AI ASIC Database
Custom AI silicon purpose-built for machine learning. 71 chips across 28 vendors, from Google TPUs to Cerebras wafers, compared on peak performance, memory, power, and how you can actually get one. Click any name for full specs.
71
Chips Tracked
28
Vendors
42
Cloud Only
8
You Can Buy
Most of these chips are rented, not bought
Unlike GPUs, most custom AI silicon never reaches a retail shelf. Google TPUs, AWS Trainium and Inferentia, and Groq LPUs are rented by the hour or reached through an API, so those rows carry a cloud price per hour instead of a buy button. Cerebras and SambaNova sell whole systems through enterprise sales teams, which is a procurement call, not a checkout page. Where a chip genuinely can be purchased as a card or module, the row shows a buy link.
The comparison still matters even if you never buy one: the availability column tells you whether a chip is a real option for your stack or just a benchmark headline. The few Amazon links on this page are tagged affiliate links; using them costs you nothing and helps fund the testing.
Showing 71 of 71
| Peak Perf | Memory | TDP | Buy | |||
|---|---|---|---|---|---|---|
AMD Instinct MI400 [VERIFY roadmap]Cloud Training AMD | 40 PFLOPS (FP4) [VERIFY] | 432 GB HBM4 | ~1400W | n/a | 2026-09 | Amazon |
NVIDIA Vera RubinCloud Training NVIDIA | 50 PFLOPS (NVFP4) [VERIFY] | 288 GB HBM4 | ~1800W | n/a | 2026-09 | Amazon |
Groq 3 LPU / LPX rackCloud Inference NVIDIA | 35x tok/MW vs Blackwell NVL72 (vendor claim) | 128 GB (rack SRAM) SRAM + 12TB DDR5 | Not disclosed (rack-scale) | n/a | 2026-09 | Amazon |
Maia 200Cloud Inference Microsoft | >10 PFLOPS (FP4, vendor-stated) | 216 GB HBM3E | 750W | n/a | 2026-01 | Amazon |
Trainium3Cloud Training AWS | ~5 PFLOPS (FP8) [VERIFY ship] | 144 GB HBM3e | 1000W | $3.50 | 2025-12 | Cloud Only |
Inferentia3 [VERIFY]Cloud Inference AWS | TBD [VERIFY ship] | 64 GB HBM3 | 250W | $0.85 | 2025-12 | Cloud Only |
SohuCloud Inference Etched | 500K+ tok/s on Llama 70B | 144 GB HBM3e | 750W | n/a | 2025-09 | Cloud Only |
LPU v2Cloud Inference Groq | 1,500+ tok/s on Llama 3 | 230 MB SRAM | 350W | $0.45 | 2025-06 | Cloud Only |
AMD Instinct MI350XCloud Training AMD | 18.4 PFLOPS (FP4, sparse) | 288 GB HBM3E | 1000W | $4.20 | 2025-06 | Amazon |
TPU v7 (Ironwood)Cloud Training Google | 4,614 TFLOPS (BF16) | 192 GB HBM3 | 800W | $4.50 | 2025-03 | Cloud Only |
Trn2 UltraServer (64-chip)Cloud Training AWS | 83.2 PFLOPS (FP8) | 6 TB HBM3 (aggregate) | 25,600W | $110.00 | 2024-12 | Cloud Only |
Trainium2Cloud Training AWS | 1.3 PFLOPS (FP16) | 64 GB HBM3 | 400W | $1.89 | 2024-11 | Cloud Only |
Maia 100Cloud Training Microsoft | 1,600 TFLOPS (MXInt8) | 64 GB HBM2e | 700W | $2.20 | 2024-11 | Cloud Only |
QuietBoxOn-Premise Tenstorrent | 1.86 PFLOPS (BF16) | 96 GB GDDR6 | 1,500W | n/a | 2024-11 | Available to Buy |
Maverick-2On-Premise NextSilicon | 32 TFLOPS (FP64) | 96 GB HBM3 | 750W | $2.40 | 2024-11 | Cloud Only |
Blackhole TensixOn-Premise Tenstorrent | 664 TFLOPS (BF16) | 32 GB GDDR6 | 350W | n/a | 2024-10 | Available to Buy |
TPU v6e (Trillium)Cloud Training Google | 918 TFLOPS (BF16) | 32 GB HBM | 400W | $2.70 | 2024-10 | Cloud Only |
Ascend 910CCloud Training Huawei | 800 TFLOPS (FP16) | 128 GB HBM | 550W | $2.60 | 2024-10 | Cloud Only |
Galaxy Wormhole ServerOn-Premise Tenstorrent | 10.6 PFLOPS (BF16) | 384 GB GDDR6 | 8,000W | n/a | 2024-10 | Available to Buy |
Blackhole p150aOn-Premise Tenstorrent | 745 TFLOPS (BF16) | 32 GB GDDR6 | 300W | n/a | 2024-10 | Available to Buy |
Gaudi3Cloud Training Intel | 1.8 PFLOPS (BF16) | 128 GB HBM2e | 600W | $2.10 | 2024-09 | Cloud Only |
Hailo-10HEdge Hailo | 40 TOPS (INT4) | on-chip LPDDR4 | 3.5W | n/a | 2024-09 | Available to Buy |
TPU v5p (Trillium)Cloud Training Google | 918 TFLOPS (BF16) | 95 GB HBM2e | 350W | $2.40 | 2024-08 | Cloud Only |
RNGD (Renegade)Cloud Inference FuriosaAI | 512 TFLOPS (FP8) | 48 GB HBM3 | 180W | $1.00 | 2024-08 | Cloud Only |
Wormhole n300sOn-Premise Tenstorrent | 466 TFLOPS (BF16) | 24 GB GDDR6 | 300W | n/a | 2024-07 | Available to Buy |
Graviton4Cloud Inference AWS | 96-core Neoverse V2 | 768 GB DDR5 | 350W | $1.85 | 2024-07 | Cloud Only |
Wormhole n150sOn-Premise Tenstorrent | 262 TFLOPS (BF16) | 12 GB GDDR6 | 160W | n/a | 2024-07 | Available to Buy |
SN40L RDUOn-Premise SambaNova | 638 TFLOPS (BF16) | 1 TB External DDR | 400W | $3.20 | 2024-06 | Cloud + Buy |
DataScale SN40L-8On-Premise SambaNova | 5.1 PFLOPS (BF16) | 8 TB External DDR (aggregate) | 3,200W | $8.50 | 2024-06 | Cloud + Buy |
Cloud AI 100 UltraCloud Inference Qualcomm | 870 TOPS (INT8) | 128 GB LPDDR4x | 150W | $0.85 | 2024-05 | Cloud Only |
MTIA v2Cloud Inference Meta | 354 TFLOPS (INT8) | 128 GB LPDDR5 | 90W | n/a | 2024-04 | Cloud Only |
PassageOn-Premise Lightmatter | 30 Tbps optical I/O | N/A (interconnect) Photonic | 60W | n/a | 2024-04 | Cloud Only |
Metis AIPUEdge Axelera | 214 TOPS (INT8) | 1 GB LPDDR4x | 12W | n/a | 2024-04 | Available to Buy |
WSE-3On-Premise Cerebras | 125 petaFLOPS (FP16) | 900 MB SRAM (on-wafer) | 23,000W | $2.50 | 2024-03 | Cloud + Buy |
CS-3 SystemOn-Premise Cerebras | 125 petaFLOPS (FP16) | 44 GB on-chip + 1.5 TB MemoryX SRAM + DDR | 23,000W | $3.20 | 2024-03 | Cloud + Buy |
GroqRack (8x LPU)On-Premise Groq | 8 PetaOP/s | 1.8 GB SRAM (8x distributed) | 2,400W | n/a | 2024-03 | Available to Buy |
LPU (Language Processing Unit)Cloud Inference Groq | 800+ tok/s on Llama 3 | 230 MB SRAM | 300W | $0.27 | 2024-01 | Cloud Only |
MTT S4000Cloud Training Moore Threads | 100 TFLOPS (FP16) | 48 GB GDDR6 | 450W | $1.10 | 2023-12 | Cloud Only |
AtomCloud Inference Rebellions | 128 TOPS (INT8) | 16 GB GDDR6 | 65W | $0.55 | 2023-11 | Cloud Only |
TPU v5eCloud Inference Google | 393 TFLOPS (BF16) | 16 GB HBM2e | 150W | $1.20 | 2023-08 | Cloud Only |
Ascend 910BCloud Training Huawei | 320 TFLOPS (FP16) | 64 GB HBM2e | 400W | $1.80 | 2023-08 | Cloud Only |
TPU v5e Pod (256-chip)Cloud Training Google | 100 PFLOPS (BF16) | 4 TB HBM2e (aggregate) | 38,400W | $78.00 | 2023-08 | Cloud Only |
Hailo-15Edge Hailo | 20 TOPS (INT8) | on-chip LPDDR4 | 1.5W | n/a | 2023-06 | Available to Buy |
Inferentia2Cloud Inference AWS | 380 TFLOPS (FP16) | 32 GB HBM3 | 200W | $0.75 | 2023-04 | Cloud Only |
Wormhole TensixOn-Premise Tenstorrent | 320 TFLOPS (BF16) | 16 GB GDDR6 | 250W | n/a | 2023-04 | Available to Buy |
SN30 RDUOn-Premise SambaNova | 365 TFLOPS (BF16) | 512 GB External DDR | 300W | $2.10 | 2022-09 | Cloud + Buy |
WarboyCloud Inference FuriosaAI | 64 TOPS (INT8) | 32 GB LPDDR4X | 40W | $0.35 | 2022-09 | Cloud Only |
M1076 AMPEdge Mythic | 25 TOPS (INT8) | 64 MB Analog (on-chip) | 3W | n/a | 2022-08 | Amazon |
BR104Cloud Training Biren | 128 TFLOPS (FP16) | 32 GB HBM2e | 300W | $1.30 | 2022-08 | Cloud Only |
M1108 AMPEdge Mythic | 35 TOPS (INT8) | 108 MB Analog (on-chip) | 4W | n/a | 2022-08 | Available to Buy |
Ara-1Edge Kinara | 10 TOPS (INT8) | on-chip SRAM | 2.5W | n/a | 2022-06 | Available to Buy |
Gaudi2Cloud Training Intel | 432 TFLOPS (BF16) | 96 GB HBM2e | 600W | $1.40 | 2022-05 | Cloud Only |
Grayskull e150On-Premise Tenstorrent | 92 TFLOPS (FP16) | 8 GB LPDDR4 | 200W | n/a | 2022-04 | Available to Buy |
Grayskull e75On-Premise Tenstorrent | 55 TFLOPS (FP16) | 8 GB LPDDR4 | 75W | n/a | 2022-04 | Available to Buy |
Bow Pod64 (IPU)On-Premise Graphcore | 1.4 petaFLOPS (FP16) | 260 GB In-Processor | 2,400W | $4.00 | 2022-03 | Cloud + Buy |
Bow-2000 (IPU)On-Premise Graphcore | 350 TFLOPS (FP16) | 65 GB In-Processor | 600W | $2.20 | 2022-03 | Cloud + Buy |
MLU370-X8Cloud Training Cambricon | 256 TOPS (INT8) | 48 GB LPDDR5 | 250W | $1.40 | 2022-03 | Cloud Only |
Bow Pod16 (IPU)On-Premise Graphcore | 350 TFLOPS (FP16) | 65 GB In-Processor | 600W | $1.10 | 2022-03 | Cloud + Buy |
TrainiumCloud Training AWS | 190 TFLOPS (FP16) | 32 GB HBM2e | 300W | $1.25 | 2021-12 | Cloud Only |
Tiangai 100Cloud Training Iluvatar | 147 TFLOPS (FP16) | 32 GB HBM2 | 250W | $1.10 | 2021-12 | Cloud Only |
CloudBlazer i20Cloud Inference Enflame | 200 TOPS (INT8) | 32 GB HBM2 | 150W | $0.95 | 2021-10 | Cloud Only |
Cloud AI 80Cloud Inference Qualcomm | 350 TOPS (INT8) | 32 GB LPDDR4x | 75W | $0.45 | 2021-09 | Cloud Only |
GrayskullOn-Premise Tenstorrent | 160 TFLOPS (FP16) | 8 GB GDDR6 | 200W | n/a | 2021-06 | Available to Buy |
TPU v4Cloud Training Google | 548 TFLOPS (BF16) | 32 GB HBM2e | 250W | $1.80 | 2021-05 | Cloud Only |
WSE-2On-Premise Cerebras | 52 petaFLOPS (FP16) | 40 GB SRAM (on-wafer) | 23,000W | $1.80 | 2021-04 | Cloud + Buy |
Hailo-8Edge Hailo | 26 TOPS (INT8) | on-chip LPDDR4 | 2.5W | n/a | 2021-03 | Amazon |
GroqChip 1On-Premise Groq | 1 PetaOP/s | 220 MB SRAM | 250W | n/a | 2020-10 | Cloud + Buy |
InferentiaCloud Inference AWS | 128 TOPS (INT8) | 8 GB DDR4 | 40W | $0.45 | 2020-05 | Cloud Only |
Habana Gaudi (HL-205)Cloud Training Intel | 184 TFLOPS (BF16) | 32 GB HBM2 | 350W | $0.90 | 2019-06 | Cloud Only |
Edge TPUEdge Google | 4 TOPS (INT8) | 8 MB SRAM (on-chip) | 2W | n/a | 2019-03 | Amazon |
Habana Goya (HL-100)Cloud Inference Intel | 5.6 TFLOPS (FP16) | 16 GB DDR4 | 200W | $0.55 | 2019-01 | Cloud Only |
AMDCloud Training
AMD Instinct MI400 [VERIFY roadmap]432 GB HBM4~1400W
Full specsCloud Only
Check Price on AmazonNVIDIACloud Training
NVIDIA Vera Rubin288 GB HBM4~1800W
Full specsCloud Only
Check Price on AmazonNVIDIACloud Inference
Groq 3 LPU / LPX rack128 GB (rack SRAM) SRAM + 12TB DDR5Not disclosed (rack-scale)
Full specsCloud Only
Check Price on AmazonMicrosoftCloud Inference
Maia 200216 GB HBM3E750W
Full specsCloud Only
Check Price on AmazonAWSCloud Training
Trainium3144 GB HBM3e1000W
Full specs$3.50/hr
Cloud Only, no retail listing
AWSCloud Inference
Inferentia3 [VERIFY]64 GB HBM3250W
Full specs$0.85/hr
Cloud Only, no retail listing
EtchedCloud Inference
Sohu144 GB HBM3e750W
Full specsCloud Only
Cloud Only, no retail listing
GroqCloud Inference
LPU v2230 MB SRAM350W
Full specs$0.45/hr
Cloud Only, no retail listing
AMDCloud Training
AMD Instinct MI350X288 GB HBM3E1000W
Full specs$4.20/hr
Check Price on AmazonGoogleCloud Training
TPU v7 (Ironwood)192 GB HBM3800W
Full specs$4.50/hr
Cloud Only, no retail listing
AWSCloud Training
Trn2 UltraServer (64-chip)6 TB HBM3 (aggregate)25,600W
Full specs$110.00/hr
Cloud Only, no retail listing
AWSCloud Training
Trainium264 GB HBM3400W
Full specs$1.89/hr
Cloud Only, no retail listing
MicrosoftCloud Training
Maia 10064 GB HBM2e700W
Full specs$2.20/hr
Cloud Only, no retail listing
TenstorrentOn-Premise
QuietBox96 GB GDDR61,500W
Full specs$15,000
Available to Buy, no retail listing
NextSiliconOn-Premise
Maverick-296 GB HBM3750W
Full specs$2.40/hr
Cloud Only, no retail listing
TenstorrentOn-Premise
Blackhole Tensix32 GB GDDR6350W
Full specs$1,399
Available to Buy, no retail listing
GoogleCloud Training
TPU v6e (Trillium)32 GB HBM400W
Full specs$2.70/hr
Cloud Only, no retail listing
HuaweiCloud Training
Ascend 910C128 GB HBM550W
Full specs$2.60/hr
Cloud Only, no retail listing
TenstorrentOn-Premise
Galaxy Wormhole Server384 GB GDDR68,000W
Full specs$98,000
Available to Buy, no retail listing
TenstorrentOn-Premise
Blackhole p150a32 GB GDDR6300W
Full specs$1,299
Available to Buy, no retail listing
IntelCloud Training
Gaudi3128 GB HBM2e600W
Full specs$2.10/hr
Cloud Only, no retail listing
GoogleCloud Training
TPU v5p (Trillium)95 GB HBM2e350W
Full specs$2.40/hr
Cloud Only, no retail listing
FuriosaAICloud Inference
RNGD (Renegade)48 GB HBM3180W
Full specs$1.00/hr
Cloud Only, no retail listing
TenstorrentOn-Premise
Wormhole n300s24 GB GDDR6300W
Full specs$1,399
Available to Buy, no retail listing
AWSCloud Inference
Graviton4768 GB DDR5350W
Full specs$1.85/hr
Cloud Only, no retail listing
TenstorrentOn-Premise
Wormhole n150s12 GB GDDR6160W
Full specs$999
Available to Buy, no retail listing
SambaNovaOn-Premise
SN40L RDU1 TB External DDR400W
Full specs$3.20/hr
Cloud + Buy, no retail listing
SambaNovaOn-Premise
DataScale SN40L-88 TB External DDR (aggregate)3,200W
Full specs$8.50/hr
Cloud + Buy, no retail listing
QualcommCloud Inference
Cloud AI 100 Ultra128 GB LPDDR4x150W
Full specs$0.85/hr
Cloud Only, no retail listing
MetaCloud Inference
MTIA v2128 GB LPDDR590W
Full specsCloud Only
Cloud Only, no retail listing
LightmatterOn-Premise
PassageN/A (interconnect) Photonic60W
Full specsCloud Only
Cloud Only, no retail listing
AxeleraEdge
Metis AIPU1 GB LPDDR4x12W
Full specs$149
Available to Buy, no retail listing
CerebrasOn-Premise
WSE-3900 MB SRAM (on-wafer)23,000W
Full specs$2.50/hr
Cloud + Buy, no retail listing
CerebrasOn-Premise
CS-3 System44 GB on-chip + 1.5 TB MemoryX SRAM + DDR23,000W
Full specs$3.20/hr
Cloud + Buy, no retail listing
GroqOn-Premise
GroqRack (8x LPU)1.8 GB SRAM (8x distributed)2,400W
Full specs$250,000
Available to Buy, no retail listing
GroqCloud Inference
LPU (Language Processing Unit)230 MB SRAM300W
Full specs$0.27/hr
Cloud Only, no retail listing
Moore ThreadsCloud Training
MTT S400048 GB GDDR6450W
Full specs$1.10/hr
Cloud Only, no retail listing
RebellionsCloud Inference
Atom16 GB GDDR665W
Full specs$0.55/hr
Cloud Only, no retail listing
GoogleCloud Inference
TPU v5e16 GB HBM2e150W
Full specs$1.20/hr
Cloud Only, no retail listing
HuaweiCloud Training
Ascend 910B64 GB HBM2e400W
Full specs$1.80/hr
Cloud Only, no retail listing
GoogleCloud Training
TPU v5e Pod (256-chip)4 TB HBM2e (aggregate)38,400W
Full specs$78.00/hr
Cloud Only, no retail listing
AWSCloud Inference
Inferentia232 GB HBM3200W
Full specs$0.75/hr
Cloud Only, no retail listing
TenstorrentOn-Premise
Wormhole Tensix16 GB GDDR6250W
Full specs$999
Available to Buy, no retail listing
SambaNovaOn-Premise
SN30 RDU512 GB External DDR300W
Full specs$2.10/hr
Cloud + Buy, no retail listing
FuriosaAICloud Inference
Warboy32 GB LPDDR4X40W
Full specs$0.35/hr
Cloud Only, no retail listing
BirenCloud Training
BR10432 GB HBM2e300W
Full specs$1.30/hr
Cloud Only, no retail listing
MythicEdge
M1108 AMP108 MB Analog (on-chip)4W
Full specs$249
Available to Buy, no retail listing
IntelCloud Training
Gaudi296 GB HBM2e600W
Full specs$1.40/hr
Cloud Only, no retail listing
TenstorrentOn-Premise
Grayskull e1508 GB LPDDR4200W
Full specs$799
Available to Buy, no retail listing
TenstorrentOn-Premise
Grayskull e758 GB LPDDR475W
Full specs$599
Available to Buy, no retail listing
GraphcoreOn-Premise
Bow Pod64 (IPU)260 GB In-Processor2,400W
Full specs$4.00/hr
Cloud + Buy, no retail listing
GraphcoreOn-Premise
Bow-2000 (IPU)65 GB In-Processor600W
Full specs$2.20/hr
Cloud + Buy, no retail listing
CambriconCloud Training
MLU370-X848 GB LPDDR5250W
Full specs$1.40/hr
Cloud Only, no retail listing
GraphcoreOn-Premise
Bow Pod16 (IPU)65 GB In-Processor600W
Full specs$1.10/hr
Cloud + Buy, no retail listing
AWSCloud Training
Trainium32 GB HBM2e300W
Full specs$1.25/hr
Cloud Only, no retail listing
IluvatarCloud Training
Tiangai 10032 GB HBM2250W
Full specs$1.10/hr
Cloud Only, no retail listing
EnflameCloud Inference
CloudBlazer i2032 GB HBM2150W
Full specs$0.95/hr
Cloud Only, no retail listing
QualcommCloud Inference
Cloud AI 8032 GB LPDDR4x75W
Full specs$0.45/hr
Cloud Only, no retail listing
TenstorrentOn-Premise
Grayskull8 GB GDDR6200W
Full specs$599
Available to Buy, no retail listing
GoogleCloud Training
TPU v432 GB HBM2e250W
Full specs$1.80/hr
Cloud Only, no retail listing
CerebrasOn-Premise
WSE-240 GB SRAM (on-wafer)23,000W
Full specs$1.80/hr
Cloud + Buy, no retail listing
GroqOn-Premise
GroqChip 1220 MB SRAM250W
Full specsCloud + Buy
Cloud + Buy, no retail listing
AWSCloud Inference
Inferentia8 GB DDR440W
Full specs$0.45/hr
Cloud Only, no retail listing
IntelCloud Training
Habana Gaudi (HL-205)32 GB HBM2350W
Full specs$0.90/hr
Cloud Only, no retail listing
IntelCloud Inference
Habana Goya (HL-100)16 GB DDR4200W
Full specs$0.55/hr
Cloud Only, no retail listing
Affiliate links, we earn from qualifying purchases at no cost to you.