadcom's growth come with lower profit margins than the company's traditional business. CEO Tan explicitly warned that gross margins would be pressured throughout the year because—plot twist—more AI revenue = a less profitable revenue mix.​

Think of it this way: If your pizza joint suddenly goes viral for $5 pizza (instead of $20 gourmet pies), sure, you're selling more pizzas, but your margin per pie just got demolished. Wall Street, apparently, would prefer you sell fewer pizzas at higher prices even if it means missing the entire market opportunity.

The Market's Logic: "Wow, record AI revenue! ... Oh wait, you mean it's LOWER margin? Abort! ABORT! Sell it all! The horror! The suffering! Won't somebody think of the gross margins?!"​

THE SPECIFICS: What Spooked the Analysts

Let's break down the actual concerns (some valid, some... less so):

1. The Margin Squeeze (The Legitimate Worry)

Broadcom's custom AI accelerators—the "XPUs" designed with hyperscalers—require complex supply chains, custom engineering, and lower pricing power than, say, their high-margin networking silicon. If AI becomes 50%+ of revenue but AI carries 200-300 basis points lower margins, the operating leverage gets compromised.​

The Response: Morningstar analysts actually made a good point here—these chips are accretive to operating margins even if gross margins dilute. Translation: Yes, gross margin might dip 2-3%, but the AI volume is so large that net operating profit still grows faster than expected. But nuance doesn't sell, so this got buried in the panic.​

2. The Vague Guidance Curse

Hock Tan played it cagey on 2026 revenue guidance. No specific numbers. Just confidence that demand remains "substantial across the board". For investors who've been conditioned to expect crystal-clear projections, this felt like driving with the headlights off on a foggy night.​

(Translation: "We know demand is crazy, but we're not going to give you the comfort of a specific number, because things move too fast." This is actually reasonable. Investors hated it.)

3. The Backlog Concentration Problem

That $73 billion backlog? Roughly 5 customers, with Anthropic alone accounting for ~$11 billion.​

What this means: Broadcom is utterly dependent on 5 companies. If even one of them hits pause—whether due to financial pressure, capital allocation scrutiny, or competitive switches—the backlog could contract materially.​

What this doesn't mean: That hyperscalers will stop building AI infrastructure. These are companies with the deepest pockets and the most at stake. But concentration risk is real, and it got the analyst community anxious.​

4. The Customer Risk That Nobody Wants to Say Out Loud

All of Broadcom's hyperscaler customers—Google, Meta, Amazon, Microsoft—employ teams of brilliant chip designers. Theoretically, they could decide to in-source more of their AI silicon, reducing dependence on Broadcom. Why pay Broadcom for custom chips when you can build it yourself?

Broadcom management has downplayed this risk, noting that their R&D velocity and design wins are sticky. And history suggests they're right—custom chip development is brutally hard, and Broadcom has two decades of infrastructure expertise. But the risk exists, and in a sell-off, every risk gets magnified.​

THE BULL CASE: Why This Dip Is Actually Ridiculous

Here's where the contrarians—and they're legion—make a compelling argument:

Broadcom didn't miss. It overdelivered.

Revenue: $64 billion (record)

EPS growth: 93% year-over-year​

AI revenue: Still accelerating, now ~27% of total revenue​

$73 billion backlog: Locked in demand for 1.5 years​

The margin concern is, frankly, a feature, not a bug.

Lower AI margins now = higher volume + scale + pricing power over time. As Broadcom ships more units and supply chain efficiencies kick in, margins will recover. This is a temporary headwind in service of multi-year hypergrowth.​

Analyst Consensus Is Actually Bullish:

Bernstein analyst Stacy Rasgon: BUY (and upgraded recently)​

Melius Research: "The AI spending pipeline remains substantial; this sell-off is an overreaction to near-term margin concerns"​

Morningstar: The gross margin dip is not worth the panic; these chips are operating-margin accretive​

Price targets? Most analysts see $400-$460 upside from current levels, implying 15-40% returns.​

The Bigger Picture:

The custom silicon market for AI is worth $60-$90 billion by 2027. Broadcom is one of only a handful of companies winning this market. The company isn't facing existential risk; it's facing a temporary margin squeeze while capturing a multi-hundred-billion-dollar market opportunity.​

THE BEAR CASE: Why the Panic Might Be Justified

Let's steelman the opposition:

1. Valuation Is Stretched

Broadcom trades at a hefty forward P/E (multiples vary, but 35x forward earnings has been cited). If margin expansion doesn't materialize, multiples could compress from 35x to 25x, eating a third of the value.​

2. Concentration Risk Is Real

Five customers drive most of the backlog. A single customer delay could trigger a domino effect of disappointment.​

3. Hyperscaler Capital Discipline

We're in a rising-rate environment. Hyperscalers are getting questioned on ROI. If data center spending slows—even slightly—Broadcom's backlog converts to revenue more slowly, pressuring near-term growth.​

4. Custom Silicon Competition

AMD's MI300, Intel's Gaudi, and in-house chip teams are improving. Broadcom's moat isn't impenetrable. If general-purpose GPU improvements continue faster than expected, custom silicon demand could plateau sooner than expected.​

SO: BUY THE DIP OR STAY BEARISH?

My takes (for entertainment purposes only, not financial advice):

For Bull Investors:

This 12-15% dip is a gift if you believe in AI infrastructure tailwinds. Broadcom is one of three companies—alongside Nvidia and maybe one other—that will dominate AI semiconductor design for the next decade. Margin dilution is temporary. $73 billion backlog is real. Buy at $360-$370 if you have a 3+ year horizon.

For Bear Investors:

If you think AI spending slows, hyperscaler capital discipline tightens, or in-house chip