• Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
    Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.

NVIDIA to Deliver 2 Million Additional GPUs through 2028

erek

erek

8=D
2FA
Joined
Dec 19, 2005
Messages
17,855
"To meet this surging demand from frontier labs, global enterprises, startups and governments, AWS and NVIDIA are building on 16 years of joint innovation to expand AI compute capacity and bring new co-engineered solutions to customers faster. As part of the expanded collaboration, the companies are working to:
  • Deploy 2 million additional NVIDIA GPUs across AWS's global infrastructure in 2027-2028
  • Bring NVIDIA Vera CPU-based infrastructure to AWS
  • Extend NVIDIA NVLink Fusion with custom NVIDIA high-bandwidth memory (NVHBM)
  • Build AI factories for the U.S. government, including 100,000 GPUs on secure AWS infrastructure for running federal and national-security workloads
  • Integrate the NVIDIA platform with the AWS Nitro System and Elastic Fabric Adapter (EFA) for enhanced security and reliability
  • Continue to support NVIDIA Nemotron open models on Amazon Bedrock and Amazon SageMaker, giving customers more open model choice
  • Accelerate data processing and vector indexing on Amazon EMR and Amazon OpenSearch with NVIDIA cuDF and cuVS CUDA-X libraries for faster, more cost-efficient analytics and AI applications
  • Further advance robotics workloads through Amazon Robotics' adoption of NVIDIA's physical AI platform, speeding innovation in warehouse automation and next-generation robots
"Customers want the freedom to choose the best tools for their AI workloads, and they want confidence that everything works seamlessly together," said Matt Garman, CEO of AWS. "That's why we've invested deeply with NVIDIA to make AWS the best place to run NVIDIA AI technologies, optimizing performance across our infrastructure from networking and security to deployment. This expanded collaboration gives frontier labs, enterprises and governments even more ways to build and deploy AI on AWS."
Click to expand...
"NVIDIA and AWS have built one of the great growth engines of the AI era, and demand is running ahead of every forecast," said Jensen Huang, founder and CEO of NVIDIA. "For 16 years, we have scaled NVIDIA computing in the cloud together. Now, we are expanding our partnership across the full stack—GPUs, CPUs, networking, open models and software—to make agentic and physical AI real at an unprecedented pace and scale that only AWS and NVIDIA can deliver. This expansion reflects customers' demand for NVIDIA's platform on AWS."
Click to expand...
Massive Expansion of AI Compute Capacity
AWS offers the widest range of GPU-based instances of any cloud provider to power a diverse set of AI and machine learning workloads. At NVIDIA GTC 2026, AWS announced plans to add more than 1 million NVIDIA GPUs starting in 2026. Since then, demand has exceeded those expectations. AWS plans to deploy an additional 2 million NVIDIA Blackwell Ultra, Rubin and Rubin Ultra GPUs in 2027-2028 across AWS Global Infrastructure, including AI factories. This additional capacity will help power customer workloads ranging from agentic AI and scientific discovery to enterprise automation and physical AI. In addition, AWS will expand NVIDIA Blackwell capacity, including NVIDIA RTX PRO 4500 Blackwell Server Edition GPUs for Amazon EC2 G7 instances. G7 instances deliver 4.6x AI inference performance and 2.1x graphics performance compared to previous-generation G6 instances. AWS is the first major cloud provider to offer compute instances accelerated by RTX PRO 4500. AWS and NVIDIA are also collaborating on NVIDIA Spectrum networking to further optimize network performance for large-scale AI training workloads across GPU clusters.

Support for NVIDIA Vera CPUs on AWS
AWS and NVIDIA are working to bring Vera CPU-based infrastructure to AWS, providing an additional option to support agentic AI workloads that require high-performance CPU compute alongside accelerated infrastructure. Purpose-built for the next generation of AI, Vera complements AWS's strategy to offer the broadest choice of compute—from AWS custom silicon to the latest accelerators and CPUs from partners.

Heterogeneous AI Infrastructure Using NVIDIA NVLink Fusion With NVHBM
At re:Invent 2025, AWS announced support for NVIDIA NVLink Fusion high-speed chip interconnect technology in next-generation Trainium chips. NVIDIA and Amazon's Annapurna Labs are expanding that support to work on NVIDIA's new custom high-bandwidth memory (NVHBM) technology, in partnership with memory suppliers, which would give Trainium access to faster, more power-efficient memory. Combined with NVLink Fusion, Annapurna Labs can now tap NVIDIA's custom memory technology and scale-up architecture to enhance performance and efficiency for AI workloads while seamlessly integrating Trainium and GPUs within a common rack-scale architecture.

Powering Federal AI at the Highest Levels of Security
Government agencies need secure AI infrastructure to keep pace with the demands of national security. AWS and NVIDIA plan to build AI factories for the U.S. government, delivering NVIDIA's AI stack, including plans to deliver 100,000 GPUs on AWS's secure infrastructure for federal and national-security workloads. This collaboration puts AWS and NVIDIA at the center of federal AI advancement for national security, enabling government agencies to deploy AI at scale for workloads classified at Impact Level 6 (IL6) and above.

These new commitments build on a foundation of deep technical integrations between AWS and NVIDIA that are already delivering results for customers today, including:
  • Enhanced security and reliability with AWS Nitro System and EFA—Across this expanded collaboration, all NVIDIA GPU-based and Trainium-based EC2 instances—including those leveraging NVLink Fusion are built on the AWS Nitro System and interconnected through EFA. Both GPU-accelerated and Trainium-based EC2 instances will continue to be built on the Nitro System and scaled out through EFA. Together, Nitro and EFA help ensure that as AWS expands its NVIDIA GPU fleet and integrates new interconnect technologies, customers retain the security, reliability and network performance they depend on for production AI workloads at scale.
  • NVIDIA Nemotron models on AWS—As part of AWS's commitment to offering customers the broadest choice of AI models, NVIDIA's Nemotron family of open models is available on Amazon Bedrock as fully managed, serverless models and on Amazon SageMaker for customers who want to deploy and fine-tune on their own infrastructure. This integration gives customers access to NVIDIA's latest open models with the security, scalability and operational tooling of AWS.
  • GPU-accelerated data processing and vector indexing—As data volumes grow, workloads such as feature engineering, large-scale ETL and real-time analytics require increasingly faster processing. AWS and NVIDIA are collaborating to deliver GPU-accelerated data processing on Amazon EMR using Amazon EC2 G7 instances and the NVIDIA cuDF library, delivering up to 3.7x faster processing speeds and a 30% better price performance compared to CPU-based configurations. Separately, as AI applications, retrieval-augmented generation pipelines and semantic search push vector databases to billions of records, index building and tuning becomes a bottleneck. GPU-accelerated vector indexing on Amazon OpenSearch Service offloads index construction onto dedicated GPUs, delivering up to 9x faster vector indexing at a quarter of the cost—available across both managed clusters and Amazon OpenSearch Serverless.
  • Physical AI for robotics—Amazon Robotics is collaborating with NVIDIA to accelerate the development of next-generation robots integrating NVIDIA's full-stack physical AI platform, including the NVIDIA Jetson platform, NVIDIA Omniverse libraries and the NVIDIA Isaac open robotics development platform. The collaboration spans simulation, synthetic data generation, robot training, route optimization, functional safety and real-to-sim validation—all running on GPU-accelerated Amazon EC2 instances. Together, AWS and NVIDIA are helping advance the capabilities that robotics workloads require at scale: massive simulation, diverse training data and continuous real-world validation."
Source: https://www.techpowerup.com/352009/aws-and-nvidia-to-deliver-2-million-additional-gpus
 
Dan_D said:
Notice how none of that will put new GPU's in the hands of gamers. NVIDIA stopped giving a shit about that market some time ago.
Click to expand...
To be fair, nobody is increasing their consumer component supply except Apple…. Because it’s all they have.
 
Lakados said:
To be fair, nobody is increasing their consumer component supply except Apple…. Because it’s all they have.
Click to expand...
I felt like intel is still quite about consumer like component has well ( imagine a lot is commercial enterprise client in laptop fleet, just that they use very similar intel product than household consumer0, less than half their revenues are from data center +AI still in 2026 I think and margin higher on the CCPG than DCAI side of things
 
erek said:
2 million over multiple years doesn’t seem like a lot of GPUs especially at scale
Click to expand...

In addition to 1 million+ already, to one buyer, over a 2 year period.......
1787800852430.png
 
Not surprised at all. Publicly owned companies chase the dollar. They always have. Niche markets are rarely served unless the ROI is there. Same as it has always been.

Until more fabs spin up the consumer space is going to pay the price in lack of availability and high prices.
 
erek said:
2 million over multiple years doesn’t seem like a lot of GPUs especially at scale
Click to expand...
Truth thought the same thing, the title is a bit misleading.....

But you need to remember how small of a part GPU are of those system now, going by "standard" NVL72 system,, 2 millions GPU would mean 25k/30k rack type of order, they are $5-$7 millions, 3000 pounds computers with a lot of stuff in them, that a 140-195 billions type order or around 3 times Intel annual total revenues in a single PO.... (or 4-5 times AMD)

The scale are just out of this world, and why that the pressure at the compute fabs level is not that big, all that giant level of expense-infracstructure needed per single 850mm of GPU is so big, that you do not spend that much TSMC 3N capacity on that part. Networking-cooper wiring of all sorts-cooling-advanced packaging, memory, storage is where those datacenter/rack put the pressure on.

That one way to look at it, single client, single order, is many more times than the annual total business of giants competitors, yet 2 millions 850mm GPU at bad yield, that will be using about 2-3 weeks of TSMC 3N capacity.... over an 100 weeks period... in a world with TSMC 2 being the latest and 5/7 family of nodes still going strong.
 
Last edited:
philb2 said:
So maybe we should return the favor, by buying AMD or even Intel GPUs.
Click to expand...
Will see with the PS6, but AMD gaming revenues were only 780 millions last quarter, it was 42% higher a year ago during the same quarter, that down to around 6.7% of their business and low margin. I am not sure how much more they care or have time to do so
 
They both are companies, not charities.

To think they'd leave (massive amounts of) money on the table, just to specifically cater to gamers only, for less money, is a bit delusional.

They all still make consumer/gaming products though, so it's not like 'gamers are abandoned' as some try to paint it.

Are gamers less important and less affluent, yes. Abandoned? No.
 
Lakados said:
AMD is putting even more of their resources into AI than Nvidia is.

We’re just not on any of their radars anymore.
Click to expand...
AMD doesn't have the resources that NV has. Their effort is a losing effort from that standpoint. They are second fiddle.
 
philb2 said:
So maybe we should return the favor, by buying AMD or even Intel GPUs.
Click to expand...
Yeah no. AMD doesn't have anything remotely close to a 5090. Not to mention, they are still behind on features and performance in ray tracing. AMD isn't the hero or champion of the people some seem to think they are. It would be doing exactly what NVIDIA's doing right now if it had the capability. Right now AMD is mid-range only and Intel's and even bigger disappointment in that arena.

That's part of the problem. NVIDIA has no real competition. It hasn't for years now.
 
polonyc2 said:
they are in the CPU space
Click to expand...
No, they aren't. They competitively price Opterons in the server space because of Intel's marketshare. It remains completely unchallenged in the HEDT workspace and the entry into that is $1,899 for a CPU. Let's also not forget that when Intel had $1,000 Extreme Editions, AMD had $1,000 FX edition CPU's. Why? Because AMD could get away with it.

AMD is only a little better behaved than Intel because it's position in the market has historically been different.
 
frustrated Maxim Boldson of Game GPU reports:

Gamescom 2026 has finally confirmed the cancellation of the GeForce RTX 50 Super.​

Maxim Boldson HARDWARE August 27 2026

The recent Gamescom 2026 exhibition finally confirmed the cancellation of the mid-term refresh of the GeForce RTX 50 Super series of graphics cards. NVIDIA has completely abandoned the production of updated Blackwell-generation graphics accelerators for the consumer market. The main reasons for this decision were a shortage of high-speed GDDR7 memory, overloaded production capacity, and a shift in the chipmaker's key priorities toward server chips for artificial intelligence systems.

The full release of consumer RTX 60 series models is scheduled for late 2027 or early 2028.

https://en.gamegpu.com/news/zhelezo...telno-podtverdila-otmenu-geforce-rtx-50-super
 
Dan_D said:
No, they aren't. They competitively price Opterons in the server space because of Intel's marketshare. It remains completely unchallenged in the HEDT workspace and the entry into that is $1,899 for a CPU. Let's also not forget that when Intel had $1,000 Extreme Editions, AMD had $1,000 FX edition CPU's. Why? Because AMD could get away with it.

AMD is only a little better behaved than Intel because it's position in the market has historically been different.
Click to expand...

yet you ignore the last few generations of Ryzen CPU's
 
Dan_D said:
No, they aren't. They competitively price Opterons in the server space because of Intel's marketshare. It remains completely unchallenged in the HEDT workspace and the entry into that is $1,899 for a CPU. Let's also not forget that when Intel had $1,000 Extreme Editions, AMD had $1,000 FX edition CPU's. Why? Because AMD could get away with it.

AMD is only a little better behaved than Intel because it's position in the market has historically been different.
Click to expand...

Don't forget Zen 6 is server first, consumer second, same shit people have dogged on Nvidia for
 
polonyc2 said:
yet you ignore the last few generations of Ryzen CPU's
Click to expand...
No, I'm not. AMD making a product to increase its market share is a business decision. They created those to make money on the consumer market. They saw an opportunity to make some money, so they did. It's not charity. They aren't a champion of the people simply because they created something that people wanted to buy.
kram182 said:
Don't forget Zen 6 is server first, consumer second, same shit people have dogged on Nvidia for
Click to expand...
I don't have a problem with that in the slightest. Taking a product and adapting it to different markets only makes sense from a business standpoint. The problem we have now is that these companies are only serving the commercial market and almost completely ignoring us now. The drip feed we get of GPU's are now so coveted that almost no one can afford them anymore. It's the same thing with RAM, SSD's, etc.
 
Last edited:
Dan_D said:
I don't have a problem with that in the slightest. Taking a product and adapting it to different markets only makes sense from a business standpoint. The problem we have now is that these companies are only serving the commercial market and almost completely ignoring us now. The drip feed we get of GPU's are now so coveted that almost no one can afford them anymore. It's the same thing with RAM, SSD's, etc.
Click to expand...

Neither do I have a problem with cut down designs/yields/etc, even for Zen 6 - just saying, if we're comparing 'our heroes'/'apples' to 'our heroes'/'apples'....

The entire 'cut down'/'bad yield' concept is why consumer/gamer will never go away IMO, despite all the people that 'doom porn' that scenario saying 'they' want us all on cloud etc.....
 
You must log in or register to reply here.
Back
Top