Mark Zuckerberg Has Not Earned This Much Money to Spend
Three years ago, Mark Zuckerberg renamed his company after a product nobody wanted, lit roughly $90 billion on fire chasing it, and asked shareholders to trust his vision. Reality Labs has now posted cumulative losses north of $90 billion since Meta began breaking it out, and last quarter alone delivered $402 million in revenue against a $4 billion operating loss. The metaverse is, by any honest accounting, the most expensive personal hobby in the history of capitalism. So when the same CEO returns to investors and announces he is now raising 2026 capital expenditures to as much as $145 billion — nearly double last year's spend, and more than the GDP of most countries — the burden of proof should sit squarely on him. On Wednesday's earnings call, an analyst asked Zuckerberg the only question that matters: what signs are you watching to know this investment is paying off? His answer, in full, was "that's a very technical question." That is not an answer. That is a man who does not have one.
The contrast with his peers is what makes the moment damning. On the same day, Sundar Pichai walked through Google Cloud's 63% revenue growth, an enterprise backlog of $462 billion, and 800% year-over-year growth in revenue from products built on its generative AI models. Satya Nadella laid out exactly where Microsoft's $190 billion is going — GPUs and CPUs to meet Azure demand he can quantify. Both stocks rose. Meta's fell 9% after hours. The market is not punishing AI capex; it is punishing a CEO who cannot explain his own. And the spillover from his inability to articulate a thesis is not contained to his shareholders. Meta is now an incremental buyer in a hyperscaler cohort whose combined 2026 capex will exceed $600 billion, and the company itself blamed memory component pricing for the upward revision — meaning Zuckerberg's open-ended bid for compute is one of the forces driving up costs for every other firm trying to build a real AI business with a real plan.
There is a version of this story where Zuckerberg is right and the rest of us look foolish in five years. Meta's ad business is genuinely strong, the company grew revenue 33% last quarter, and superintelligence, if it arrives, will pay for everything. But "trust me" is not a capital allocation framework, and Zuckerberg has not earned the trust. He missed on the metaverse, then quietly cut Reality Labs' budget by 30% and laid off 8,000 people on May 20 to fund the next bet. He has now told his own employees, on the record, that he does not have a "crystal ball plan for the next three years" — while committing the company to a spending trajectory that requires exactly such a plan to be coherent. Public companies are not personal sandboxes. At some point, a board, a shareholder base, or a regulator is going to insist that the man spending $145 billion of other people's money be able to articulate, in sentences a sixth-grader could follow, what he expects to get for it. Until then, the only honest answer to "should we trust Mark Zuckerberg's vision again" is the one he gave the analyst: that's a very technical question.
@nypost@nytimes@WSJbusiness@elonmusk@Meta
@OGCapital25 word of advice: do not lose it! you caught a once in your lifetime move and might be confusing a good process with a good process and some good luck
Google $GOOG selling $35B+ of their chips to their second largest competitor, only to then lease chips from SpaceX $SPCX for ~$1B/month is a real head scratcher.. If Google is as capacity constrained as their $85B equity raise and deal with SpaceX would have you believe, why did they commit to give even a single TPU to Anthropic..?
We call it like we see it. Mike Wilson at MS was so bearish and dead wrong for so long that when he turned bullish he only sees green! Ignores everything, rates, oil, misses, bogeys, just sees solve for up!
I wrote this article on $AMD when it was about $150. I was was banging on the table for everyone to buy at a time when people called Lisa Su DEI and thought $AMD had no chance of holding its own against $NVDA.
Fast forward a mere 8 months, and my position has just about 4xed. With real size.
If you want, real, deep, and thoughtful critiques and analysis of companies, I'll give it to you.
Here is how my portfolio has performed over the last few years:
1. PLTR at $17 - now $145
2. ASTS at $20 - now $100
3. SIVE at 8 SEK - now 80 SEK
4. AMD at $80 - now $500
5. HIMS at $13 - now $28
6. TSLA at $220 - now $420
No fluff, no pumpy content - just how I see things.
Also big shout out to @alc2022 for being bang on about $AMD. His analyses on the company really helped me form my conviction. Antonio takes a longer term view than most, but his thinking is excellent.
@dylan522p@SemiAnalysis_ As a paid subscriber I get it but the markets sure didn’t. It was like “I shot the sheriff (Micron Samsung Hynix) but I didn’t not shoot no deputy (QQQ). ☮️
FWIW even @ChatGPTapp knows the math is 5x not 50%
Yes — the VR data point is real, not just “MS said it.”
Nvidia’s own VR NVL72 spec says:
* 54 TB LPDDR5X CPU memory per VR NVL72 rack
* 20.7 TB HBM4 GPU memory per rack
* 72 Rubin GPUs / 36 Vera CPUs
So the core Morgan Stanley direction is right: VR has a ton more rack-level non-HBM DRAM/SOCAMM than GB-class systems.
The nuance:
* True: VR uses 54 TB LPDDR5X per rack.
* True: VR uses 20.7 TB HBM4 per rack.
* Not proven from Nvidia: the exact $2.0M memory BOM number. That is MS’s cost estimate, not an official Nvidia price.
* Also important: Nvidia labels specs as preliminary / subject to change.
So my answer: yes, the physical VR memory content is confirmed as huge. The dollar BOM is an estimate.
We are clearly so back to doomsayers on X! So back! Everyone runs with same narratives too. Maybe X is a secret AI agent! Good news is that costs tokens!
Did Sam Really Say This? Yes...
Sam: "The second is cost. AI spend has become a genuine problem, and fast."
Sam: “People are really saying — that’s kind of become a meme now — ‘my company spent my entire 2026 budget in Q1. Can you make this more efficient?’”
Me: Watch if usage slows at all as next narrative (or narrative shift). Too early to say imho