News Oracle has reportedly placed an order for $40 billion in Nvidia AI GPUs for a new OpenAI data center

[Admin]

Oracle aims to deploy 400,000 Nvidia GB200 AI chips at a Texas data center to support the U.S. Stargate Project.

Oracle has reportedly placed an order for $40 billion in Nvidia AI GPUs for a new OpenAI data center : Read more

 

[A Stoner]

These kinds of sales will keep us gamers as secondary for the foreseeable future. NVidia cannot ignore these customers, which is why the 50 series pretty much only got AI enhancements that were developed while serving the AI buyers needs.

 

[Findecanor]

Computing power is not mentioned in Gigawatts. Its power-draw is.
How will this data centre get its power?

It wasn't long ago that Texas suffered rolling blackouts in its electricity grid.
Grid infrastructure can take decades to prospect and build.

 

[Stomx]

Couple months ago just a week before Deepseek R1 release Larry E was advertising AI trying to convince everyone to join him to spend $500B on it. No one responded?

 

[rluker5]

$40,000,000,000/400,000 = $100,000. Is that the going rate for a GB200 nowadays? When you are buying 400,000 of them? Either Nvidia is printing money, some accountants aren't completely honest or this article doesn't have the full context.

 

[rluker5]

These kinds of sales will keep us gamers as secondary for the foreseeable future. NVidia cannot ignore these customers, which is why the 50 series pretty much only got AI enhancements that were developed while serving the AI buyers needs.

Somebody has figured out the equation;
Build extremely expensive AI focused datacenters + ????? = $profit!
Any day now we will learn what the ????? stands for.

 

[3en88]

Somebody has figured out the equation;
Build extremely expensive AI focused datacenters + ????? = $profit!
Any day now we will learn what the ????? stands for.

Ads.

 

[lmcnabney]

TLDR - gaming GPUs will only get minimal production because Nvidia can make 10-100x the money using those wafers to make AI chips.

So enjoy those currently sky-high GPU prices. They are going to get a lot higher as production slows to a trickle.

 

[SomeoneElse23]

Computing power is not mentioned in Gigawatts. Its power-draw is.
How will this data centre get its power?

It wasn't long ago that Texas suffered rolling blackouts in its electricity grid.
Grid infrastructure can take decades to prospect and build.

"going green" used to be popular.

Now the corporations don't seem to care.

Maybe they never cared, it just sounded good?

 

[stuff and nonesense]

TLDR - gaming GPUs will only get minimal production because Nvidia can make 10-100x the money using those wafers to make AI chips.

So enjoy those currently sky-high GPU prices. They are going to get a lot higher as production slows to a trickle.

Perhaps devs could target AMD levels of performance using DX12 and not using the Nvidia special sauce. Make Nvidia irrelevant?

 

[lmcnabney]

Perhaps devs could target AMD levels of performance using DX12 and not using the Nvidia special sauce. Make Nvidia irrelevant?

It really doesn't make much difference. There simply isn't enough fab capacity on Earth to meet the needs of the market. GPU chips are HUGE. AMD only has so much contracted capacity with TSMC. A 9950X3D needs 70 sq mm and sells for $700. The 9070XT uses 357 sq mm and requires RAM and more hardware and sells for the same - and AMD shares that money with AIB partners.
There really is no reason for either Nvidia or AMD to make ANY GPUs. They are both fabless and every wafer they have contracts for can make them a lot more money being turned into something else.

 

[rluker5]

It really doesn't make much difference. There simply isn't enough fab capacity on Earth to meet the needs of the market. GPU chips are HUGE. AMD only has so much contracted capacity with TSMC. A 9950X3D needs 70 sq mm and sells for $700. The 9070XT uses 357 sq mm and requires RAM and more hardware and sells for the same - and AMD shares that money with AIB partners.
There really is no reason for either Nvidia or AMD to make ANY GPUs. They are both fabless and every wafer they have contracts for can make them a lot more money being turned into something else.

I mostly agree with you, but a 9950x3d needs a 70.6mm2 and a 70.6mm double stacked so 70.6mm2 + 141.2mm2 + 122mm2 = 333.8mm2 not including the extra work of the double stack when compared to the 357mm2 of the 9070xt. I'm pretty sure the 9950x3d silicon costs more to make.

Oddly enough the 9950x at 4,6nm uses more total silicon area per Techpowerup than the "10nm" 13900k and so does the 9800x3d. That and the GTX970 had a bigger die size than the RX9070.

But your point of the limit in total fab capacity decreasing the supply of gaming GPUs when AI GPUs are more in demand, and thus increasing gaming GPU prices is valid.

 

[BFG-9000]

There's going to also be tremendous demand by the space-based U.S. Golden Shower missile defense system, given the aggressive target of initial operation within 3 years.

We've come full circle in that the original computers were created for calculating ballistic trajectory tables, and now AI is needed to target steerable hypersonic missiles capable of evasive maneuvers, using missiles with extremely limited fuel.