Cloud · AI infrastructure · In-house silicon

Microsoft

Big-tech operator of Windows, Office and the Azure cloud. FY2025 (ended June 2025) revenue was $281.7B. To expand AI data centers it raised cash capex about 45% in one year to $64.6B. As Nvidia’s largest GPU buyer, plus AMD, TSMC and its own Maia silicon, it is a key demand sink for AI chips.

MSFT Updated

Money in기업·소비자 (Azure·M365·Windows)Revenue $281.7BOpenAI
MicrosoftFY2025
Money outNVIDIAUnits 485,000AMDTSMCAstera LabsIntelMarvell메모리·HBM (SK하이닉스·삼성·마이크론)

Financials & Segments

  • Total revenue · FY2025 (ended June 2025) revenue $281.7B (+15%)
  • Customer concentration · Demand sink for AI infrastructure. Cash capex $64.6B (FY2025, up 45% from $44.5B). GPUs from Nvidia (485k Hopper units in 2024, its largest buyer) and AMD, foundry at TSMC (its own Maia chip), connectivity from Astera Labs.
  • Segments · Intelligent Cloud · Productivity and Business Processes · More Personal Computing

Revenue from

  • 기업·소비자 (Azure·M365·Windows) High confidence· Revenue $281.7B · FY2025 · 10-K annual total revenue
    The demand end of the chain: sells cloud (Azure), productivity software (M365) and Windows to enterprises and consumers. FY2025 total revenue $281.7B.
    Confirms EN SEC EDGAR Published Jul 30, 2025: "FY2025 total revenue $281.7B: Intelligent Cloud $106.3B, Productivity and Business Processes $120.8B, More Personal Computing $54.6B."
  • OpenAI Medium confidence
    Azure is OpenAI’s primary cloud compute provider; Maia was co-designed with OpenAI for its model workloads.
    Confirms EN Microsoft Azure Published Nov 15, 2023: "Microsoft says Maia was designed for workloads such as Azure OpenAI Service, developed in collaboration with OpenAI."

Pays to

  • NVIDIA High confidence· Units 485,000 (est.) · CY2024 · Omdia estimate
    Nvidia’s largest Hopper GPU buyer: about 485k units (H100/H200) in 2024, roughly double the next biggest (Omdia estimate).
    Mentions EN SEC EDGAR Published Jul 30, 2025: "10-K: FY2025 cash capex of $64.6B (up 45% from $44.5B) reflects a surge in AI data-center and accelerator investment (vendors not named)."
    Confirms EN heise online Published Dec 19, 2024: ""Microsoft purchased 485,000 Hopper GPUs, i.e. H100 and H200, in 2024," making it "by far Nvidia’s largest customer for AI accelerators.""
    Mentions KO M Today Accessed Jun 20, 2026: "Microsoft says it will gradually replace the AMD and Nvidia chips it uses in data centers with its own silicon, confirming heavy current Nvidia GPU use."
  • AMD High confidence
    Runs Azure OpenAI Service (GPT-3.5/4) and Copilot on AMD Instinct MI300X; offers ND MI300X v5 VMs.
    Confirms EN AMD Investor Relations Published May 21, 2024: ""The AMD Instinct MI300X and ROCm software stack is powering the Azure OpenAI Chat GPT 3.5 and 4 services.""
    Confirms KO Seoul Economic Daily Accessed Jun 20, 2026: "AMD names Microsoft, alongside OpenAI and Meta, among MI300X customers."
  • TSMC High confidence
    Has its own Maia AI chips (Maia 100/200) fabricated at TSMC on N5 with CoWoS-S packaging; Maia 100 carries 64GB HBM2E.
    Confirms EN ServeTheHome Published Aug 26, 2024: "Maia 100 is a "TSMC 5nm part" using "TSMC CoWoS-S" packaging with "64GB of HBM2E.""
    Confirms KO M Today Accessed Jun 20, 2026: "Reports Microsoft will deploy its own inference chip, Maia 200, in 2026."
  • Astera Labs Medium confidence
    A key hyperscaler customer buying PCIe/Ethernet retimers and fabric connectivity directly (concentrated customer base).
    Confirms EN SEC EDGAR Published Feb 10, 2026: "Astera Labs’ 10-K lists Microsoft among the concentrated hyperscaler customers buying its retimers and fabric directly."
  • Intel Medium confidence
    Buys Intel Xeon CPUs for Azure servers; the ND MI300X v5 VM uses 4th-gen Xeon as host CPU.
    Mentions EN AMD Investor Relations Published May 21, 2024: "The ND MI300X v5 series pairs 8 MI300X GPUs with "two 4th Gen Intel Xeon Scalable processors.""
  • Marvell Medium confidence
    Cited as a hyperscaler custom-silicon partner (specific program and volume undisclosed); attribution kept cautious.
    Mentions KO Korea Economic Daily Published Apr 14, 2026: "Reports Marvell supplies custom AI silicon to hyperscalers (no public attribution to a specific Maia program)."
  • 메모리·HBM (SK하이닉스·삼성·마이크론) Low confidence
    Buys server DRAM/SSD and Maia HBM2E from the three memory makers (specific supplier and volume undisclosed).
    Mentions EN ServeTheHome Published Aug 26, 2024: "Maia 100 uses 64GB of HBM2E (HBM supplied by SK hynix, Samsung and Micron)."

← All companiesMarkdown · JSON

For information only, not investment advice. Data is refined from public sources and may contain errors.

ValueChain.wikiCompany value-chain data