About us

Zettabyte is building the AI infrastructure of tomorrow. our mission is to fulfill all of the ai computing needs of enterprises and entrepreneurs so that we can propel AI forward for all of humanity.

OUR story
We’re Building AI Infrastructure for All

Drawing on our expertise from hyperscale cloud data centers.
‍
We support startups and businesses globally, ensuring reliable access to AI computing .

We specialize in software enhanced performance optimization of the GPU hardware stack, pushing up MFU efficiency by as much as 20%.

Our partners

Characteristics of professional IDC/Tele-Center service

Custom-made high quality infrastructures
  • Professional facility construction and management service.
  • Professional management team of custom-built facilities.
  • The 99.999% availability via dual loop design.
  • 20A plug-in switch of every circuit, uninterrupted upgrade.
  • Complete separation of electricity and telecom wiring ducts.
  • High quality copper busway class F design.
  • Ground resistance < 0.5 Ω
  • Dual loop electric power, water cooling and telecommunication ducts.
99.999% electricity and backup systems
  • Dual loop electricity from Taiwan Power Company.
  • Dual loop high-voltage transformer.
  • Dual loop uninterruptible power supply system.
  • Emergency power supply battery.
  • Continuous operation generator.
  • K-13 dual loop Trans-isolate power transformer.
Dual loop air conditioning system
  • Water-cool microcomputer thermostatic air conditioner.
  • Dual loop design.
  • Water cooling tower.
  • Refrigerant compressor.
  • 7-day water-storage tank.
  • Leakage monitoring.
  • The best working environment (Temperature of the air conditioner stands at 22°C +/- 1°C, humidity keeps at 50+/- 5%).
Fire precautionary detection and prevention system
  • Concept: Prevention is certainly better than cure.
  • Connect the environmental monitoring system with smoke detecting system.
  • Use FM200 fire extinguisher.
  • Multi-detection to prevent false alarm & action from happening.
  • Use CO2 and powder extinguisher on the basement.
  • Provided with emergency operation procedure to avoid accidents.
Year-round Nonstop maintenance center
  • 24-hr patrol of guards and vehicle lane control.
  • Utilize fingerprint and card readers to conduct the entry and exit control.
  • Set up 360-degree monitoring and motion-detection systems for surveillance.
  • Year-round Nonstop network, security and building environment management.
  • Year-round Nonstop standby maintenance staff.

See our most popular specifications used for AI training

Liquid Cooling H100 Entry Level

**GPU: NVIDIA DELTA-NEXT Liquid Cool (Vulcan) GPU Baseboard, 8** **CPU: 8558P 2P 48C/2.7G/350W(48/2.5,32/3.1/330) 260M SGX512 1xDIQD **Memory: 64GB DDR5-5600 2Rx4 LP (16Gb) ECC RDIMM ,HF, RoHS **Storage: Samsung PM9A3 960GB NVMe PCIe4x4 U.2 7mm 1DWPD 5YR SED Opal **Chassis: 4U X13 Liquid cooled HGX H100 8GPU, X13DEG-M, CSE-GP401 **Networking: 900-9X766-003N-SQ0 1-p NDR 400G IB/ETH OSFP Crypto Disabled **NIC: Std LP 2-port 10G RJ45, Intel X550 (Retail Pack) **Socket: SOCKET E LGA4677, E1A XCC CARRIER, DG 1.0, NO SHIM, BLK HSG **IO Board: NIC IO board for 4U Liquid-Cool GPU System,RoHS **Adapter Board: NIC adaptor board with Switch for 4U Liquid-Cool GPU Sys **Cable 1: I2C, (1x6F to 1x6F)/PH2.0,40CM,24AWG,2A/p,RoHS **Cable 2: 2 MCIOx8 (STR to STR),LtF,FOLD,19cm,32AWG,RoHS **Cable 3: MicroHi,(2x4 to 2x4),PH3.0,30CM,8A/pin,18AWG,RoHS **TPM: SPI TPM 2.0 using SLB9670 (Retail Pack) **Warranty: 0% 3 YRS LABOR, 3 YRS PARTS, 1 YR CRS UNDER LIMITED WRNTY **Assembly Fee

Liquid Cooling H100 High Level

**GPU: NVIDIA DELTA-NEXT Liquid Cool (Vulcan) GPU Baseboard, 8 **CPU: SPR 8480+ 2P 56C 2.0G 350W 105MB BI(1111) E5 4677 **Memory: 96GB DDR5-5600 2Rx4 LP (24Gb) ECC RDIMM,RoHS **Storage: Intel D7-P5620 1.6TB NVMePCIe4.0X4 3D TLC U.2 15mm3DWPD **Chassis: 4U X13 Liquid cooled HGX H100 8GPU, X13DEG-M, CSE-GP401 **Networking: 900-9X766-003N-SQ0 1-p NDR 400G IB/ETH OSFP Crypto Disabled **NIC: Std LP 2-port 10G RJ45, Intel X550 (Retail Pack) **Socket: SOCKET E LGA4677, E1A XCC CARRIER, DG 1.0, NO SHIM, BLK HSG **IO Board: NIC IO board for 4U Liquid-Cool GPU System,RoHS **Adapter Board: NIC adaptor board with Switch for 4U Liquid-Cool GPU Sys **Cable 1: I2C, (1x6F to 1x6F)/PH2.0,40CM,24AWG,2A/p,RoHS **Cable 2: 2 MCIOx8 (STR to STR),LtF,FOLD,19cm,32AWG,RoHS **Cable 3: MicroHi,(2x4 to 2x4),PH3.0,30CM,8A/pin,18AWG,RoHS **TPM: SPI TPM 2.0 using SLB9670 (Retail Pack) **Warranty: 0% 3 YRS LABOR, 3 YRS PARTS, 1 YR CRS UNDER LIMITED WRNTY **Assembly Fee

Liquid Cooling H200 Entry Level

**GPU: NVIDIA DELTA-NEXT Liquid Cool (Vulcan) GPU Baseboard, 8 **CPU: 8558P 2P 48C/2.7G/350W(48/2.5,32/3.1/330) 260M SGX512 1xDIQD **Memory: 64GB DDR5-5600 2Rx4 LP (16Gb) ECC RDIMM ,HF, RoHS **Storage: Samsung PM9A3 960GB NVMe PCIe4x4 U.2 7mm 1DWPD 5YR SED Opal **Chassis: 4U X13 Liquid cooled HGX H100 8GPU, X13DEG-M, CSE-GP401 **Networking: 900-9X766-003N-SQ0 1-p NDR 400G IB/ETH OSFP Crypto Disabled **NIC: Std LP 2-port 10G RJ45, Intel X550 (Retail Pack) **Socket: SOCKET E LGA4677, E1A XCC CARRIER, DG 1.0, NO SHIM, BLK HSG **IO Board: NIC IO board for 4U Liquid-Cool GPU System,RoHS **Adapter Board: NIC adaptor board with Switch for 4U Liquid-Cool GPU Sys **Cable 1: I2C, (1x6F to 1x6F)/PH2.0,40CM,24AWG,2A/p,RoHS **Cable 2: 2 MCIOx8 (STR to STR),LtF,FOLD,19cm,32AWG,RoHS **Cable 3: MicroHi,(2x4 to 2x4),PH3.0,30CM,8A/pin,18AWG,RoHS **TPM: SPI TPM 2.0 using SLB9670 (Retail Pack) **Warranty: 0% 3 YRS LABOR, 3 YRS PARTS, 1 YR CRS UNDER LIMITED WRNTY **Assembly Fee

Liquid Cooling H200 High Level

**GPU: NVIDIA DELTA-NEXT Liquid Cool (Vulcan) GPU Baseboard, 8 **CPU: SPR 8480+ 2P 56C 2.0G 350W 105MB BI(1111) E5 4677 **Memory: 96GB DDR5-5600 2Rx4 LP (24Gb) ECC RDIMM,RoHS **Storage: Intel D7-P5620 1.6TB NVMePCIe4.0X4 3D TLC U.2 15mm3DWPD **Chassis: 4U X13 Liquid cooled HGX H100 8GPU, X13DEG-M, CSE-GP401 **Networking: 900-9X766-003N-SQ0 1-p NDR 400G IB/ETH OSFP Crypto Disabled **NIC: Std LP 2-port 10G RJ45, Intel X550 (Retail Pack) **Socket: SOCKET E LGA4677, E1A XCC CARRIER, DG 1.0, NO SHIM, BLK HSG **IO Board: NIC IO board for 4U Liquid-Cool GPU System,RoHS **Adapter Board: NIC adaptor board with Switch for 4U Liquid-Cool GPU Sys **Cable 1: I2C, (1x6F to 1x6F)/PH2.0,40CM,24AWG,2A/p,RoHS **Cable 2: 2 MCIOx8 (STR to STR),LtF,FOLD,19cm,32AWG,RoHS **Cable 3: MicroHi,(2x4 to 2x4),PH3.0,30CM,8A/pin,18AWG,RoHS **TPM: SPI TPM 2.0 using SLB9670 (Retail Pack) **Warranty: 0% 3 YRS LABOR, 3 YRS PARTS, 1 YR CRS UNDER LIMITED WRNTY **Assembly Fee

Air Cooled H100 Entry Level

**GPU: NVIDIA DELTA-NEXT HGX GPU Baseboard, 8 H100 80GB SXM5** **CPU: 8558P 2P 48C/2.7G/350W(48/2.5,32/3.1/330) 260M SGX512 1xDIQD **Memory: 64GB DDR5-5600 2Rx4 (16Gb) LP ECC RDIMM **Storage: Samsung PM9A3 960GB NVMe PCIe4x4 U.2 7mm 1DWPD 5YR SED Opal **Chassis: 8U X13 8GPU (Rear I/O), X13DEG-OAD, CSE-GP801TS **Networking: 900-9X766-003N-SQ0 1-p NDR 400G IB/ETH OSFP Crypto Disabled **NIC: Std LP 2-port 10G RJ45, Intel X550 (Retail Pack) **Socket: SOCKET E LGA4677, E1A XCC CARRIER, DG 1.0, NO SHIM, BLK HSG **System Management Software: SMC System Management Software Suite Node License **Warranty: 0% 3 YRS LABOR, 3 YRS PARTS, 1 YR CRS UNDER LIMITED WRNTY **Assembly Fee

Air Cooled H100 High Level

**GPU: NVIDIA DELTA-NEXT HGX GPU Baseboard, 8 H100 80GB SXM5 **CPU: SPR 8480+ 2P 56C 2.0G 350W 105MB BI(1111) E5 4677 **Memory: 96GB DDR5-5600 2RX4 (24Gb) ECC RDIMM **Storage: Intel D7-P5620 1.6TB NVMePCIe4.0X4 3D TLC U.2 15mm3DWPD **Chassis: 8U X13 8GPU (Rear I/O), X13DEG-OAD, CSE-GP801TS **Networking: 900-9X766-003N-SQ0 1-p NDR 400G IB/ETH OSFP Crypto **Disabled NIC: Std LP 2-port 10G RJ45, Intel X550 (Retail Pack) **Socket: SOCKET E LGA4677, E1A XCC CARRIER, DG 1.0, NO SHIM, BLK HSG **System Management Software: SMC System Management Software Suite Node License **TPM: SPI TPM 2.0 using SLB9670 (Retail Pack) **Warranty: 0% 3 YRS LABOR, 3 YRS PARTS, 1 YR CRS UNDER LIMITED WRNTY **Assembly Fee

Air Cooled H200 Entry Level

**GPU: NVIDIA DELTA-NEXT (HGX) GPU Baseboard, 8 H200 141GB SXM5 **CPU: 8558P 2P 48C/2.7G/350W(48/2.5,32/3.1/330) 260M SGX512 1xDIQD **Memory: 64GB DDR5-5600 2Rx4 (16Gb) LP ECC RDIMM **Storage: Samsung PM9A3 960GB NVMe PCIe4x4 U.2 7mm 1DWPD 5YR SED Opal **Chassis: 8U X13 8GPU (Rear I/O), X13DEG-OAD, CSE-GP801TS **Networking: 900-9X766-003N-SQ0 1-p NDR 400G IB/ETH OSFP Crypto Disabled **NIC: Std LP 2-port 10G RJ45, Intel X550 (Retail Pack) **Socket: SOCKET E LGA4677, E1A XCC CARRIER, DG 1.0, NO SHIM, BLK HSG **System Management Software: SMC System Management Software Suite Node License **Warranty: 0% 3 YRS LABOR, 3 YRS PARTS, 1 YR CRS UNDER LIMITED WRNTY **Assembly Fee

Air Cooled H200 High Level

**GPU: NVIDIA DELTA-NEXT (HGX) GPU Baseboard, 8 H200 141GB SXM5 **CPU: SPR 8480+ 2P 56C 2.0G 350W 105MB BI(1111) E5 4677 **Memory: 96GB DDR5-5600 2RX4 (24Gb) ECC RDIMM **Storage: Intel D7-P5620 1.6TB NVMePCIe4.0X4 3D TLC U.2 15mm3DWPD **Chassis: 8U X13 8GPU (Rear I/O), X13DEG-OAD, CSE-GP801TS **Networking: 900-9X766-003N-SQ0 1-p NDR 400G IB/ETH OSFP Crypto Disabled **NIC: Std LP 2-port 10G RJ45, Intel X550 (Retail Pack) **Socket: SOCKET E LGA4677, E1A XCC CARRIER, DG 1.0, NO SHIM, BLK HSG **System Management Software: SMC System Management Software Suite Node License **TPM: SPI TPM 2.0 using SLB9670 (Retail Pack) **Warranty: 0% 3 YRS LABOR, 3 YRS PARTS, 1 YR CRS UNDER LIMITED WRNTY **Assembly Fee

8U Air Cooled System With NVIDIA HGX H100/H200 8-GPU

CPU: Dual 5th/4th Gen Intel® Xeon® or AMD EPYC 9004 Series Processors** Memory: 2TB DDR5 (recommended)** GPU: NVIDIA HGX H100/H200 8-GPU (80GB HBM3 or 141GB HBM3e per GPU)900GB/s NVLink GPU-GPU interconnect with NVSwitch** Networking: 8x NVIDIA ConnectX®-7 Single-port 400Gbps/NDR OSFP NICsNetworking2x NVIDIA ConnectX-7 Dual-port 200Gbps/NDR200 QSFP112 NICs1:1 networking to each GPU to enable NVIDIA GPUDirect RDMA and Storage** Storage: 30.4TB NVMe (4x 7.6TB U.3)Storage3.8TB NVMe (2x 1.9TB U.3, Boot) [Optional M.2 available]** Power Supply: 6x 3000W Redundant Titanium Level power supplies**

Fully Integrated air Cooled 32-node Cluster With 256 H100/H200 GPUs

Compute Fabric Leaf: 8x SSE-MQM9700-NS2F, 64-port NVIDIA Quantum-2 InfiniBand 400G NDR, 32OSFP ports switch** Compute FabricSpine: 4x SSE-MQM9700-NS2F, 64-port NVIDIA Quantum-2 InfiniBand 400G NDR, 32OSFP ports switch** In-band Management Switch:
2x SSE-MSN4600-CS2FC 64-port 100GbE QSFP28, 2U switch** Out-of-band Management Switch:
2x SSE-G3748R-SMIS, 48-port 1Gbps Ethernet ToR management switch 1x SSE-F35485R, 48-port 10Gbps Ethernet ToR management switch** Rack: 9x 48U 750mm x 1200mm** PDU: 34x 208V 60A 3Ph

Xwing 4U H100

**GPU: 8 x H100 NVL (Nvidia) **CPU: 2 x AMD Turin EPYC 9575F 64 cores, 400W **Memory: 32 x 64GB RDIMM DDR5-6400 (Hynix) **Storage: 16 x E1.S SSD + Tray (Kioxia XD7p E1.S Gen4 1.92 TB, 15mm Samsung) **Boot Device: 1 x M.2 SSD (M.2 2280 Gen4, 1TB, Micron) **NIC: 1 x Dual Port 25G NIC (Intel® Ethernet Network Adapter OCP3.0 E810-XXVDA2, Pull Tab) **Infiniband: 4 x NVIDIA® ConnectX®-7 400GbE/NDR200 IB (Single-Port OSFP HHHL Adapter Card, PCIe 5.0 x16, QSFP112) **Ethernet: 1 x NVIDIA® DPU BlueField-3 200GbE (Dual-Port FHHL Adapter Card, 2P x QSFP112, 200G, >75W) **NVL Bridge: 6 x NVL Bridge (Nvidia)

Miramar 8U H100

**GPU: 1 x H100 Module (Nvidia) **CPU: 2 x AMD Epyc 9454 48 cores **Memory: 24 x 96GB DDR5 5600 (SKHynix) **Storage: 16 x U.2 2.5” PCIe 5.0 SSD 1 DWPD 1.92TB (KIOXIA CM7) **Additional Storage: 8 x 2.5” SATA Gen4 1.6TB SAS (Kioxia) **Boot Device: 2 x M.2 SSD (M.2 2280 1TB Micron MTFDKBA960TFR-1BC1ZABYY) **RAID **Card: 1 x MegaRAID 9660-16i (Broadcom) **NIC (E-W): 8 x IB CX7 400G OSFP NDR Single Port NIC (Nvidia MCX75310AAS-NEAT) **NIC (N-S): 1 x NVIDIA® ConnectX®-7 200GbE/NDR200 IB (Single-Port OSFP HHHL Adapter Card, PCIe 5.0 x16, Nvidia MCX75310AAS-HEAT) **OCP3.0 NIC (N-S): 1 x Dual Port 10GbE RJ-45 PCIe Card, Intel X710-AT2, OCP3.0

Anakin Rack GPU Systems

**GPU: 72x Blackwell NVL72 (18x Compute Trays, 4 GPUs per tray) **CPU: Grace CPUs (2 per tray, integrated with GPUs) **Memory: 4x E1.S NVMe SSD per tray for GPU Cache **Networking: 9x 5th Generation NVLink™ Switch Trays for intra-rack communication • 1x 32-port 400GbE/IB Switch for North-South traffic • 3x 48-port 1GbE OOB Management Switches **Power: 6x 33KW Power Shelves with N+N redundancy **Cooling: Direct liquid cooling with real-time monitoring and rapid leakage detection **Rack Size: 48RU ORv3 Rack **System Management: Universal Management Server with real-time monitoring

X-Wing Series GPU Systems

**GPU: Supports up to 8 PCIe-based GPU accelerators (e.g., NVIDIA® L40S, H100 NVL, H200 NVL, B100 NVL) **CPU: Dual AMD® EPYCTM Turin 64C TDP 400W Processors **Memory: Up to 32x DDR5 RDIMM **Storage: Default 8x E1.S SSD, expandable to 16x **Expansion Slots: 1x Gen5 x16 HHHL AIC, 5x Gen5 x16 FHHL AIC **Power: 4x 12V 3200W (N+1) Redundant Titanium Grade Power Supplies **Networking: Up to 4x 400Gb/s High Bandwidth RDMA-Enabled Network Ports **System Cooling: Air-cooled with 5x 8080 Hot-swappable fans **Management: Open BMC and GPU management, with whole system health monitoring

Miramar Series GPU Systems

**GPU: Supports Nvidia® HGXTM H100/H200/B200A, AMD® Instinct Mi300X **CPU: Dual AMD® EPYCTM 9004 / 9005 Series Processors, PCIe Gen5 **Memory: 24x DDR5 RDIMM, up to 3072 GB **Storage: Up to 16x 2.5” Hot-Swap NVMe U.2 + 8x 2.5” SAS/SATA (Optional by SAS Raid Card) **Expansion Slots: 8x Gen5 x16 HHHL AIC, 5x Gen5 x16 FHHL AIC, 2x OCP3.0 SFF **Power: 6x 54V 3000W (N+2) Redundant Titanium Grade PSUs • 2x 12V 2400W or 2700W (N+1) PSUs depending on region (NA/EU) **System Cooling: 12x 80105 Hot-swappable fans for system cooling, 4x 6056 fans for SSDs cooling **Management: DC-SCM 2.0 Form Factor, Open BMC, GPU Management

North-Island Series GPU Systems

**GPU: Supports Nvidia® HGXTM B200A Ultra, AMD® Mi325X/Mi355X **CPU: Dual AMD® EPYCTM Turin 64C TDP 400W or Intel® Granite Rapids-SP 350W Processors **Memory: 32x 128GB DDR5 RDIMM or 16x 192GB DDR5 RDIMM **Storage: Up to 8x 2.5” Hot-Swap E1.S NVMe SSD for GPU Cache **Expansion Slots: 8x FHHL PCIe AIC, 2x FHHL PCIe AIC **Power: 8x 54V 3200W (N+N) Redundant Titanium Grade PSUs **System Cooling: Liquid-cooled GPU/CPU/DDR RDIMM, supported by internal fans **Management: Open BMC and GPU management, whole system health monitoring

Give Zettabyte a try and see for yourself, superior MFU in both training and inference.

zettabyte®

Delivering AI Computing to
Enterprises and Entrepreneurs Worldwide.

zettabyte® ai computing
info@zettabyte.space

© 2024 All Rights Reserved.