# AI Server List > AI servers ship as 8-GPU nodes 3U to 10U tall, or as 72-GPU NVL72 racks. This table lists 68 models from 31 makers: 49 carry 8 GPUs, 13 are rack-scale systems and 31 offer liquid cooling. NVL72 racks are rated at 120 to 140 kW; an 8-GPU HGX B300 node peaks at 15 kW. Every figure links to the maker’s own page. Every AI server model: GPUs, chassis, cooling, rack kW, straight off the datasheet. 68 servers, each field linked to its public source, last verified 2026-11-11. When citing, link the canonical HTML page. ## Data - [AI servers by OEM and ODM](https://aiserverlist.com/): the full comparison table - [Data (JSON)](https://aiserverlist.com/data.json): every record with sources - [Data (CSV)](https://aiserverlist.com/data.csv): one row per server - [Methodology](https://aiserverlist.com/methodology/): what each field means and how it is verified ## Servers - [AMAX AceleMax AXG-428ILC2](https://aiserverlist.com/amax-acelemax-axg-428ilc2/): A liquid-cooled HGX B200 server for AI training, inference, and HPC deployments. CPU and GPU cold plates include leak detection. - [ASRock Rack 4U8X-GNR2/DLC SYN B200](https://aiserverlist.com/asrock-rack-4u8x-gnr2-dlc-b200/): A compact HGX B200 platform for AI infrastructure buyers and system integrators. PCIe switch synthetic mode supports GPU, network, and storage data movement. - [ASUS XA GB721-E2](https://aiserverlist.com/asus-xa-gb721-e2/): ASUS's liquid-cooled GB300 NVL72 rack-scale solution in a 48RU NVIDIA MGX rack, part of the ASUS AI POD line. For AI cloud and data center buyers who want ASUS rack deployment tooling. - [ASUS XA NB3I-E12](https://aiserverlist.com/asus-xa-nb3i-e12/): An HGX B300 server for demanding AI workloads. Eight embedded ConnectX-8 interfaces support GPU networking at up to 800G per SXM. - [Bull (formerly Eviden) BullSequana AI 641-A02/H02](https://aiserverlist.com/bull-bullsequana-ai-641/): An eight-GPU HGX H200 platform for large-model training and inference. Bull offers air or hybrid cooling configurations in this family. - [Cisco UCS C880A M8](https://aiserverlist.com/cisco-ucs-c880a-m8/): An air-cooled B300 platform for enterprise and service-provider AI training, fine-tuning, and inference. It integrates one ConnectX-8 adapter per GPU for cluster expansion. - [Colfax CX7850q-XK9](https://aiserverlist.com/colfax-cx7850q-xk9/): A 7U HGX H200 server for AI and HPC teams, built to order. It carries eight NVLink-connected H200 GPUs, eighteen front NVMe bays and 3+3 redundant 4000 W power supplies. - [Compal SG720-2A](https://aiserverlist.com/compal-sg720-2a/): An eight-GPU AMD Instinct MI350 Series server from Compal, offered with air, single-phase or two-phase liquid cooling, with an ORv3 21-inch version (OG720-2A). - [Dell PowerEdge XE9680](https://aiserverlist.com/dell-poweredge-xe9680/): An eight-accelerator platform for enterprise AI training, inference, and HPC. This record captures the H200 GPU option and the server’s air-cooled rack format. - [Dell PowerEdge XE9712](https://aiserverlist.com/dell-poweredge-xe9712/): Dell's 1RU Grace Blackwell Ultra compute sled for GB300 NVL72, shipped in the 48RU IR9048 rack with 33 kW power shelves. Aimed at large-scale LLM training and real-time inference buyers who want a Dell-integrated rack. - [Dell PowerEdge XE9780](https://aiserverlist.com/dell-poweredge-xe9780/): An air-cooled Intel Xeon 6 server for large-model training and inference. Dell lists both B200 and B300 accelerator configurations. - [Dell PowerEdge XE9780L](https://aiserverlist.com/dell-poweredge-xe9780l/): Dell's liquid-cooled 3 OU node with eight HGX B300 GPUs for ORv3 racks and a 33 kW power shelf. Built for AI factories that run IR7000 rack-scale deployments. - [Dell PowerEdge XE9785](https://aiserverlist.com/dell-poweredge-xe9785/): Dell's air-cooled 10U eight-GPU server on dual 5th Gen EPYC, with AMD Instinct MI355X accelerators. For AI training and inference without a liquid-cooling build-out. - [Dell PowerEdge XE9785L](https://aiserverlist.com/dell-poweredge-xe9785l/): A liquid-cooled AMD EPYC rack sled for dense enterprise and service-provider AI deployments. The server supports AMD MI355X and NVIDIA B300 accelerator configurations. - [Exxact TensorEX TS4-128275176](https://aiserverlist.com/exxact-tensorex-ts4-128275176/): A configurable HGX B200 system for AI development, research, and simulation. Exxact offers deployment with its machine-learning software environments. - [Fsas Technologies (Fujitsu) PRIMERGY GX2570 M8s](https://aiserverlist.com/fsas-primergy-gx2570-m8s/): A B200 GPU server for AI and high-performance computing buyers. Fsas Technologies offers both air-cooled and water-cooled chassis configurations. - [GIGABYTE G4L3-SD1-LAX5](https://aiserverlist.com/gigabyte-g4l3-sd1-lax5/): 4U dual-Xeon server with a direct-liquid-cooled NVIDIA HGX B200 8-GPU baseboard. Built for high-density B200 racks that run on facility liquid cooling. - [GIGABYTE G4L4-SD3-LAX7](https://aiserverlist.com/gigabyte-g4l4-sd3-lax7/): Gigabyte's 4U liquid-cooled HGX B300 server with an integrated-manifold DLC design and leak detection. Built for dense liquid-cooled B300 racks. - [GIGABYTE G893-SD1-AAX5](https://aiserverlist.com/gigabyte-g893-sd1-aax5/): An Intel-based HGX B200 server for AI training, inference, and HPC. It combines NVLink GPU connectivity with PCIe Gen5 expansion. - [GIGABYTE G893-SG1-AAX1](https://aiserverlist.com/gigabyte-g893-sg1-aax1/): 8U dual-Xeon server carrying eight Intel Gaudi 3 HL-325L OAM cards on an HLB-325 baseboard. Aimed at enterprise Gaudi 3 AI training and inference clusters. - [GIGABYTE G894-AD3-AAX7](https://aiserverlist.com/gigabyte-g894-ad3-aax7/): Gigabyte's 8U HGX B300 server with dual Xeon 6900 and 24 DIMM slots. Built for memory-heavy AI and HPC clusters. - [GIGABYTE G894-SD3-AAX7](https://aiserverlist.com/gigabyte-g894-sd3-aax7/): Gigabyte's 8U HGX B300 server with dual Xeon 6700/6500 and 32 DIMM slots. Built for AI training and inference clusters. - [GIGABYTE G894-ZD3-AAX7](https://aiserverlist.com/gigabyte-g894-zd3-aax7/): Gigabyte's 8U HGX B300 server with dual AMD EPYC 9005/9004 and 24 DIMM slots. Built for AI clusters that standardize on EPYC hosts. - [GIGABYTE GIGAPOD NVIDIA GB300 NVL72](https://aiserverlist.com/gigabyte-gigapod-gb300-nvl72/): GIGABYTE's direct-liquid-cooled GB300 NVL72 rack built from 18 XN15-CB0-L01 compute trays and 9 NVLink switch trays, also sold as 16+8 rack GIGAPOD scalable units. For AI factory builders who scale from one rack to a full pod. - [H3C UniServer R5500 G7 X](https://aiserverlist.com/h3c-uniserver-r5500-g7-x/): A modular eight-GPU server for distributed model training and enterprise AI workloads. GPU and CPU modules support separate power planes and servicing. - [HPE Compute XD690](https://aiserverlist.com/hpe-compute-xd690/): HPE's 10U air-cooled server with eight Blackwell Ultra B300 GPUs and two Xeon 6 processors. Built for large AI training, tuning and inference clusters. - [HPE ProLiant Compute XD685](https://aiserverlist.com/hpe-proliant-compute-xd685/): An AMD EPYC platform for AI service providers, large enterprises, and sovereign AI deployments. Chassis height depends on the air or direct liquid cooling configuration. - [Huawei Atlas 800T A2](https://aiserverlist.com/huawei-atlas-800t-a2/): An AI training server for government, finance, education, research, and telecom deployments. It uses Kunpeng CPUs and provides eight 200GE network interfaces. - [Hypertec CIARA TITAN GS685R-G6 B200](https://aiserverlist.com/hypertec-ciara-titan-gs685r-g6-b200/): An AMD EPYC HGX B200 platform for enterprise-scale model training, inference, and scientific computing. It uses passive GPU air cooling and redundant power. - [IEIT SYSTEMS (Inspur) NF5688G7 AMD](https://aiserverlist.com/ieit-inspur-nf5688g7-amd/): A hyperscale AI training server for on-premises and cloud deployments. The AMD configuration combines eight Hopper accelerators with EPYC 9004 processors. - [Ingrasys (Foxconn) NVIDIA GB300 NVL72](https://aiserverlist.com/ingrasys-nvidia-gb300-nvl72/): A liquid-cooled rack-scale system for AI reasoning, inference, and training. Ingrasys integrates Grace CPUs, Blackwell Ultra GPUs, and NVLink networking. - [Inventec Artemis II Rack (GB300 NVL72)](https://aiserverlist.com/inventec-artemis-ii-gb300-nvl72/): Inventec's liquid-cooled rack for NVIDIA GB300 NVL72 building blocks with sidecar or row-based cooling and leak detection. For hyperscalers and cloud builders sourcing ODM racks. - [Inventec P9000IG6(AC)](https://aiserverlist.com/inventec-p9000ig6-ac/): An air-cooled HGX B200 platform from Inventec’s server business. It targets AI training, scientific computing, and other data-intensive enterprise workloads. - [Lenovo NVIDIA GB300 NVL72](https://aiserverlist.com/lenovo-nvidia-gb300-nvl72/): Lenovo's 48U MGX rack with 18 compute trays and 9 NVLink switch trays, cooled by warm-water hybrid liquid and air. For enterprises and cloud providers that want Lenovo services and warm-water efficiency. - [Lenovo ThinkSystem SC777 V4 Neptune](https://aiserverlist.com/lenovo-thinksystem-sc777-v4/): Fanless, fully warm-water-cooled GB200 NVL4 node with four Blackwell GPUs and two Grace CPUs, mounted eight to a 13U N1380 enclosure. For HPC and AI training sites with facility water. - [Lenovo ThinkSystem SR680a V3 with B200](https://aiserverlist.com/lenovo-thinksystem-sr680a-v3-b200/): An Intel-based B200 server for data-center AI training, simulation, and scientific research. The system fits air-cooled facilities. - [Lenovo ThinkSystem SR680a V4](https://aiserverlist.com/lenovo-thinksystem-sr680a-v4/): Lenovo's 8U air-cooled server with eight HGX B300 GPUs and two Xeon 6700 processors. Built for enterprise AI training and inference. - [Lenovo ThinkSystem SR780a V3](https://aiserverlist.com/lenovo-thinksystem-sr780a-v3/): A liquid-cooled Intel server for AI training, simulation, and scientific workloads. It supports H100, H200, and B200 GPU configurations. - [MiTAC Olympus G8825Z5U2BC-350X-755](https://aiserverlist.com/mitac-olympus-g8825z5u2bc/): An all-AMD server for large language models, AI training and inference, and HPC. The named configuration includes EPYC 9755 processors and MI350X accelerators. - [MSI CG480-S5063](https://aiserverlist.com/msi-cg480-s5063/): A 4U NVIDIA MGX platform for enterprise AI training, inference, simulation, and HPC. It supports up to eight dual-width PCIe GPUs. - [Netweb Tyrone Camarero PDA200A2HG-812](https://aiserverlist.com/netweb-tyrone-camarero-pda200a2hg-812/): An AMD EPYC HGX B200 server for AI and high-performance computing buyers. The platform includes NVSwitch and twelve hot-swap NVMe bays. - [NVIDIA DGX B200](https://aiserverlist.com/nvidia-dgx-b200/): An integrated B200 AI server for organizations developing and running large models. NVIDIA supplies the compute, networking, and system software platform. - [NVIDIA DGX B300](https://aiserverlist.com/nvidia-dgx-b300/): NVIDIA's turnkey 10U system with eight Blackwell Ultra GPUs, 2.3 TB of GPU memory and eight 800 Gb/s ConnectX-8 ports. Built for enterprises and AI factories that want a ready-to-run training and inference node. - [NVIDIA DGX GB200 NVL72](https://aiserverlist.com/nvidia-dgx-gb200-nvl72/): A complete NVLink-connected rack for large-scale AI training and inference. Liquid cold plates cool CPUs and GPUs while fans cool networking and storage components. - [NVIDIA DGX GB300](https://aiserverlist.com/nvidia-dgx-gb300/): A liquid-cooled, rack-scale DGX that links 72 Blackwell Ultra GPUs and 36 Grace CPUs in one NVLink domain. Built for enterprises and AI factories that want NVIDIA's own turnkey reasoning and training system. - [NVIDIA GB300 NVL72 by HPE](https://aiserverlist.com/hpe-nvidia-gb300-nvl72/): HPE's direct-liquid-cooled GB300 NVL72 rack with 72 Blackwell Ultra GPUs and 36 Grace CPUs, with in-row and in-rack CDU options. For hyperscale and sovereign AI builders who want HPE integration and services. - [PEGATRON AS205-2T1](https://aiserverlist.com/pegatron-as205-2t1/): A compact dual-socket NVIDIA MGX server for agentic AI, scientific computing, rendering, and video workloads. Its GPU and storage configurations vary with accelerator choice. - [Pegatron AS404-2A1-8I2](https://aiserverlist.com/pegatron-as404-2a1-8i2/): A 4U liquid-cooled server with eight AMD Instinct MI355X GPUs and dual EPYC 9005 CPUs for AI training, inference and HPC. - [Penguin Solutions Altus XE4318GTS-DTC](https://aiserverlist.com/penguin-altus-xe4318gts-dtc/): An AMD EPYC HGX B200 server for AI training, inference, and HPC. Direct liquid cooling supports a compact chassis. - [QCT QuantaGrid D75H-10U](https://aiserverlist.com/qct-quantagrid-d75h-10u/): QCT's 10U HGX B300 server with two Xeon 6 processors, 12 front-access PSUs and eight 800G ConnectX-8 ports. Built for large AI training and inference deployments. - [QCT QuantaGrid D75L-3U](https://aiserverlist.com/qct-quantagrid-d75l-3u/): QCT's 3U liquid-cooled HGX B300 node for 19-inch MGX racks with a busbar and power shelves, up to nine nodes (72 GPUs) per rack. - [QCT QuantaGrid D75T-7U](https://aiserverlist.com/qct-quantagrid-d75t-7u/): An AMD EPYC-based GPU platform from Quanta’s data-center business. AMD’s partner catalog identifies air-cooled MI325X and MI350X configurations. - [Supermicro A+ Server AS-4126GS-NBR-LCC](https://aiserverlist.com/supermicro-as-4126gs-nbr-lcc/): A liquid-cooled AMD EPYC HGX B200 server for AI and scientific computing. Supermicro requires a complete rack solution and onsite service with this system. - [Supermicro AS-4126GS-NMR-LCC](https://aiserverlist.com/supermicro-as-4126gs-nmr-lcc/): A direct-liquid-cooled 4U dual-EPYC server with eight OAM accelerators for dense AI racks. Takes AMD Instinct MI355X or MI325X. - [Supermicro AS-8126GS-TNMR](https://aiserverlist.com/supermicro-as-8126gs-tnmr/): An air-cooled 8U dual-EPYC server with eight OAM accelerators, built for AI training and inference in standard air-cooled racks. Takes AMD Instinct MI350X or MI325X. - [Supermicro GPU SuperServer SYS-822GS-NBRT](https://aiserverlist.com/supermicro-sys-822gs-nbrt/): An eight-GPU B200 server for enterprise AI training, inference, and scientific computing. The front-access chassis uses air cooling. - [Supermicro SRS-GB200-NVL72](https://aiserverlist.com/supermicro-srs-gb200-nvl72/): Liquid-cooled rack with 72 Blackwell GPUs and 36 Grace CPUs on one NVLink domain, delivered as an integrated Supermicro rack. For AI factories that want exascale compute per rack. - [Supermicro SuperServer SRS-GB300-NVL72](https://aiserverlist.com/supermicro-srs-gb300-nvl72/): Supermicro's liquid-cooled 48U GB300 NVL72 rack: 18 x 1U compute trays, 72 B300 GPUs, and an in-rack 250 kW CDU. For AI factories that want a rack delivered integrated and tested. - [Supermicro SYS-222GS-NB3OT-ALC](https://aiserverlist.com/supermicro-sys-222gs-nb3ot-alc/): Supermicro's 2-OU liquid-cooled HGX B300 node for 21-inch OCP Open Rack V3, with nine nodes per 72-GPU rack. Built for hyperscale-style B300 deployments. - [Supermicro SYS-422GS-NB3RT-ALC](https://aiserverlist.com/supermicro-sys-422gs-nb3rt-alc/): Supermicro's 4U direct-to-chip liquid-cooled HGX B300 server, sold with full liquid-cooled rack integration. Built for dense B300 racks. - [Supermicro SYS-422GS-NBRT-LCC](https://aiserverlist.com/supermicro-sys-422gs-nbrt-lcc/): 4U direct-liquid-cooled server with eight NVIDIA HGX B200 GPUs and dual Xeon 6700 CPUs. Sold with full liquid-cooled rack integration for dense AI deployments. - [Supermicro SYS-822GA-NGR3](https://aiserverlist.com/supermicro-sys-822ga-ngr3/): 8U air-cooled AI training server with eight Intel Gaudi 3 OAM accelerators and dual Xeon 6900 CPUs. Built for teams that want Gaudi 3 scale-out over six onboard 800 GbE ports. - [Supermicro SYS-822GS-NB3RT](https://aiserverlist.com/supermicro-sys-822gs-nb3rt/): Supermicro's 8U air-cooled HGX B300 server with dual Xeon 6700 and eight 800GbE ConnectX-8 ports. Built for data centers that deploy B300 without liquid cooling. - [Thinkmate GPX GE95-2808-8HGX-B2](https://aiserverlist.com/thinkmate-gpx-ge95-2808-8hgx-b2/): A configurable AMD EPYC HGX B200 system for AI and accelerated-computing buyers. Thinkmate assembles and tests configurations and offers advanced parts-replacement warranty options. - [Wistron GS-18G00](https://aiserverlist.com/wistron-gs-18g00/): An AMD-based AI server platform from the Taiwanese ODM. AMD lists it among partner platforms supporting the Pollara 400 AI network adapter. - [Wiwynn NVIDIA GB300 NVL72](https://aiserverlist.com/wiwynn-nvidia-gb300-nvl72/): Wiwynn's fully liquid-cooled, fan-less GB300 NVL72 AI rack for hyperscale data centers, with in-rack CDU and sidecar options. - [Wiwynn NVIDIA Vera Rubin NVL72](https://aiserverlist.com/wiwynn-nvidia-vera-rubin-nvl72/): Wiwynn integrates rack-scale NVIDIA systems for hyperscale cloud providers. Its Vera Rubin offering uses the third-generation MGX framework. - [xFusion FusionServer G8600 V7](https://aiserverlist.com/xfusion-fusionserver-g8600-v7/): An eight-GPU platform for model training, HPC, and data analysis. Separate GPU and CPU power planes support modular servicing. ## Optional - [All records in one file](https://aiserverlist.com/llms-full.txt) - [Sitemap](https://aiserverlist.com/sitemap.xml)