AI accelerators · CPU · Adaptive SoC

AMD

A key Nvidia rival. FY2025 revenue $34.6B (+34%, data center $16.6B); Q1 FY2026 data center $5.8B (+57%) as MI350 ramps. Supplies Instinct MI350 (288GB HBM3E) to Microsoft, Meta, Oracle and OpenAI. Chips at TSMC (N3P) and GlobalFoundries; HBM from Samsung and Micron.

AMD Updated Jun 21, 2026 1 AI consumers· 1 recurring

Money in ▸Oracle●Units 50,000OpenAI●Microsoft●Meta Platforms●

AMD$34.6B

▸ Money outTSMC●Samsung Electronics●Micron Technology●Synopsys●Cadence Design Systems●GlobalFoundries◐Samsung Electro-Mechanics◐SK hynix◐ASE Technology◐Arm Holdings◐Ibiden◐

Financials & Segments

Total revenue · $34.6B (FY2025, +34%); Q1 FY2026 revenue $10.3B (+38%), data center $5.8B (+57%)
Customer concentration · No single customer ≥10% (FY2025, diversified). Data Center $16.6B
Segments · Data Center($16.6B) · Client + Gaming($14.6B) · Embedded($3.5B)

Revenue from

Oracle High confidence· Units 50,000 · 2026 · MI450 supercluster
First hyperscaler to offer a 50,000-GPU MI450 supercluster.
Confirms EN Oracle Published Oct 14, 2025: "Oracle to offer a publicly available 50,000 MI450 supercluster (starting Q3 2026)."
OpenAI High confidence
6 gigawatt AI infrastructure deal (MI450, from 2H 2026).
Confirms KO The Elec Published Oct 7, 2025: "AMD and OpenAI signed a 6GW GPU supply deal."
Confirms KO Korea Economic Daily Published Oct 6, 2025: "Samsung HBM customer AMD will supply AI accelerators to OpenAI in bulk."
Microsoft High confidence
Runs production Copilot workloads on MI300X in Azure.
Confirms EN AMD Published Jun 12, 2025: "Microsoft, Meta and OpenAI committed to MI350 deployments."
Meta Platforms High confidence
Uses MI300X for Llama 3 and 4 inference.
Mentions EN Data Center Dynamics Published May 6, 2026: "Q1 FY2026 data-center revenue $5.8B (+57%), driven by MI350 and EPYC demand."
Confirms EN AMD Published Jun 12, 2025: "Meta uses MI300X for Llama inference."

Pays to

TSMC High confidence
MI350 XCDs are made on TSMC's N3P node (foundry).
Confirms EN SEC EDGAR Published Feb 4, 2026: "AMD 10-K: utilizes TSMC for wafer production of HPC, FPGA and adaptive SoC products."
Confirms EN TrendForce Published Jun 13, 2025: "MI350X/MI355X are built on TSMC's N3P node."
Samsung Electronics High confidence
HBM3E 12-high supplier for MI350 plus some foundry. Supplying since June 2025.
Confirms KO Korea Economic Daily Published Jun 13, 2025: "Samsung supplies 12-high HBM3E to AMD's MI350 from June 2025."
Confirms EN TrendForce Published Jun 13, 2025: "MI350 uses Samsung and Micron as dual HBM3E suppliers."
Micron Technology High confidence
HBM3E 36GB supplier for MI350 (dual with Samsung).
Confirms EN TrendForce Published Jun 13, 2025: "MI350's 288GB HBM3E comes from Micron and Samsung."
Mentions EN AMD Published Jun 12, 2025: "MI350 series carries 288GB HBM3E."
Synopsys High confidence
Uses/buys EDA tools and IP from Synopsys.
Confirms KO Seoul Economic Daily Published Jul 1, 2025: "Samsung and others use Synopsys EDA."
Confirms EN SEC EDGAR Published Dec 18, 2024: "10-K: EDA tools/IP revenue and major chip customers."
Confirms EN SemiAnalysis Published Jun 1, 2024: "Synopsys leads EDA at ~31%."
Cadence Design Systems High confidence
Uses/buys EDA tools and IP from Cadence Design Systems.
Confirms KO Seoul Economic Daily Published Mar 1, 2025: "Cadence grows on AI-chip design."
Confirms EN SEC EDGAR Published Feb 24, 2025: "10-K: EDA tools/IP revenue."
Confirms EN SemiAnalysis Published Jun 1, 2024: "Cadence is No.2 in EDA at ~30%."
GlobalFoundries Medium confidence
Foundry for certain products.
Confirms EN SEC EDGAR Published Feb 4, 2026: "AMD 10-K: also utilizes GLOBALFOUNDRIES as a foundry."
Samsung Electro-Mechanics Medium confidence
Buys FC-BGA semiconductor substrates for AI accelerators.
Confirms KO The Elec Published Jan 31, 2026: "Samsung Electro-Mechanics added Broadcom to FC-BGA customers (after NVIDIA, AMD, Google, Amazon, Apple), supplying AI-accelerator substrates."
SK hynix Medium confidence
Buys HBM3E for its AI accelerators (MI300/MI350) from SK hynix (one of its suppliers).
Mentions EN Notebookcheck Accessed Jun 19, 2026: "Supplies HBM to global AI customers such as NVIDIA and Google."
ASE Technology Medium confidence
Outsources full-process packaging of Venice CPUs to ASE (from 2026).
Confirms EN TrendForce Published Jan 14, 2026: "AMD Venice to use ASE full-process packaging."
Arm Holdings Medium confidence
Uses/buys CPU-architecture IP (license and royalty) from Arm Holdings.
Confirms KO Economy6 Published Jan 15, 2026: "Apple, Qualcomm, Samsung, NVIDIA use ARM IP."
Confirms EN Arm IR Published Jul 1, 2024: "FY2024 revenue $3.23B."
Confirms EN SEC EDGAR Published May 22, 2024: "20-F: licensing and royalty revenue."
Ibiden Medium confidence
Buys FC-BGA package substrates from Ibiden.
Confirms KO Jabon Published Feb 1, 2026: "¥500B expansion for NVIDIA AI substrates."
Confirms EN Ibiden Published May 1, 2025: "IR: IC package-substrate revenue."
Confirms EN DigiTimes Published Jan 2, 2025: "Exclusive FC-BGA supply to Intel and NVIDIA."

Value-chain ripple (tiers 2–3)

Following node-to-node links up to 3 tiers; each firm is shown once, at its nearest tier.

Leading firms (revenue chain)

T1 Microsoft, Meta Platforms

Beneficiaries (payment chain)

← All companiesMarkdown · JSON