To The Moon

Hey everyone, Muthu boy is on leave today, so I’m taking over for him.

The Nasdaq and semiconductor stocks really got hammered this week. I have to be honest — I’m hurting quite badly, especially on Fluence Energy. I never expected it to drop this much. Late at night, I quickly cut my GLW at 265 and rotated into PLTR, HOOD, Fig, Bitmine, and RKLB. My boss has been chasing me for my schedule, but I’ve been secretly checking SK Hynix’s share price every morning.

Everyone’s panicking with the same thought: “If these big companies have so much excess computing power, they won’t need to spend on new AI infrastructure anymore. Then what happens to us hardware sellers?” That’s exactly why the entire semiconductor sector plunged these past two days.

But let’s slow down and unpack this properly. Let’s play detective and get the real story straight.

First: Meta is renting out computing power. The big question is — what kind of compute are they actually offering? Their top-tier stuff, or just the idle portion they’re not using?

Let me explain it simply. Unlike Amazon, Google, and Microsoft, Meta never really sold public cloud services. They kept everything for internal use — mainly two things.

One is advertising. We all scroll Facebook, Instagram, and Threads, and you’ve seen how insanely accurate their ads have become. That precision comes from throwing massive computing power at it, and it works.

Second is their own model, Llama. It’s open-source, they started early, and they’re already on the fifth generation. But usage has been dropping. Everyone’s on ChatGPT, Claude, or Gemini now. Meta feels like they’re stuck in 4th or 5th place.

So they asked themselves: “Why keep burning huge amounts of compute training a model when the returns aren’t worth it?” If they slow down Llama training, all that infrastructure they built becomes idle. That’s the “excess” everyone’s talking about.

Simply put: If I can’t win, I’ll join them and make money instead.

Elon’s xAI rented out compute to Anthropic two months ago on a long-term deal and made good money. Meta saw that and thought, “We can do the same.” Instead of wasting money on a struggling model, they’re renting out the GPUs they already own for instant cash flow.

Important note: They’re renting compute hours, not selling the actual GPUs. Big difference.

What they’re offering is mostly older H100 chips from three years ago. The new Blackwell B200s are much more powerful, so they’re smartly monetizing the unused capacity and improving their ROI.

Why do this? Training a big model needs tens of thousands of GPUs running at once. After training, there’s a natural “breathing period” before inference ramps up. If they’re not using the compute this month, why let it sit idle? Rent it out, then take it back when needed.

It’s like owning three houses but living in only one — you rent the other two out. When your family needs them again, you take them back. It’s flexible management.

Meta even turned this into a proper new business called Meta Compute, similar to their own AWS. This shows AI computing power is becoming a public utility, just like electricity and water.

That’s why Meta’s stock jumped while everyone else got crushed. They’re turning idle assets into real cash.

This is actually positive for the whole industry. Startups, SMEs, and universities can now access high-quality compute more easily, which helps more AI applications grow — and that eventually benefits Meta too.

Many people think this means Meta will cut CapEx. I disagree. If this renting business succeeds and boosts EPS, they can easily tell shareholders, “See? It works. Let’s go bigger.” CapEx could actually increase, not decrease.

We’ll find out more at their earnings call next month. If CapEx stays around $130 billion or even goes higher, that’s a very bullish signal.

For the AI supply chain — TSMC, SK Hynix, PCBs, DRAM, etc. — this is not bad news. If anything, it’s supportive.

Bottom line: The market overreacted. AI compute is not in oversupply. We’re still very early. Soon we’ll have AI agents (not just chatbots) running 24/7, burning hundreds of times more tokens. Add robots, self-driving cars, smart glasses… the real demand explosion is still coming.

So even though I’m sitting on some painful losses right now, I’m not bearish on AI infrastructure at all. The revolution isn’t ending — it’s just entering its golden growth phase.

Stay strong, guys. We got this 💪