Oracle announced Zettascale10 in October 2025 as an AI-compute design targeting up to 800,000 NVIDIA GPUs and up to 16 zettaFLOPS of peak performance. Those are announced targets, not a verified count of GPUs now running or proof of measured performance. As of October 4, 2026, Oracle reports that 42% of Abilene’s total data-center capacity has been delivered; that campus figure does not say how many GPUs are installed or serving workloads.
What is Oracle Zettascale10?
Zettascale10 is Oracle’s announced multi-data-center AI cluster fabric, designed to connect large numbers of NVIDIA GPUs across data-center campuses. Oracle presented it as the compute infrastructure underpinning its Stargate collaboration with OpenAI at Abilene, Texas. The name refers to the scale of the system Oracle says it is designing—not to a published, independently verified measure of a running cluster.
The design combines Oracle Acceleron RoCE networking with NVIDIA AI infrastructure. Oracle says the data centers are intended to sit within a two-kilometer radius of one another, a layout meant to reduce GPU-to-GPU latency during large-scale AI training. Those are descriptions of the intended architecture and benefits; the material available does not independently test its latency, reliability, cost, or efficiency.
How many GPUs and how much compute does Oracle say it will have?
Up to 800,000 GPUs is a target, not a current operating count
Oracle’s October 14, 2025 announcement described a system spanning “hundreds of thousands” of GPUs and set an initial Zettascale10 deployment target of up to 800,000 NVIDIA GPUs. The first phrase describes the broad scale of the planned system; the second is its announced upper target. Neither establishes how many GPUs are installed, connected, or available to customers today.
#1 Best Overall
- [ Maximum AI Compute Power ] Dominate complex workloads with the ASUS ESC8000A-E13. This 4U rack server is a powerhouse engineered for mass-scale AI, machine learning, and deep training. Featuring support for dual AMD EPYC 9005/9004 processors and up to eight dual-slot GPUs, it delivers the raw computational muscle required to train LLMs and run complex simulations effortlessly. Accelerate your data science pipeline and transform raw data into actionable intelligence faster than ever.
- [ Advanced Thermal Efficiency ] High performance demands elite cooling. The ESC8000A-E13 features a cutting-edge aerodynamic design with independent CPU and GPU airflow tunnels. Equipped with redundant hot-swap fans and optimized for liquid cooling integrations, this 4U server ensures maximum uptime under heavy, sustained workloads. Keep your data center running cool, quiet, and highly efficient while preventing thermal throttling during mission-critical enterprise operations.
- [ Scale with Flexible Storage ] Future-proof your infrastructure with unmatched storage and expansion flexibility. This offers comprehensive front-panel drive bays supporting Gen5 NVMe, SAS, or SATA drives alongside multiple PCIe 5.0 slots. Designed as a high-density 4U server capable of housing eight dual-slot GPUs: NVD H200, RTX PRO 6000 Blackwell, RTX PRO 4500 Blackwell or AMD Instinct MI350P PCIe Card, each supporting up to 600 watts.
- [ Enterprise-Grade Reliability ] Minimize downtime and secure your ecosystem with server-grade redundancy. The ESC8000A-E13 is built for 24/7 continuous operation, boasting 2+2 redundant (3200W total) 80 PLUS Titanium power supplies and integrated ASUS ASMB11-iKVM for comprehensive out-of-band management. Ideal for cloud service providers, rendering farms, and large enterprise infrastructure, it combines robust physical hardware with smart remote monitoring to safeguard your digital assets.
- [Reliability Guaranteed] Shop with total peace of mind knowing that every new computer component we sell is backed by our EPC 3-year warranty. Whether you are investing in high-speed DDR5 RAM or a powerhouse GPU, we protect your build against defects and performance failures. We stand firmly behind the quality of our hardware, ensuring that your setup remains fast, stable, and secure for years to come.
16 zettaFLOPS is peak performance
Oracle also announced up to 16 zettaFLOPS of peak performance. A FLOP is a floating-point operation; one zettaFLOP per second is 1021 such operations per second. “Peak” describes a theoretical maximum under specified computing conditions, not the speed a customer should expect on a particular AI model. It is not the same as measured training time, inference throughput, or sustained application performance. The announcement does not provide an independent workload benchmark for Zettascale10.
Is Oracle already operating that many GPUs at Stargate Abilene?
The available figures do not establish that. Oracle’s Abilene data-center page, accessed October 4, 2026, reports that 42% of the campus’s total capacity has been delivered, with the remainder due in subsequent quarters. The page includes site imagery dated May 27 and June 4, 2026, but its delivery statement is not explicitly dated. Most importantly, the percentage measures campus capacity, not GPUs: it cannot be converted into a GPU count or used to show how much compute is online.
Rank #2
- NVIDIA Volta GV100 Architecture — 4,608 CUDA Cores, 640 1st-Gen Tensor Cores delivering 14 TFLOPS FP32 and 112 TFLOPS deep learning performance for AI training, inference, HPC, and scientific computing workloads
- 32GB HBM2 ECC Memory — 900 GB/s Bandwidth — High-bandwidth memory on a 4096-bit bus with ECC error correction provides the memory capacity and throughput required for the largest AI models, simulations, and datasets
- PCIe 3.0 x16 Interface — 250W TDP — Standard PCIe Gen3 connectivity with passive cooling designed for enterprise rack server deployment in HPE ProLiant, Dell PowerEdge, and Supermicro platforms with adequate chassis airflow
- NVLink — Scale to 96GB Unified Memory — Connect two V100 GPUs via NVLink at 300 GB/s bi-directional bandwidth to scale GPU memory from 32GB to 96GB for larger AI training and HPC workloads
- Multi-Precision Computing — Supports FP64 (7 TFLOPS), FP32 (14 TFLOPS), FP16 (112 TFLOPS) and INT8 precision modes for flexible deployment across training, inference, and scientific simulation workloads
A separate Associated Press report from October 2025 described Abilene as an eight-building project. At the time, one building was operating and a second was nearly complete; the report said the complex would require about 900 megawatts to power all eight buildings. That is a contemporaneous account of the project’s planned power scale and phased construction, not a statement of current energy use or building status.
Availability was a forecast in the announcement
Oracle said it was taking orders for Zettascale10 when it announced the system and expected availability in the second half of the following calendar year—H2 2026. The sources available here do not confirm that the full 800,000-GPU configuration became generally available on that schedule. An announced order window or expected availability date should not be read as confirmation that the maximum build is deployed.
Rank #3
- AI-Optimized: Designed to support up to 4 GPUs, it is perfect for handling intensive AI and machine learning tasks, ensuring high performance and scalability for advanced computational needs.
- Intelligent Storage: Equipped with 8 hot-swappable 3.5" SATA/SAS drives (12Gbps), featuring SGPIO and temperature control, it ensures efficient data management and reliable storage performance.
- Robust Cooling: The system includes 3x 12038 hot-swap PWM fans and 2x 8038 rear fans, providing advanced thermal management to maintain optimal temperatures and ensure stable operation under heavy workloads.
- Rack-Ready: Comes with a pre-installed rail kit, allowing for quick and easy installation in standard 19-inch server racks, making it ideal for data center environments and enterprise setups.
- Versatile Connectivity: Offers USB 3.0 and the latest USB 3.2 Type-C ports, ensuring high-speed data transfer and compatibility with a wide range of peripherals and devices for enhanced connectivity options.
How does Zettascale10 differ from Oracle’s earlier supercomputer figures?
Oracle’s March 18, 2025 engineering account covered an earlier OCI Supercluster configuration. Its figures are useful for chronology, but they should not be combined with Zettascale10’s later headline as if they describe one deployed system.
| Oracle announcement | GPU scale | Peak compute | Network figure |
|---|---|---|---|
| OCI Supercluster, March 18, 2025 | Up to 131,072 NVIDIA Blackwell GPUs | Up to 2.4 zettaFLOPS | 52 Pbps aggregate line-rate throughput across the 131,072-GPU configuration, at 400 Gbps per port |
| Zettascale10, October 14, 2025 | Initial target of up to 800,000 NVIDIA GPUs | Up to 16 zettaFLOPS | No comparable aggregate network-throughput figure stated in the cited announcement |
Oracle’s 2024 CloudWorld post also listed Supercluster configurations using H200, B200, and GB200 hardware at different scales and availability stages, including up to 131,072 B200 GPUs and more than 100,000 GB200 GPUs. Those are older announcements, not evidence of current installed capacity. The key comparison is between announced configurations and their stated targets—not a confirmed count of operating GPUs.
Rank #4
- Professional AI & Creator Workstation: AMD Radeon AI PRO R9700 GPU with 32GB GDDR6 is engineered for AI development, professional content creation, and compute-intensive workloads.
- Massive 32GB Memory Capacity: 32GB of GDDR6 memory on a 256-bit bus provides ample bandwidth for large AI models, 8K video editing, and complex 3D rendering.
- Advanced RDNA 4 with AI Accelerators: 64 Compute Units with 3rd Gen Ray Tracing and dedicated 2nd Gen AI Accelerators for groundbreaking AI performance and visual computing.
- Professional Blower Cooling: Efficient single blower design exhausts heat directly out of the chassis, ideal for multi-GPU workstation and server configurations.
- Enterprise-Grade Thermal Solution: Vapor chamber heatsink with industrial Honeywell PTM7950 thermal interface material ensures reliable cooling under sustained professional loads.
Why does the network matter at this scale?
Connecting a very large GPU cluster is not just a matter of adding processors: GPUs must exchange data, and network congestion or long communication paths can constrain distributed training. Oracle’s Zettascale10 announcement describes an Acceleron RoCE fabric that uses switching capabilities built into modern GPU network interface cards, multiple isolated network planes, and traffic shifting if a plane has problems. These are Oracle’s descriptions of its design; the announcement does not quantify realized uptime or savings.
For its earlier Supercluster, Oracle described a three-tier Clos network, 400 Gbps connections per GPU, placement based on network locality, and congestion-control methods for RoCE. Oracle also reported latency figures for network tiers serving different GPU group sizes. Those details explain the engineering challenges the company says it addressed in that earlier configuration; they are not independent benchmarks of Zettascale10.
Outdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware matchPC Slower Than It Used to Be?
A free scan shows the junk files, broken settings and background clutter dragging Windows down - then fixes them in one click.Free scan · Windows 10 & 11Best Value
What do the figures mean for infrastructure investment?
The scale claims point to a project dependent on coordinated delivery of compute, networking, and data-center capacity. The Abilene capacity update and the AP’s 2025 power estimate help frame the physical buildout, but neither reveals the full project cost, current electricity draw, operating GPU count, or revenue attributable to Zettascale10. A peak-FLOPS headline alone cannot answer those financial questions.
For anyone assessing Oracle’s infrastructure plans, keep four measures separate: announced GPU target, campus capacity delivered, hardware actually operating, and performance measured on a workload. The sources cited here provide an announced target and a campus-level progress update, but not a current operating-GPU count or an independent Zettascale10 benchmark. They are not enough on their own to determine whether the investment will earn a particular return.
What is not part of the Zettascale10 announcement?
BlueField-4 timing
Oracle’s October 28, 2025 post described NVIDIA BlueField-4 DPUs as part of its next-generation platform, while saying BlueField-3 DPUs were already in production and BlueField-4 was “soon to be integrated.” The post included a future-product disclaimer. It therefore does not establish that BlueField-4 or every capability described was generally available or deployed in Zettascale10 at that time.
Related NVIDIA software and analytics
NVIDIA’s Oracle AI World post describes related Oracle integrations with NVIDIA AI Enterprise, NIM, and GPU-accelerated analytics, and characterizes Zettascale10 as designed for AI training and inference. The existence of those integrations does not show that each product is included in every Zettascale10 deployment.
Free tools Windows power users keep installed
One-click scans. No signup required.
The Argonne supercomputer project
NVIDIA separately announced a collaboration with Oracle and the U.S. Department of Energy to build AI supercomputers for scientific discovery at Argonne. That is a distinct project from Stargate Abilene and Zettascale10; its specifications should not be attributed to the Abilene system.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
