TBC Home →
← All chips

AI accelerator · Global

AMD Instinct MI355X

CDNA4 flagship (June 2025) with 288GB HBM3e at 8 TB/s, 185B transistors on TSMC 3nm compute dies, and MXFP4/MXFP6 support. AMD discloses no price; analyst estimates cluster around $25k-30k per card.

GlobalShippingAMDCDNA 4 familyLaunched 2025-06TSMC 3nm + 6nm

Latest observation ~$25K (per card) as of 2026-08 · researched 2026-08-27

~$25Klatest observation · analyst estimate · 2026-08
2dated price observations on record
10sourced specs (all linked below)

Specifications

Sourced specs

SpecValueSource
MXFP4 compute10.1 PFLOPS (dense)via amd.com ↗2026-08-27
MXFP6 compute10.1 PFLOPSvia amd.com ↗2026-08-27
OCP-FP8 matrix5.0 PFLOPS dense / 10.1 PFLOPS with sparsityvia amd.com ↗2026-08-27
INT8 matrix5.0 POPS dense / 10.1 POPS with sparsityvia amd.com ↗2026-08-27
FP16/BF16 matrix2.5 PFLOPS dense / 5.0 PFLOPS with sparsityvia amd.com ↗2026-08-27
TF32 matrix1.25 PFLOPS dense / 2.5 PFLOPS with sparsityvia amd.com ↗2026-08-27
FP32 / FP64 compute157.3 TFLOPS FP32 / 78.6 TFLOPS FP64via amd.com ↗2026-08-27
Memory288GB HBM3E, 8 TB/svia amd.com ↗2026-08-27
InterconnectInfinity Fabric 4th gen: 7 scale-up links at 153 GB/s + 1 scale-out link at 128 GB/s; PCIe Gen5via amd.com ↗2026-08-27
Package / TDP185B transistors, TSMC 3nm + 6nm, 16,384 SPs / 256 CUs @ 2400 MHz, 1400W TBP (OAM LC)via amd.com ↗2026-08-27

Pricing

Price history, newest first

DatePriceTypeSourceConfidence
2026-08~$25K per cardMI355X modeled at $25k-30k; other estimates run to ~$30k-40k (getdeploying/tech commentary)Analyst estimatevia Epoch AI (citing Seeking Alpha) ↗low
2026-08~$9 per GPU-hour (cloud on-demand)From ~$8.60/hr per MI355X GPUMarket indexvia getdeploying cloud price index (retrieved Aug 2026) ↗low

vendor-official or multiple independent reports · single major report or several secondary · estimate or lone secondary source

Prices in USD, converted at each observation year's average exchange rate; the original figure and rate print under every converted price. How prices are sourced and graded ↗

Context

Notes

  • Pricing not publicly disclosed by AMD; only analyst estimates available.
  • Launched 2025-06-12 at Advancing AI; volume shipments H2 2025. A Morgan Stanley TCO analysis (Sept 2025) benchmarked MI355X racks against GB200 NVL72.

Where it sits

In the stack

AI accelerators / GPUs →All chip pricing →

All specs and prices link to their original sources, dated as observed. Figures are market observations, not offers; treat each as of its date.