Do these 3 things before closing this tab:
1Scan for outdated or missing drivers - takes under a minute2Repair Windows errors before they cause bigger problems3Fix the driver behind crashes, sound loss and screen glitchesAI companies want access to large clusters of NVIDIA H100 GPUs because training and running advanced AI models takes substantial computing capacity. But a company that uses H100s may not own them: it can rent cloud capacity, while cloud providers may own GPUs and lease them to customers. Public estimates expressed as “H100 equivalents” are modeled comparisons of processing power—not verified counts of physical H100 cards.
What the H100 is—and why companies want it
The H100 is NVIDIA’s Hopper-generation data-center GPU, designed for AI workloads including large language models. NVIDIA highlights its Transformer Engine and training and inference capabilities on its H100 product page.
Building and operating AI models requires computing capacity, and a cluster of accelerators can supply more of it than a single device. That capacity can support model training and inference—the work of generating responses after a model has been trained. Companies therefore compete not just to buy individual chips, but to secure enough coordinated compute for their workloads.
The contest is not simply a matter of ordering GPUs. In January 2024, OpenAI CEO Sam Altman told Axios at the World Economic Forum that “none of the pieces are ready” for delivering AI infrastructure “at the scale that people want it.” The comment described a broader infrastructure challenge, not just a shortage of H100s. Axios reported his remarks.
#1 Best Overall
- PLEASE NOTE: Exporting an NVIDIA RTX Pro 6000 GPU outside the US requires strict adherence to the U.S. Export Administration Regulations (EAR) and issuance of an export license from the Bureau of Industry and Security (BIS). Compliance and Know Your Customer (KYC) screening may be required as a condition of order acceptance. [NVIDIA Blackwell Streaming Multiprocessor] The new SM features increased processing throughput, and new neural shaders that integrate neural networks inside of programmable shaders | DLSS 4: Multi Frame Generation ensures ultra-smooth frame pacing for lifelike simulations.
- [Double-Flow-Through Design] The RTX PRO 6000 Blackwell features a double-flow-through cooling design, optimizing efficiency and airflow to sustain peak performance under 600W power loads. | [5th Gen Tensor Cores] Deliver up to 3X the performance of the previous generation and support for FP4 precision for faster AI model processing times with reduced memory usage, enabling local fine-tuning of LLMs and generative AI | [4th Gen Ray Tracing Cores] Double the ray-triangle intersection rate of the previous generation to create photoreal, physically accurate scenes and immersive 3D designs with RTX Mega Geometry, which enables up to 100X more ray-traced triangles.
- [PCIe Gen 5] Support for PCIe Gen 5 provides double the bandwidth of PCIe Gen 4, improving data-transfer speeds from CPU memory and unlocking faster performance for data-intensive tasks like AI, data science, and 3D modeling. | [GDDR7 Memory] With 96 GB of GPU memory and 1.8 TB ps bandwidth, it can tackle massive 3D and AI projects, fine-tune AI models locally, explore large-scale VR environments, and drive larger multi-app workflows.
- [DisplayPort 2.1] Achieve unparalleled visual clarity and performance, driving high resolution displays at up to 8K at 240 Hz and 16K at 60 Hz. Increased bandwidth enables seamless multi-monitor setups while HDR and higher color depth support ensures superior color accuracy for precision work, such as video editing, 3D design, and live broadcasting.
- [Universal MIG] Divide a single RTX PRO 6000 Blackwell into multiple isolated instances, each with dedicated resources, allowing for concurrent execution of multiple workloads, optimized GPU utilization, and secure isolation of different applications or users. [WARRANTY] 3 YR Manufacturer's Warranty. Bulk OEM Packaging. Retail Packaging is NOT included.
Physical H100 counts and “H100 equivalents” are different
A physical H100 count means a number of actual H100 accelerators. An H100-equivalent estimate instead expresses modeled computing capacity in terms of H100 performance. It can represent NVIDIA accelerators other than H100s; it is not a count of H100 cards.
Epoch AI estimates companies’ NVIDIA accelerator holdings using sales and estimated customer allocations. Its figures are ranges from the 25th to 75th percentile, with a median estimate. They are modeled, not audited inventories, and the cited estimates are not explicitly dated as a live snapshot.
| Company or group | Modeled H100-equivalent range | Median estimate |
|---|---|---|
| Google (all of Alphabet) | 270,000–390,000 | 320,000 |
| Microsoft | 540,000–800,000 | 660,000 |
| Meta | 330,000–490,000 | 400,000 |
| Amazon | 240,000–370,000 | 290,000 |
These are Epoch AI’s modeled estimates, not audited physical H100 counts. Its page says some of the compute held by large companies is rented to other users, and that large companies also rent compute themselves. Google’s estimate includes all of Alphabet; Epoch AI estimates Google’s TPU capacity separately. See Epoch AI’s computing-capacity estimates and methodology.
Rank #2
- NVIDIA Ampere Architecture-based CUDA Cores - Double-speed processing for single-precision floating point (FP32) operations and improved power efficiency provide significant performance improvements for graphics and simulation workflows, such as complex 3D computer-aided design (CAD) and computer-aided engineering (CAE), on the desktop.
- Second-Generation RT Cores - With up to 2X the throughput over the previous generation and the ability to concurrently run ray tracing with either shading or denoising capabilities, second-generation RT Cores deliver massive speedups for workloads like photorealistic rendering of movie content, architectural design evaluations, and virtual prototyping of product designs. This technology also speeds up the rendering of ray-traced motion blur for faster results with greater visual accuracy.
- Third-Generation Tensor Cores - New Tensor Float 32 (TF32) precision provides up to 5X the training throughput over the previous generation to accelerate AI and data science model training without requiring any code changes. Hardware support for structural sparsity doubles the throughput for inferencing. Tensor Cores also bring AI to graphics with capabilities like DLSS, AI denoising, and enhanced editing for select applications.
- Third-Generation NVIDIA NVLink - Increased GPU-to-GPU interconnect bandwidth provides a single scalable memory to accelerate graphics and compute workloads and tackle larger datasets.
- 48 Gigabytes (GB) of GPU Memory - Ultra-fast GDDR6 memory, scalable up to 96 GB with NVLink, gives data scientists, engineers, and creative professionals the large memory necessary to work with massive datasets and workloads like data science and simulation.
What is publicly known about Meta’s H100s?
Axios reported on January 23, 2024, that Meta had amassed 340,000 H100 GPUs. That is a dated reported figure, not a current count. It should not be treated as directly comparable to Epoch AI’s modeled H100-equivalent estimate for Meta, which covers estimated NVIDIA accelerator processing power rather than only physical H100s. Axios’s January 2024 report provides the dated example.
The Tool Desk
Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →Companies can use H100s without owning them
Buying accelerators is only one way to secure compute. A company can rent capacity from a cloud provider, which operates the hardware and makes it available through cloud services. As a result, the company running a model may not own the GPUs doing the work.
In March 2024, NVIDIA and Google announced H100-powered A3 virtual machines and DGX Cloud availability through Google Cloud. In NVIDIA’s published customer case study, Runway CTO and co-founder Anastasis Germanidis said: “Using GKE to orchestrate our training jobs enables us to scale to thousands of H100 GPUs in a single fabric to meet our customers’ growing demand.” This is a vendor-published customer statement, not independent performance testing. The announcement establishes a dated example of cloud access; it does not establish current instance availability. NVIDIA’s March 18, 2024 announcement describes the offering.
Rank #3
- 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
Ownership and access answer different questions. A provider’s hardware may serve several customers over time, while a customer’s use of a large cluster does not mean it owns that cluster. Estimates of capacity held by a cloud company also do not necessarily show how much is available to a particular AI lab or model.
Why GPUs are not the only bottleneck
Accelerators need facilities and infrastructure before they can be deployed. NVIDIA’s Form 10-Q for the quarter ended July 26, 2026, identifies land, power, data-center shells and capital as requirements for deployment, and says shortages can delay deployments. Export-control restrictions can also affect shipments of H100 chips and systems. These constraints mean that acquiring GPUs alone does not guarantee that a company can bring new computing capacity online immediately. NVIDIA’s July 2026 filing describes these deployment and shipment risks.
Free tools Windows power users keep installed
One-click scans. No signup required.
What historical wait times and prices can—and cannot—tell you
The European Commission’s 2024 competition policy brief reported that H100 waits were nearly 12 months at the end of 2023, then fell to three to four months. It also cited reported H100 costs of up to $30,000–$40,000 per unit, possibly more. These are historical figures in the Commission’s brief, not current wait-time estimates or a current price quote.
Rank #4
- Standard Memory: 40 GB
- Host Interface: PCI Express 4.0
- Cooler Type: Passive Cooler
- Product Type: Graphics Card
| Measure | Reported figure | Qualification |
|---|---|---|
| Wait time | Nearly 12 months at end-2023; later three to four months | Historical figures reported in the European Commission’s 2024 brief |
| Reported cost per H100 | Up to $30,000–$40,000, possibly more | Historical reported cost cited in the European Commission’s 2024 brief; not a current quote |
The brief’s figures help show the pressure companies faced during that period, but they cannot tell a buyer how long delivery would take or what a GPU costs now. Read the European Commission DG Competition brief.
How to interpret claims about who has the most H100s
- Check whether the number is hardware or modeled capacity. A literal H100 count is not interchangeable with H100-equivalent processing power.
- Check what the estimate covers. A company’s estimated capacity may include accelerators other than H100s; a cloud provider’s capacity may also be rented to customers.
- Look for the date and method. A dated report, a modeled estimate and a current audited inventory are different kinds of evidence.
- Separate capacity from access. Total capacity held by a company does not say how much is allocated to a specific customer or workload.
- Consider deployment constraints. Power, facilities, capital and export rules can affect when purchased hardware becomes usable.
The cited public material does not establish current physical H100 inventories for each company. Comparisons are most useful when they preserve the distinctions above rather than presenting estimates as a definitive league table.
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.
