Comment by jmyeet
15 hours ago
So what people seem to be unaware of or are purposely ignoring is that OpenAI and Anthropic have invested trillions in a rapdily depreciating asset. There was a HN post from a day or two ago where someone bought a V100 for 150 pounds and connected it to their computer. Well that was a $10k GPU in 2017. That's the fate of H100/B100 GPUs in 5-10 years (and I suspect closer to 5). What do you do if you've invested $1 trillion that will be worth $100 billion or less in 5 years? I think it'll be worse than that because modern hardware at that time will still probably be the same Wattage but have much higher performance so you'll be getting much higher performance-per-Watt and that's going to really matter.
The only company I'm confident will survive this hardware crunch and still be relatively successful in this space is Google.
OpenAI in particular is a bet that there will be an AI moat and that OpenAI will "win". I don't think there will be a moat and China is a big reason why (eg DeepSeek).
SpaceX is a little different. Yes, launching rockets is a business but it's not a trillion dollar business. 100 Falcon 9 launches doesn't even break $10 billion in revenue. Plus, Starship faces cost overruns, delays and significant headwinds.
But the real kicker is that SpaceX was used to bail out Elon from the Twitter purchase and the xAI investors from the first Twitter bailout. That's a problem because xAI is burning $1 billion a month in a company where that really matters and I don't think Grok will "win" here. Like, at all. SpaceX would be a significantly more attractive company without xAI.
The big potential growth area is Starlink. For that to justify this valuation I think you need handheld Starlink phones. That requires a lot of satellites at a relatively low orbit, which also means they have a relatively short life (because they burn up in the atmosphere). And for that Starship must succeed.
All the AI data center in space stuff is complete bullshit. It makes no sense. It'll never be viable. It's not going to happen.
EDIT: let me clarify because I was careless in my wording. So, Anthropic individually has not spent "trillions". That was more of a general statement on AI spending. Anthropic has raised ~$100B, the last round of which was $65B (at $965B post-money IIRC). This industry as a whole needs to recoup trillions.
Anthropic seems to be in a better position (as a business) than OpeNAI is but I do think the it's a race to cash out before depreciating assets, well, drepreciate and there's the real risk as compute becomes cheaper and the AI craze wears off, Claude just may not have the growth trajectory that is built into the price.
> What do you do if you've invested $1 trillion that will be worth $100 billion or less in 5 years?
I think the aim would be to generate at least $900bn of cash flow from those assets.
As I was reading the start of your argument, I thought you were gonna call the models a depreciating asset! Totally agree about GPUs too, but literally everything they’re spending money on has to be rebuilt to stay competitive. They have to go for the moonshot of training a full new model when better tech comes, they have to upgrade GPUs to keep their data centers efficient.
Technically, the model is a depreciating asset too. Just consider the difference between a model you need a B200 cluster to run vs one you can run on a Raspberry Pi. One's going to have a moat around it that gives it value and the other isn't. It's a hyperbolic argument to be sure but the nature of "enthusiast" hardware is that we're currently running, say, ~27B parameter models on hardware for a few thousand. What's that going to look like in 2 years?
Anthropic/OpenAI really need to train ever-bigger models to keep their moat. But that assumes there isn't a law of diminishing returns and also that a compressed model isn't sufficient for what many people need.
You mihgt say that the training is a barrier. And it is, kind of. Notice how it's Chinese companies coming out with open-source models like DeepSeek and Qwen? That's no accident. As soon as DeepSeek came out I knew what was going on: China is going to make sure no single Western company "owns" AI. It's in their national interest for that not to happen.
I wouldn't be surprised if the rush-to-IPO is motivated, at least in part, by getting ahead of Chinese AI commoditization.
How many failed foundation model training run cycles do you think these companies can tank before the bubble pops and deepseek/etc. catch up to frontier quality?
If Ant, OAI, etc. aren't able to make 20-30% improvements on Opus 4.6 in 2026, does the music stop playing altogether? It seems like they'd lose their ability to charge >10% gross margin on inference in a span of 3-6 months.
"OpenAI and Anthropic have invested trillions in a rapdily depreciating asset". Anthropic raised a bit over 100B and has 47B ARR. Where are you getting trillions from ?
Now that Moores law is dead I think GPUs will depreciate a lot slower. I mean there's already a lot of hardware that has gotten more expensive in the last 5 years.
> I mean there's already a lot of hardware that has gotten more expensive in the last 5 years.
The vast majority of the price rise is mainly due to AI companies sucking all the air out of the room and everyone investing in "AI" regardless.
If China gets their process down to match US/Korea/Taiwan and they decide to flood the market to drown out competitors then hardware is going to be an order of magnitude (or two) cheaper than it is today.
Source on the trillions invested?
[dead]
Starlink Mobile (i.e. Starlink direct-to-cellphone without modifications) is already happening, and fast. Phones that have the recently announced Qualcomm X105 modem will support Starlink Mobile 5G at speeds up to 150Mbps, direct from satellite. The Qualcomm X105 modem will be in most Android flagship phones coming later this year, and by 2027 most new phones will support Starlink direct-to-cell. The next iPhone that supports the 3GPP Rel-19 standard will too.
The rollout relies on Starlink V3 sats, which can only be launched Starship, but Starship progress is going well and is already able to deploy satellites from orbit. SpaceX is capable of launching Starlink V3 on the current iteration of Starship, but they want more testing. We'll probably see Starlink V3 launching late this year or early next year.