원문 보기
Hi everyone. Welcome to Seattle Fire. Nvidia's revenue hit 96.2 billion dollars, over double last year. In his latest interview, Jensen Huang isn't phased by 10x, but the stock is flat from June to September 24th. The episode he was on has a pretty blunt title. Jensen Huang thinks AI alarmism has gone too far.
That's directly relevant to the AI stocks you hold. If AI labs hit the brakes, would compute demand shrink? Today, let's walk through this interview, then follow the compute money to see where it flows. I'll give you one number to track every quarter. Here's my analysis and observations, not investment advice.
This clip runs a bit over 2 minutes. It's from the Ezra Klein show. Halfway through, the host reads an open letter. First, listen for three things. First, where the labs are shifting R&D focus. Second, the tone he uses for that number 10. Third, what should happen if a lab loses control? Are they going to get a lot more issues associated with the product.
This is very normal. And when they have a lot Now now they have so much market footprint, they have to shift their R&D or total R&D from just capability to a lot of verification evaluation and testing. And so to the point where I wouldn't be surprised if the amount of compute necessary to develop these models increased by a factor of 10 because the evaluation is so rigorous.
And but that doesn't that's not where they are today. They're making that transition and I hear them saying it and I'm delight I'm delighted to hear them saying it. But I think the if they believe they're out of control then the right answer is don't ship products until they're in control. It is really quite that simple.
See, I I I I find this perplexing, honestly. Because you just you have so many people these ops professing one, that they're out of control. Yeah. Two, that they are seeing things that are frightening them. why they have that whistle-take whistleblower. And you take the pacing letter that 1,300 plus employees signed.
To realize AI's potential, industry, government, and society at large may need the option to buy time to address emerging risks, develop security measures, and strengthen oversight. But each company and country is under intense competitive pressure not to unilaterally slow down its advances. The labs.
No, no, that last sentence. Nobody's putting the pressure on them. The US I've got a Listen, there are 400 million Americans here. I believe that if everybody were just to take a vote, just right now, let's let's just do this. If they need this, if they need if that's what they need, I'll give them my vote.
Don't ship the product. If your product is not ready to ship, don't ship the product. I have no This is the first time that I've heard a a company or CEO say that I need the laws I need the antitrust laws to be relieved. I need the liability laws of products to be relieved so that I can pace myself.
That paragraph's fantastic. I completely agree. Um auditors, I completely agree. We have financial auditors. That's great. Third-party audits, safety auditors, financial auditors. Um that's all great. That's terrific. Okay, let's start with the first point. He says once an AI product gets useful, problems grow, too.
That's completely normal. Once the market's big, R&D shifts to evaluation and testing. Think of it like a food stall becoming a chain. Food safety can't rely on just the owner tasting it. It needs a full inspection process instead. Same for AI. More testing burns more compute. So, he says the evaluation is so rigorous, he wouldn't be surprised if compute rose tenfold.
He adds that labs are making that shift right now. What does that 10x actually mean? Same model, but 10 times the compute needed. If it gets there, that's a whole new block of demand. Does testing really burn that much compute? Every new version means re-running the whole test suite, checking for wrong answers or dangerous tricks.
More users means more test cases, all needing compute. Now, the second point. Halfway through, Ezra reads an open letter. It calls for labs to slow down. He says it's signed by over 1,300 employees. The letter says industry, government, and society may need the option to buy time for emerging risks, time for safety measures, and stronger oversight.
But competitive pressure makes it hard to slow down alone. Honestly, I get the concern. They want time for safety. Jensen Huang is very direct about it. If a lab believes it's lost control, don't ship. Wait until you're back in control. That's simple. The way I read it, he hands the brakes back to each company itself.
If you feel you can't control it, don't ship it. Bringing up Hinton's doomsday predictions, he gets sharper. He says, "Don't assume being an alarmist is a social good. That's not true." he adds. That's a heavy line, but safety still matters to him. He says he's glad labs are shifting to testing. So, what does this have to do with investing?
To Jensen Huang, safety is work that costs compute. The more seriously it's done, the more machines needed. The third point matters most for your money. You might think he sells compute for a living. He says 10x wouldn't surprise him. That helps him, too. Fair point. So, let's not just take his word for it.
While some call for a slow down, let's check the numbers. Nvidia reported August 26th, $96.2 billion in revenue, up 106% from a year ago, more than double. Net income, almost $60 billion. In other words, $62 of every $100 becomes net income. Earlier, Jensen Huang dropped another number, $500 billion in VC funding went into AI-native firms.
But look at the stock from June 1st to September 24th. It hit $190 at one point, then jumped 9% after earnings. Nvidia is up only about 0.1% start to finish. Before jumping to conclusions, let's zoom out. The extra compute for safety doesn't just go to Nvidia. Let's follow this chain stop by stop. The buyers, TSMC who makes the chips, Micron who makes the memory, then back to Nvidia.
Every number here comes from each company's own filings. Perfect for checking if Jensen Huang is overselling this. Stop one, the buyers. Let's look at Meta. Capex is just money spent on data centers and gear. Meta spent $31 on this in Q2 alone. That's over $300 a day on average. Zuckerberg said AI is accelerating their core business, so they're doubling down.
Meta expects full-year capex of $130 to $145 billion. That's up from a previous low of $125 billion. Buyers aren't pulling back. The source hasn't dried up. Stop two, the foundry. Let's look at TSMC. On September 10th, it reported August revenue of 514.8 billion New Taiwan dollars, up about 10% from July, over 50% from last August.
A company this big still grew 50% plus in August, and the pace is actually speeding up. Jan-August revenue is up nearly 40% from last year. August alone was over 50%. Stop three, memory. AI chips need a fast memory next to them, called HBM. Think of it as the chips' dedicated warehouse. Fast enough data means the chip never sits idle.
Micron, the memory maker, has an even wilder number. Its latest quarter, reported in June, $41.4 billion. Same quarter last year, just $9.3 billion. Up about 346% more than quadrupling in a year. Micron says its new HBM4 is shipping in high volume. The CEO says they're investing at record levels. Micron reports its next results on September 30th.
That's the nearest checkpoint on this chain. Stop four, back to Nvidia. Data center revenue alone was $89 billion last quarter. That's over 90% of total revenue. Data center revenue grew 117% year-over-year, faster than the company's overall 106%. That's exactly the compute safety evaluation eats up.
Put all four stops together. Buyers doubling down, TSMC up 50% plus in August, Micron 4x plus, and Nvidia's data center more than doubled. Nvidia's stock ended four months flat while the chain races. Safety's extra compute stacks on top of a fast chain. Revenue doubled, stock didn't move. How to read that?
Let me share my take. I'm a long-term holder of AI semiconductors myself. This compute demand connects to my long-term position. I agree with the direction Jensen Huang points to. Safety isn't just breaking, testing burns compute, too. The time the letter asks for seems reasonable. That time still gets spent on tests using compute.
Safety evaluation is just one more source of demand. GPUs, foundry, and memory all get used along that chain. A stock flat for months, to me, is just noise. What changes my mind? Big clouds cutting capex. Meta's full-year low and just got up. The day that number gets revised down, I'll reassess. If you already hold AI semiconductor positions, check earnings first, not headlines.
No open letter matters more than slowing revenue growth. Waiting on a pullback? Think this through first. Are you betting on one company or the whole industry? This chain, foundry to memory, moves together. A large semiconductor ETF spreads that bet. So, back to the opening question. Will demand shrink?
By his logic, safer AI needs more compute, not less. That money won't stay with Nvidia alone. Foundries and memory makers get a share, too. The number I promised, Nvidia's YOY revenue growth. This quarter, it's 106%. Compare each new quarter's growth against this 106%. If it drops quarter after quarter, I'll recheck this.
Will AI labs slow down shipping first or buy more compute to run more tests? Comment below and tell me your reasoning. If this cleared up the compute chain for you, hit like, subscribe, and share with friends into AI chips. Turn on notifications, the bell. See you at the next earnings report.
댓글 0