Report: Deal marks shift from exclusive Microsoft partnership as AI compute needs soar.
See full article...
See full article...
OpenAI signs surprise deal with Google Cloud despite fierce AI rivalry
- Thread starter JournalBot
- Start date
Google sells elastic infrastructure to a company which has heavy demands for elastic infrastructure.
I'd happily take a pile o' cash from a company which is planning to burn billions of dollars in the next few years. Even if they're a rival... it'll push Google forward too.
From Yahoo Finance
I'd happily take a pile o' cash from a company which is planning to burn billions of dollars in the next few years. Even if they're a rival... it'll push Google forward too.
Upvote
74
(74
/
0)
Why would google turn down money for infrastructure services? It's totally separate from its AI business.
Upvote
63
(63
/
0)
I'm not surprised that Google is playing nice right now. Google is probably feeling "I just got caught, but Dad hasn't made up his mind so I should be an extra good boy" at the moment.
Upvote
-2
(7
/
-9)
It's not like a pie, in which a bigger piece for me means a smaller piece for you. When the AI bubble bursts all will share equally in the failure.
Upvote
43
(46
/
-3)
In a gold rush, the best way to make money is selling shovels. And Google just sold a lot of shovels.
Upvote
74
(76
/
-2)
Slightly off-topic: I guess I'm approaching the "get off my lawn" stage of life, because using "compute" as a shorthand for processor cycles or processing power will always sound strange to me.
Proceeds to yell at cloud about those kids and their newfangled words.
Proceeds to yell at cloud about those kids and their newfangled words.
Upvote
14
(14
/
0)
You’re right, it was not a noun, but a verb which has become a noun exactly for that reason - as a shorthand for processor cycles or processing power. Such is the development of a language.
Upvote
15
(15
/
0)
You know how some businesses have their first dollar framed and hanging on a wall somewhere? I can certainly imagine a future Google CEO hanging the last dollar OpenAI paid them as OpenAI went bankrupt hanging on a wall.
Upvote
25
(26
/
-1)
https://www.wheresyoured.at/openai-is-a-systemic-risk-to-the-tech-industry-2/
Considering some of their existing obligations it feels almost like they're just flailing around, hoping to find the next dumb f*cker who'll sign a deal with them, so at least the headlines seem positive until they crash and burn...
Considering some of their existing obligations it feels almost like they're just flailing around, hoping to find the next dumb f*cker who'll sign a deal with them, so at least the headlines seem positive until they crash and burn...
Upvote
28
(28
/
0)
Yeah, I know, and I'm not about to go all prescriptivist on language use. My generation had its way of talking, today's has theirs. Just noticing that I'm getting too old to bother staying current with it.
Upvote
12
(12
/
0)
sigmasirrus
Ars Scholae Palatinae
I think LLMs will become commodified. Already the open source, open weight models are pretty good. OpenAI doesn’t really have a moat. They merely have attention. The one use I have found LLMs to be helpful for is coding. But I don’t care who makes the model, as long as it works well. Given OpenAI’s legal troubles and their being forced to retain everything you give them, I will not be using them at all for coding. Claude seems to be pretty good at it. But if there were a local model that specialized in coding that I could run on my own hardware and would be as good as Claude, I would use that instead. So the money can’t really be in proprietary models in the long run…
Upvote
15
(15
/
0)
Upvote
1
(2
/
-1)
Yup, Google needs to make very sure that they're getting paid. This client has eleven figures in debts and loss. The funding from Softbank is still +/- 10 BILLION USD of uncertainty, pointing towards "no," after the for-profit conversion stalled out. Gold rush era shovel-makers got their cut up front and were dealing with much, much smaller scales.
Or they can see if they get hit by the waves of the bubble popping and the anti-trust remedies at the same time!
Upvote
20
(20
/
0)
LOL damn AI autospell/correct thinking it knows better than a Human... well sometimes at least.
Upvote
1
(1
/
0)
Scone Muncher
Ars Centurion
I read Ed Zitron's writing and I'm right there with him; I feel like I'm taking crazy pills when I see the dollar numbers being slung around by/at AI outfits. It's "AI Week" at my [very large global] employer. I spent 90 minutes yesterday watching a half-dozen senior directors blather on about how transformative AI will be for us. In those 90 minutes, not a single example of an actual use case.
Upvote
32
(32
/
0)
Upvote
0
(0
/
0)
The money's supposed to eventually come from somewhere! Never mind that there's currently an absolute overload of money going in. Economy-killer application, some day; maybe, they think. Or it'll kill the economy, one way or another. I've got Ed's latest from yesterday on the to-read list.
Upvote
11
(11
/
0)
I would have thought that cloud computing was ideal for smaller companies, for example, those looking to benefit from economies of scale and access elastic compute without owning infrastructure. But when you’re OpenAI, and your compute requirements are so vast that you’re essentially one of the reasons those economies of scale exist in the first place, how do the economics work out?
At OpenAI’s scale, you’re not just a customer, you’re shaping the market. You influence hardware roadmaps, availability, energy consumption patterns, and even where new data centers are built. The traditional cloud value proposition starts to blur when your needs are large enough to drive hyperscale infrastructure decisions.
Is this just about trying to minimise CapEx? At some stage it has to work out cheaper and more flexible at OpenAI’s scale to have their own data centres?
At OpenAI’s scale, you’re not just a customer, you’re shaping the market. You influence hardware roadmaps, availability, energy consumption patterns, and even where new data centers are built. The traditional cloud value proposition starts to blur when your needs are large enough to drive hyperscale infrastructure decisions.
Is this just about trying to minimise CapEx? At some stage it has to work out cheaper and more flexible at OpenAI’s scale to have their own data centres?
Upvote
9
(9
/
0)
It shows Google has presumably a lot of available AI compute sitting around not being used for their own efforts, so in this case Google is also a gold miner selling their shovels to other gold miners, which might lead the market to question how good are Google's gold mining efforts.
Upvote
7
(8
/
-1)
Rule of thumb: anybody who is talking up a new technology and says that “it works great for my use cases”, without saying exactly what any of those use cases actually are, is either a professional spammer or unemployed.
Upvote
12
(12
/
0)
I guess they're finally reached the limits of Redmond's indulgence. If you've already stretched the capacity of one of the world's largest data centre vendors I cannot see how the business will ever be viable.
Upvote
2
(2
/
0)
"Professional spammer" just means MBA, right?
The ones sending out actual mass SPAM probably work harder and are less likely to crash the entire economy when the house of cards collapses.
Upvote
7
(8
/
-1)
Scipio Africanus
Ars Centurion
We forget these days how these very large companies let their divisions operate with degrees of independence.
Like why is tsmc letting intel use its fabs? Intel is about to or is going to let others use their own fabs. Why do car companies manufacture key parts to competitors?
There's many many examples.
Like why is tsmc letting intel use its fabs? Intel is about to or is going to let others use their own fabs. Why do car companies manufacture key parts to competitors?
There's many many examples.
Upvote
2
(2
/
0)
Alternative rule of thumb:
Always skip "____ Week" at your [very large] employer.
My blind guess though is that they do actually have a transformative application in mind-- replace the call center.
Last edited:
Upvote
3
(3
/
0)
The fact that a rivalry is perceived to exist at all is a great argument for why Google should be broken up.
Upvote
0
(0
/
0)
Just to be clear, there was nothing surprising about this, and the reason isn't because Microsoft couldn't supply data centres fast enough. Microsoft has essentially abandoned OpenAI. They've cancelled plans for over 2GW of data centres, as well as pulling out of multiple plans for future developments:
https://www.wheresyoured.at/power-cut/
They've also handed contracts for compute for OpenAI over to other companies (who don't actually have anywhere near enough, or the money to build it), while ending the requirement for OpenAI to exclusively use MS:
https://www.wheresyoured.at/openai-is-a-systemic-risk-to-the-tech-industry-2/
It's often pointed out that despite all the hype, and even if the actual capabilities of AI could meet some of that hype, not a single AI company has slightest clue how they could even dream of actually making any profit from it. What is often missed in that is that this includes the companies providing the computing power in the background as well. Building and running data centres is expensive, and if your main customer is massively unprofitable, building even more data centres solely for them with the promise that they might make money to pay you some time in the future, that might not be a great investment.
Nvidia are the only ones actually selling shovels. MS et al are the ones buying shovels. Sure, they hope to make money by renting them to miners, but so far none of the miners have been able to actually pay. MS bought into OpenAI by providing cheap compute in exchange for part ownership. Now, in exchange for more compute, they want actual money. OpenAI does not have any. So MS have pulled out and left OpenAI to swing in the wind.
What's weird is how little the media, even on the more niche techy side of things, seem to have noticed this. All the talk is about how magic AI will be in the future, or how many billions are being promised here and there. Somehow, the biggest provider of AI compute effectively abandoning the biggest AI company and leaving it with no-one willing or able to provide the compute it demands has gone almost completely unmentioned.
Yeah, see the above links for how well that's going. CoreWeave can't build on that scale and doesn't have the money. SoftBank doesn't have the money and, since OpenAI failed to turn for-profit, no longer has any obligation to pay the majority of what had been promised. Oracle's involvement is largely limited to subcontracting data centre builds to a cryptocurrency scammer that has never built a data centre. There is no $500 billion Stargate project, it's all smoke, mirrors, and people pulling out of commitments before they risk getting called on them.
This "surprise" deal most likely means that Google has some spare compute capacity lying idle, probably as part of the usual cycle of overbuilding and then waiting for demand to catch up. Of course they'll be happy to siphon off some of OpenAI's investor money when given the opportunity. But unless they've suddenly committed to building several GW of data centres in the next couple of years, they won't come close to making up for the capacity that MS has just cancelled. And while MS certainly hasn't got everything right over the years, it's a fair bet they're not the only ones seeing which way the AI winds are blowing and wondering where the money to build all that compute is actually going to come from.
Upvote
9
(9
/
0)
Imagine being the Google sales rep (team) who closed this deal. Legendary. Anyway this continues Google's tradition of profiting from trends even when they aren't winning, by selling solid gold shovels and picks to VC-backed customers. Look how much money they took off of Snap while completely whiffing their own social strategy.
Upvote
0
(0
/
0)
Eh, this is a misunderstanding of the size and shape of Google Cloud. Google can carve out these deals because the Google part of Google is huge, way bigger than you are imagining, and even large cloud customer deals aren't that big by comparison.
Upvote
-1
(0
/
-1)
(Silly post removed, please do better if you want to stay)
Last edited by a moderator:
Upvote
-3
(0
/
-3)
No it isn't. You're massively overestimating how big Google is compared to Microsoft and Amazon. Google can't just magically pull multiple GWs of data centre from their rear, any more than MS can. See the links to Ed Zitron's posts earlier, and follow some of the links there. For comparison, Google claims to be planning around $75 billion capex this year for AI. MS already spent over $200 billion in the last two years (and claim $13b in revenue as a result). Google is yet another company planning to rent (non-existent) compute from CoreWeave, specifically because they don't have anywhere near enough to meet the claimed demand.
This isn't simply a random cloud customer asking to borrow a few of Google's spare servers. The amount of compute OpenAI and others are demanding simply does not exist. Google does not have it. Microsoft does not have it. No has it, and no-one with either the money or capability is actually planning on building it. This deal is not Google being so huge they can just sell OpenAI whatever they ask for without even noticing; this deal is Google letting them have whatever minimal spare capacity they currently have, while giving some vague commitment for some billions in future build, almost certainly with heavy overlap with the existing, highly questionable, CoreWeave and Stargate projects and not much at all in the way of new, Google-funded capex.
Upvote
1
(1
/
0)