Entity — AMD (Instinct)
Status: active Owner: Finance / Charlie AGT-002 Entity ID: ENTITY-AI-AMD Visibility: PUBLIC Last verified: 2026-07-10
Snapshot
AMD is the merchant GPU challenger with the Instinct line. Current shipping flagship is the MI355X (CDNA 4, 288 GB HBM3E, 5 PF dense FP8), which matches NVIDIA Blackwell-class memory capacity. At CES 2026 AMD announced the full MI400 series (CDNA 5, HBM4) and its first rack-scale product, Helios (72x MI455X), targeted at 2H2026 — AMD's first true rack-unit answer to NVL72-class systems.
Chip Table
| Chip | Status | Key spec highlights | As-of | Source | Grade |
|---|---|---|---|---|---|
| Instinct MI355X | shipping | CDNA 4; TSMC 3nm+6nm chiplets, 185B transistors; 288 GB HBM3E, 8 TB/s; 5.0 PF dense FP8, 10.1 PF FP4 (MXFP4); 1,400 W TBP liquid-cooled; Infinity Fabric 7x153 GB/s | 2026-07-10 | https://www.amd.com/en/products/accelerators/instinct/mi350/mi355x.html | 🟢 |
| Instinct MI350X | shipping | air-cooled sibling of MI355X (1,000 W-class); same 288 GB HBM3E platform | 2026-07-10 | https://www.amd.com/en/products/accelerators/instinct/mi350/mi355x/platform.html | 🟢 (positioning; full sheet not re-verified this pass) |
| Instinct MI455X (Helios) | announced (2H2026) | CDNA 5; 12 TSMC N2 compute chiplets, ~320B transistors; 432 GB HBM4, 19.6 TB/s; vendor claims 40 PF FP4 / 20 PF FP8; Helios rack: 72 GPUs, 2.9 EF FP4, 31 TB HBM4 | 2026-07-10 | https://www.amd.com/en/newsroom/press-releases/2026-1-5-amd-and-its-partners-share-their-vision-for-ai-ev.html | 🟡 (CES announcement; datasheet pending) |
| Instinct MI430X / MI440X / MI450 | announced | MI400-family variants for HPC/sovereign/volume segments; MI450 shares 432 GB HBM4 at lower power | 2026-07-10 | https://www.tomshardware.com/tech-industry/artificial-intelligence/amd-touts-instinct-mi430x-mi440x-and-mi455x-ai-accelerators-and-helios-rack-scale-ai-architecture-at-ces-full-mi400-series-family-fulfills-a-broad-range-of-infrastructure-and-customer-requirements | 🟡 |
| Instinct MI500 | announced (2027, claims only) | vendor "1,000x" AI-performance claim vs unspecified baseline — A1 narrative, no spec sheet; logged for cadence only | 2026-07-10 | https://www.datacenterdynamics.com/en/news/amd-unveils-full-mi400-product-lineup-claims-mi500-chips-will-deliver-1000x-increase-in-ai-performance/ | 🟠 |
Technical Trajectory
- Cadence: annual — MI300X (2023) → MI325X (2024) → MI350/355X (2025) → MI400 series (2H2026) → MI500 (2027 announced). CES 2026 press release (2026-01-05, 🟢 for cadence).
- Memory strategy: capacity leadership — 288 GB HBM3E now, 432 GB HBM4 on MI455X (vs NVIDIA Rubin's claimed 288 GB HBM4). Bandwidth 8 → 19.6 TB/s claimed (🟡 until datasheet).
- Rack scale: Helios (72x MI455X + EPYC "Venice" Zen 6) is AMD's first NVL72-class rack product; a double-wide variant is reported for later 2026 (🟡). Scale-up fabric details (UALink vs Infinity Fabric evolution) not yet on an official spec page — n/a.
- Node: first major accelerator on TSMC N2 compute chiplets if MI455X ships as described (🟡) — would leapfrog NVIDIA Rubin's 3nm-class node on paper.
- Software: ROCm maturity remains the adoption gate; MLPerf Training v5.1 had AMD-ecosystem submissions (AMD, MangoBoost among submitters, 🟢 via MLCommons) — worth pulling exact results next pass.
Watch Items
- MI455X official datasheet (upgrade 40/20 PF and 19.6 TB/s claims to 🟢; confirm dense-vs-sparse basis).
- Helios rack general availability and first hyperscaler deployment disclosures in 2H2026.
- AMD MLPerf Training v6.x / Inference v6.x submissions — MI355X and then MI455X measured results vs spec-sheet ratios.
- FY2026 earnings: Instinct revenue disclosure cadence (data-center segment detail).
Claims
- CLAIM-AI-SILICON-001 — AMD is the second merchant leg of the merchant-vs-custom-ASIC share question.
- CLAIM-AI-BOTTLENECK-001 — 432 GB HBM4 per GPU makes AMD a large HBM demand claimant; relevant to where the supply bottleneck binds.
Sources
- 🟢 AMD MI355X product page — https://www.amd.com/en/products/accelerators/instinct/mi350/mi355x.html (accessed 2026-07-10)
- 🟢 AMD MI355X platform page / datasheets — https://www.amd.com/en/products/accelerators/instinct/mi350/mi355x/platform.html (accessed 2026-07-10)
- 🟢 AMD CES 2026 press release — https://www.amd.com/en/newsroom/press-releases/2026-1-5-amd-and-its-partners-share-their-vision-for-ai-ev.html (accessed 2026-07-10)
- 🟡 Tom's Hardware, MI400 family at CES 2026 — https://www.tomshardware.com/tech-industry/artificial-intelligence/amd-touts-instinct-mi430x-mi440x-and-mi455x-ai-accelerators-and-helios-rack-scale-ai-architecture-at-ces-full-mi400-series-family-fulfills-a-broad-range-of-infrastructure-and-customer-requirements (accessed 2026-07-10)
- 🟡 DCD, MI400 lineup / MI500 claim — https://www.datacenterdynamics.com/en/news/amd-unveils-full-mi400-product-lineup-claims-mi500-chips-will-deliver-1000x-increase-in-ai-performance/ (accessed 2026-07-10)
Changelog
- 2026-07-10: Created; first verification pass.