Updates

Per-node maturity is computed live from the database. Below are the update history and the next expansion schedule.

Node status

21 nodes · T4 16 · T3 3 · T2 2

Update history

• Jun 20, 2026Amazon expansion (14th): the Trainium demand hub — all four hyperscalers doneDoneDemand-end node on FY2025 revenue $716.9B, AWS $129B, cash capex $131.8B (+59%, ~$200B for 2026). Five cost edges (Nvidia 196k units, TSMC for Trainium2/3, Marvell design partner, SK HBM3e, Astera) and two revenue edges (AWS with Anthropic Project Rainier 500k chips, e-commerce). Back-filled Nvidia/Marvell reverse links and created TSMC/SK/Astera reciprocals. Re-verified Marvell to FY2026 ($8.195B, DC >$6B). T4 (13/13). All four hubs (Microsoft, Alphabet, Meta, Amazon) complete.
• Jun 20, 2026Meta expansion (13th): the MTIA demand hubDoneDemand-end node on FY2025 revenue $201.0B and cash capex $69.7B (+87%). Six cost edges (Nvidia 224k units, AMD MI300X for Llama, Broadcom MTIA co-development, TSMC 5nm, SK HBM, Astera) and two revenue edges (ads, Reality Labs). Back-filled reverse links on Nvidia/AMD/Broadcom and a TSMC revenue edge. Re-verified AMD to Q1 FY2026 (data center $5.8B, +57%). T4 (13/13).
• Jun 20, 2026Alphabet expansion (12th): the TPU demand hubDoneDemand-end node on FY2025 revenue $402.8B and cash capex $91.4B (+74%, 2026 guided $175-185B). Seven cost edges (Broadcom TPU co-design, TSMC fab, Nvidia 169k units, SK HBM3E, Marvell, Intel/Samsung foundry diversification) and two revenue edges (cloud, ads). Back-filled reverse links on Broadcom/Nvidia/Marvell/SK hynix and revenue edges on TSMC/Intel/Samsung. T4 (13/13).
• Jun 20, 2026Microsoft expansion (11th): the AI-chip demand hubDoneAdded the demand-end node on FY2025 revenue of $281.7B and cash capex of $64.6B (+45%). Seven cost edges (Nvidia — its largest Hopper buyer at 485k units in 2024 — plus AMD, TSMC for its own Maia chip, Astera, Intel) and two revenue edges (OpenAI, total demand). Closed the graph by back-filling reverse links on Nvidia/AMD/Marvell and revenue edges on TSMC/Intel. T4 (13/13).
• Jun 20, 2026Samsung Electro-Mechanics expansion (10th)DoneFC-BGA substrate supplier for AI accelerators. All six revenue edges (NVIDIA, AMD, Broadcom, Apple, Marvell, Samsung) are existing nodes, making it highly connective; reverse NVIDIA/AMD to SEMCO substrate-purchase edges added. DART primary filing (revenue ₩11.3T). T3.
• Jun 20, 2026Astera Labs expansion (9th) — demand layer doneDoneAI server interconnect. Aries and Scorpio adopted into NVIDIA Blackwell (linked to the NVIDIA node), TSMC sole foundry. T3. This completes the chip-design and demand layer (NVIDIA, AMD, Broadcom, Intel, Marvell, Apple, Astera).
• Jun 20, 2026Apple expansion (8th)DoneThe demand end of the chain. Cost edges to TSMC (processor), Samsung/SK/Micron (memory) and Broadcom (wireless) all link to nodes, closing five loops; reverse Broadcom to Apple edge added. TSMC second-largest customer (17%). T4.
• Jun 20, 2026Marvell expansion (7th)DoneDesign partner for Amazon Trainium; customers Microsoft and Google. Cost edges to TSMC (N2, CoWoS) and Samsung/SK (custom HBM) link to nodes, with reverse Samsung/SK to Marvell edges added. T4.
• Jun 20, 2026ASML expansion (6th)DoneThe monopoly EUV supplier. Customers TSMC, Samsung, SK, Intel and Micron are all existing nodes, so one node closes five cost loops. Suppliers Carl Zeiss (sole-source optics) and Cymer. T4. Revenue €32.7B; Samsung and SK took two-thirds of 2025 EUV shipments.
• Jun 20, 2026Intel expansion (5th)DoneA dual node: a foundry and also a TSMC customer. Outsources some tiles to TSMC (linked both ways), first to adopt ASML High-NA EUV. Customers Dell, Lenovo, HP. KO/EN cross-verification. T4.
• Jun 20, 2026Broadcom expansion (4th)DoneCo-designs Google TPU; customers Meta and Anthropic. Cost edges to TSMC, SK, Samsung and Micron link to existing nodes; also added the reverse SK to Broadcom HBM edge. KO/EN cross-verification. T4.
• Jun 20, 2026AMD expansion (3rd)DoneCost edges to TSMC, Samsung and Micron link back to existing nodes, closing three loops. Customers Microsoft, Meta, Oracle, OpenAI. Korean/English cross-verification (Korea Economic Daily, The Elec + AMD 10-K). T4.
• Jun 20, 2026TSMC expansion (2nd)DoneCustomers NVIDIA, Apple, AMD, Broadcom plus suppliers ASML, Applied Materials, Lam. Confirmed via 20-F that NVIDIA is TSMC's largest customer at 19% ($23.2B), and back-filled that figure onto the existing NVIDIA node's cost edge (two-way cross-verification).
• Jun 20, 2026NVIDIA node added (first expansion)DoneNVIDIA's cost side (HBM purchases) links back to the SK, Samsung and Micron nodes, closing the graph. Reaches T4.
• Jun 20, 2026Korean/English bilingual + source translationDoneCompany names, descriptions, notes, financials, metrics, source excerpts and publisher names are all bilingual. Machine-readable .md/.json/llms.txt provided.
• Jun 20, 2026HBM equipment cluster addedDoneAdded ASMPT, Hanwha Semitech and BESI as nodes to track HBM4 TC bonder share.
• Jun 20, 2026Micron and Hanmi added — 4 core nodes at T4DoneThe three memory makers (SK, Samsung, Micron) plus back-end equipment (Hanmi) all reach top maturity (T4).
• Jun 19, 2026SK hynix and Samsung seed nodes builtDoneFirst two nodes of the HBM/AI memory chain, with revenue/cost edges and multi-source cross-verification.

Schedule

• Jun 30, 2026Korean materials (Dongjin Semichem, etc.)PlannedSemiconductor materials such as photoresist and etchant. As Samsung/SK suppliers they close the memory loop, with rich DART and Korean sources.
• Jun 30, 2026Foundry and OSAT (GlobalFoundries, ASE, Amkor)PlannedOutsourced back-end packaging and test; closes fabless cost edges.
• Jun 30, 2026Materials, substrates, wafersPlannedDongjin Semichem, Soulbrain, Ibiden, Shin-Etsu, SUMCO and others.
• Jun 30, 2026Promote ASMPT and BESI to T3PlannedAdd HKEX and Euronext filings to fill cost edges and cross-verification.