Cloud · AI infrastructure · In-house silicon (TPU)

Alphabet (Google)

Big-tech operator of Search, YouTube, Android and Google Cloud. FY2025 revenue was $402.8B (+15%) and net income $132.2B. To expand AI data centers it raised cash capex from $52.5B to $91.4B in one year and guided 2026 to $175-185B. It co-designs its own TPU (Ironwood) with Broadcom and builds it at TSMC, while also buying Nvidia GPUs in volume. A key demand sink for AI chips.

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

Money in ▸광고·소비자 (Search·YouTube·Android)●Revenue $402.8BGoogle Cloud 고객 (기업·AI)◐Revenue ~$17.7B
Alphabet (Google)FY2025 revenue $402.8B
▸ Money outNVIDIA●Units 169,000Broadcom●TSMC●SK hynix◐Marvell◐Intel◐Samsung Electronics◐Arista Networks◐

Financials & Segments

  • Total revenue · FY2025 revenue $402.8B (+15%), net income $132.2B
  • Customer concentration · Demand sink for AI infrastructure. Cash capex $91.4B (FY2025, up 74% from $52.5B); 2026 guidance $175-185B. Its own TPU (Ironwood) is co-designed with Broadcom and made by TSMC; GPUs from Nvidia (169k Hopper units in 2024). Diversifying to Intel and Samsung foundries as TSMC hits capacity.
  • Segments · Google Services · Google Cloud · Other Bets

Revenue from

  • 광고·소비자 (Search·YouTube·Android) High confidence· Revenue $402.8B · FY2025 · 10-K annual total revenue
    The demand end of the chain: Search, YouTube and Android ads plus subscriptions make up most revenue. FY2025 total revenue $402.8B.
    Confirms EN SEC EDGAR Published Feb 5, 2026: "FY2025 total revenue $402.8B (+15%), net income $132.2B; YouTube annual revenue over $60B."
  • Google Cloud 고객 (기업·AI) Medium confidence· Revenue ~$17.7B · Q4 2025 · quarterly revenue (run-rate $70B+)
    Sells GCP and TPU compute to enterprise and AI customers. Q4 2025 revenue $17.7B (+48%), run-rate over $70B.
    Confirms EN SEC EDGAR Published Feb 5, 2026: "Google Cloud revenue surged on AI demand (Q4 +48%, run-rate over $70B)."

Pays to

  • NVIDIA High confidence· Units 169,000 (est.) · CY2024 · Omdia estimate
    Buys Nvidia GPUs in volume for Google Cloud: about 169k Hopper units in 2024 (Omdia). One of Nvidia’s top-4 customers.
    Mentions EN SEC EDGAR Published Feb 5, 2026: "10-K: FY2025 cash capex of $91.4B (up 74% from $52.5B) reflects a surge in AI infrastructure and accelerator investment (vendors not named)."
    Confirms EN heise online Published Dec 19, 2024: "Per Omdia, Alphabet (Google) purchased about 169,000 Hopper GPUs in 2024."
  • Broadcom High confidence
    Co-designs and co-develops its own TPU AI chips (Ironwood/TPU v7, next-gen 8t) with Broadcom, under an agreement running to 2031. Its biggest custom-AI-chip partner.
    Confirms EN TrendForce Published Nov 7, 2025: "Google unveiled its 7th-gen TPU Ironwood, co-designed with Broadcom, with a 9,216-chip superpod aimed at Nvidia."
    Confirms EN Tom's Hardware Accessed Jun 20, 2026: "Broadcom designs Ironwood and the TPU 8t training chip under an agreement running through 2031."
    Mentions KO THE ELEC Accessed Jun 20, 2026: "Korean coverage of Google’s next-gen TPU design and production chain."
  • TSMC High confidence
    Has its own TPU Ironwood fabricated at TSMC on N3P (192GB HBM3E); its Axion Arm CPU is also at TSMC.
    Mentions KO Herald Economy Published Jun 10, 2026: "TSMC has manufactured the TPUs, but Google is adding suppliers as TSMC hits capacity."
    Confirms EN TrendForce Published Nov 7, 2025: "Ironwood is built on a TSMC process with 192GB of HBM3E."
  • SK hynix Medium confidence
    Supplies HBM3E (192GB) for TPU Ironwood and data-center memory (supplier/volume undisclosed).
    Confirms EN TrendForce Published Nov 7, 2025: "Ironwood carries 192GB of HBM3E at 7.37 TB/s."
  • Marvell Medium confidence
    Cited as a custom-AI-silicon customer (its main custom partner is Broadcom); attribution kept cautious.
    Mentions KO Herald Economy Published Jun 10, 2026: "Mentioned among suppliers in coverage of Google’s custom-AI-chip supply-chain shift."
  • Intel Medium confidence
    Outsources part of next-gen TPU volume to Intel Foundry as TSMC hits capacity; a share of an estimated 6M TPUs in 2027-2028 goes to Intel.
    Confirms KO Herald Economy Published Jun 10, 2026: "Intel Foundry won Google AI-chip (TPU) orders, emerging as an alternative to TSMC."
  • Samsung Electronics Medium confidence
    Buys HBM/DRAM from Samsung for TPUs and datacenters (60%+ of Google TPU HBM); also evaluating Samsung 2nm foundry for next-gen TPU (not finalized).
    Mentions KO Financial Post Published Jan 15, 2026: "Google 10th-gen TPU in talks for Samsung 2nm."
    Confirms KO Seoul Economic Daily Published Jan 10, 2026: "Samsung supplies 60%+ of Google TPU HBM."
    Mentions KO THE ELEC Accessed Jun 20, 2026: "Google is considering Samsung’s foundry for production of some next-gen TPU components."
  • Arista Networks Medium confidence
    Buys AI datacenter Ethernet switches from Arista Networks.
    Confirms KO ITWorld Published Aug 1, 2025: "Meta builds AI clusters on the 7700R4."
    Confirms EN Arista Published Feb 18, 2025: "2024 revenue $7B (+19.5%); cloud titans 48%."
    Confirms EN SEC EDGAR Published Feb 18, 2025: "10-K: Ethernet-switch revenue concentrated in MS and Meta."

Value-chain ripple (tiers 2–3)

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

Beneficiaries (payment chain)

← All companiesMarkdown · JSON

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

ValueChain.wikiCompany value-chain data