[BLKNSLVR]

And SpaceX will spend $800M per month on Nvidia hardware purchase contacts, and Nvidia will spend $700M per month on Google services.

I'm picturing a teenager blowing a bubble gum bubble bigger and bigger. I assume it can go on forever!

[credit_guy]

The doubt you are expressing was very justified until this week. But on June 1st, Microsoft changed the pricing for business and enterprise Copilot, and people started paying real money for using AI. Until that day, Github Copilot was charging users 4 cents per request (if they exceeded their subscription's monthly limit). Now, they charge at the API rate. I monitor my usage, and it's easily 10 times more expensive than the 4 cents per request before, maybe 20 or even more. And guess what? Businesses are shocked about the changes, and thinking hard what to do, but most of them will just pay 10 or 20 times or more for AI than they used to pay until now.

We will hear projections soon, in a few months, but my guess is that the big 3 (Anthropic, OpenAI, Google) will get of the order of $10 billion per month in AI inference revenues. And it will only go up from there.

[MichaelNolan]

> but most of them will just pay 10 or 20 times or more for AI than they used to pay until now.

At my work, after the GitHub price hike, we all got the option of 1. Keep using Github, but with the same total spend as before. i.e, use 1/20th as much since the dollar cap isn't changing. Or option 2, use as much self hosted DeepSeek v4 Pro, Qwen 3.6, Gemma4 as you want since it's almost free. (to be fair, we already had the GPUs)

If the Chinese keep releasing open weight models that are "close enough" to the big 3, I expect many orgs to make the same choice. I think we will see VPs and higher start saying "you better have a good reason for using Opus 4.8 or GPT 5.5 Pro, do you have metrics showing ${cheaper_model} isn't good enough?"

[credit_guy]

> to be fair, we already had the GPUs

Most businesses don't have the GPUs, or the knowledge necessary to do self-hosted inference. So, they'd have to rely on OpenRouter, or Ollama, or some other inference provider, but there are lots of problems with that. With Microsoft, people could get comfortable with the compliance side of the problem. They already use Outlook, and Office365. Copilot is just one more point where things can go wrong, but it's less scary than your emails being captured and held for ransom, and you already think Microsoft can take care of that. But with Ollama or OpenRouter, you have no idea what is happening with your data, and you also are not sure if they are serving the real models, or quantized versions.

To be sure, there will be plenty of people finding alternate solutions, but 80-90% of the businesses will just pay the higher price to Microsoft.

[MichaelNolan]

For sure not everyone can self host (though I suspect most tech and tech adjacent companies can).

But all the existing “safe/trusted” cloud providers will offer cheaper models. The choice won’t just be GitHub vs OpenRouter. It will be DeepSeek, qwen, Gemma on GitHub vs Opus on GitHub. I’m sure AWS, GCP, Azure, etc will be happy to sell open weight models. And those lower priced models will put a cap on how many people pay for higher priced models.

[layoric]

What makes you think “most of them will just pay 10 or 20 times more for AI”?

They can’t measure ROI, and it will start costing more than their staff. You might be right, but I can’t think why any competent C suit would agree to this..

[BowBun]

Look at cloud spend - how many of your employers have measured the ROI of using cloud vs. self-hosting? At a certain point these things just become the cost of doing business I suppose.

[layoric]

Every single one minus Federal Gov. if you are in a leader position you absolutely should be assessing hosting options, trade offs, costs etc, and looking forward as things grow to make sure you aren’t just burning cash. Only my experience from Australia, the low interest environment could be different in the US, but I would expect this attitude to change if interest rates stay above 5%.

[BowBun]

Of the 4 series A-B startups I've helped grow, none have measured and made decisions on cloud spend based on measurement. The only time this came up is when bills got too large and spend needed to be controlled. You're right that it may be a difference of environment (USA here)

[jmkni]

This is an active conversation going on at my day job right now (and I suspect many other peoples too)

Every developer (we have about 100) has Github Copilot, and interestingly some barely use it while others use it a lot (about 70% of usage comes from a handful of devs), and the dashboard shows you exactly who is using which models, and how much

I definitely don't think they will just go along with paying 10/20x more than before without seeing some sort of return on that investment

We've already had the we're spending all this money on AI, why aren't we shipping software faster conversation multiple times

My prediction is that those high users, costing the most money, will be watched carefully (one colleague even suggested half-jokingly that whoever tops the leaderboard should have to give everyone else a presentation on what they spent all those AI credits on)

The sweet spot is to have good competent developers who users AI when it actually makes sense, but aren't dependent on it

[credit_guy]

As of now, most people haven't figured out how to use AI productively. It takes time. Maybe 1 in 20 developers have come up with a good workflow to get AI-assisted coding done without a lot of slop. The remaining 19 either got burned a few times, or still use AI in a 2023-style: ask AI for a code snippet in a chatbox, then copy-paste it to the code base.

But one or two years from now, many more people will have learned how to be productive with AI. Knowledge will percolate.

And for all those people, the companies will ask themselves: is this guy's 20% increase in productivity worth $200 per month? If that increase in productivity is actually worth $2000 per month, then the answer will be an unequivocal yes. Not only that, but the need to switch to lower cost AI providers, so the $200 is lowered to $20 will just not be worth the extra headache of having to go through all the approvals to onboard a new vendor.

That is the Copilot's moat.

[nish__]

What is Nvidia spending so much on Google services for?

[vdfs]

A bubble will never pop as long as stay in it's bubble form

[manmal]

Usually there’s some environmental disturbance eventually, that makes it burst.

[Waterluvian]

So… data centres in space it is.