WHY ISN’T ANYBODY ASKING THE HARD QUESTIONS?

(AVGO), (CSCO), (META), (INTC), (NVDA), (AMD) 

I was having coffee with a semiconductor engineer last week when she mentioned something that made me pause mid-sip. 

“Everyone talks about Nvidia’s AI dominance,” she said, “but half the infrastructure running their chips comes from Broadcom (AVGO). The problem is, nobody’s asking what happens when the music stops.”

That conversation stuck with me, especially after diving into Broadcom’s latest quarterly results. Sure, the numbers look impressive on the surface. 

Revenue jumped 22% year-over-year, semiconductor sales climbed 26%, and the company is sitting on $10.7 billion in cash. But here’s what keeps me up at night: trading at 92 times earnings for a company whose entire thesis depends on the AI infrastructure boom continuing indefinitely.

I’ve been watching tech cycles long enough to know that what goes up with breathtaking speed can come down just as fast. Remember when everyone thought Cisco (CSCO) would rule the internet forever? Yeah, that didn’t age well either.

The bulls make compelling arguments. Broadcom’s chips touch 99.9% of internet traffic, they’ve got partnerships with OpenAI and Meta (META), and they’re positioning themselves as the infrastructure backbone for AI inference. 

The VMware acquisition reduced customer concentration risk, and they’ve actually paid down debt from $66 billion to $62.8 billion while growing cash reserves. On paper, it’s a beautiful story.

But paper and reality don’t always match. What bothers me isn’t what the company is doing today, it’s what they’re betting their future on. 

Custom AI processors sound revolutionary until you realize they’re essentially becoming a contract manufacturer for tech giants who could decide to bring chip design in-house tomorrow. Ask Intel (INTC) how that outsourcing trend worked out for them.

The inference market opportunity is real, but it’s also speculative. Yes, inference workloads will eventually dwarf training, but when? And at what margins? 

Broadcom’s current valuation assumes they’ll capture meaningful market share from Nvidia (NVDA) and AMD (AMD) while maintaining premium pricing. That’s a lot of assumptions stacked on top of each other.

Another thing missing from the conversation is return on invested capital. 

Despite all the AI hype, Broadcom’s ROIC sits at a mediocre 13.77%, hardly the kind of capital efficiency you’d expect from a company supposedly riding the next great technology wave. Their return on assets tells a similar story at 11.34%. 

These aren’t terrible numbers, but they’re not the metrics of a company creating sustainable competitive advantages either.

The debt situation deserves scrutiny, too. Yes, they’ve reduced long-term debt, but $62.8 billion is still substantial for a company generating $18.9 billion in net income. 

When interest rates eventually normalize, that debt service will bite harder into cash flows. The timing couldn’t be worse if AI infrastructure spending slows simultaneously.

Then there’s the software question. Broadcom keeps talking about competing with Nvidia’s CUDA ecosystem, but software has never been their core competency. 

Building a software platform that developers actually want to use takes years, and Nvidia’s got a massive head start. Broadcom’s betting their customers’ engineers will fill this gap, which feels optimistic at best.

I keep thinking about my engineer friend’s comment. The infrastructure boom driving Broadcom’s growth depends entirely on continued massive AI investments from a handful of hyperscale customers. 

What happens when those customers start questioning AI ROI? What happens when the next generation of chips requires different infrastructure? What happens when geopolitical tensions disrupt the delicate global semiconductor supply chain?

The market’s pricing Broadcom for perfection, assuming everything goes right for the next several years. EPS estimates show 38% growth this year, 36% next year, and 28% the year after. 

Those are spectacular numbers, but they’re also exactly what you’d expect analysts to predict during the peak euphoria phase of a technology cycle.

I’m not saying Broadcom is a bad company. They’ve built legitimate competitive moats in networking infrastructure, and their engineering expertise is world-class. 

But at current valuations, you’re paying for a best-case scenario with little room for disappointment. 

When I see a stock trading at nearly double its five-year average multiple during what might be peak AI hype, every instinct I’ve developed over decades of investing starts flashing warning signals.

The smart money isn’t chasing momentum at these prices. It’s waiting for reality to catch up with expectations, which in this business, always happens eventually.

My coffee companion understood what most people missed: just because everyone’s using your pipes doesn’t mean you own the water rights forever. That one-liner just saved me from chasing a 92x multiple. Her next coffee’s definitely on me.