Stargate and the AI Industrial Revolution | Hacker News Reader

Stargate and the AI Industrial Revolution

davefriedman.substack.com

21 pointsmhb18 comments

Ugh, I hate hate hate these kinds of polemics.

All this hype around what is basically a data center PLI and a low-moat foundation model company is going to screw our industry over the same way the Telco Bust (remember WorldCom???) did back in the 2000s.

ML/AI is transformative, but the costs to develop and GTM a foundational model are not significant enough of a moat, nor is DC capex a significant moat either, becuase it becomes a commercial real estate play that can demand tech margins. CoreWeave's IPO before the debt offering proves the market softening as well.

The national strategy portion is off as well. Some major players have been ignored and others overstated.

Airlines have multi-billion dollar moats and are constantly on the verge of bankruptcy. If the end user can choose a competitor with near enough performance then it's just a race to the bottom with huge debts and no margins.

I can see moats in AI. Namely when i see an AI company with hard to replicate trained model weights to accomplish a specific task i see a fantastic moat. But $500million in infrastructure? That's just a big airline.

>But $500million in infrastructure? That's just a big airline.

Have I got news for you, it's $500 billion actually.. One hell of a gigantic failing airline we're talking about.

> The implication is stark: models won’t be moats. Infrastructure will.

I think the jury's still out on this one.

Infrastructure size has always been a type of moat, models may become moats, and apps (or how the models are used) may become moats.

Infrastructure is a liability. Obfuscation and/or licensing is the moat.

While that's a great slogan: The infra you need to create and update models dwarfs inference infra. Without that liability, you won't get a working model.

No, the infra itself isn't the moat. Being in an economic position that allows you to expend the resources necessary to run training is one, though. And the cost & speed advantages you realize from utilizing scale (and economies thereof) locally instead of renting means you'll have infra as a signifier of that economic advantage.

And at that point, the additional inference infra is cost-efficient enough that you might as well just have it too, because it's both profit and a better barrier than obfuscation or licensing.

In theory you need the infra V1 to help you design an improve infra v2, and so on and so on. This is a race, not fortification building.

Looking for moats makes sense in a commercial world. I thought the point of the article was that this time its different. You don't need a moat to prevent competitors moving into your market. Nation states are building their own castles, regardless of cost or business case.

This time is, as every single time before, not actually different. The singularity has no calendar appointment for end of 2027, and businesses still care about making money, and will for the foreseeable future.

Nation states with endless pockets are committing to winning the AI arms race.

Business tasked with winning the race will make heaps of money, and everyday people and companies won't be choosing whether or not AI is worth it, we'll be paying with our taxes regardless.

I don;t think the Manhattan Project was business as usual, and I suspect this will be similar.

Obfuscation of what?

There is always a platform/complexity moat. Eventually the costs to build exceed reasonable startup costs. Were starting to see this in leading vendors where so many different things are done well that it would be difficult to replicate.

Unique specialized high-value hard-to-duplicate data and datasets will be frontline moats and provide new competitive edges.

I don't agree "models won't be moats"

If that's true, we'd at least see meta having a SoTa foundational model, and they haven't put one out.

Nor has Microsoft, who would probably love to have some leverage against the OpenAI weird relationship they're in.

On a stylistic note, I was surprised to see David Friedman letting ChatGPT write his posts when he is a perfectly competent writer himself, and then I realized this is Dave Friedman. Turns out to be rather different!

Doesn't really seem like a situation where a defensive moat is the appropriate metaphor. This is a winner-take-all effort. What's the likelihood that it is and how much investment does that justify? It isn't that far-fetched that a reasonable argument could be made for a very, very large investment.

This is all going to be hilarious when the inevitable AI crash happens. Somebody will pick up all of this infrastructure for pennies, and eventually it will come into its own years later. Just like the fibre rollouts of yesteryear.

Power efficiency is the hardest thing to copy. If you're running Nvidia chips, you're going to have a hard time competing with Google.

Prose like an adventure novel, content like a space opera. Lots of overconfident crystal ball gazing. Yep, must be a “future of AI” piece.

Reply on news.ycombinator.com