{
 "_note": "The industry as a cross-section: 13 strata from bedrock energy to the model labs at the surface. Names, tickers and roles are descriptive; every bottleneck flag carries a figure that traces to a SOURCES.md row (flag_row) or is explicitly press-only (flag_row null → REPORTED chip). Compiled 2026-09-02.",
 "as_of": "2026-09-24",
 "layers": [
  {
   "id": "L01",
   "order": 1,
   "name": "Energy generation",
   "role": "Gas, nuclear, geothermal and renewables — the electrons everything else waits for.",
   "hue": "#ffa24d",
   "bottleneck": {
    "status": "tight",
    "flag": "116 GW GEV backlog",
    "flag_row": 40,
    "note": "GE Vernova gas-turbine backlog + reservations 116 GW (Jul 2026), taking reservations for 2031; Siemens Energy 95 GW; MHI 35 GW with 2028–30 delivery. New firm generation is sold out for years."
   },
   "players": [
    {
     "name": "Constellation",
     "ticker": "CEG",
     "role": "nuclear fleet, PPAs"
    },
    {
     "name": "Vistra",
     "ticker": "VST",
     "role": "gas + nuclear IPP"
    },
    {
     "name": "NRG",
     "ticker": "NRG",
     "role": "gas IPP, hyperscaler CCGT"
    },
    {
     "name": "Talen",
     "ticker": "TLN",
     "role": "Susquehanna nuclear → Amazon"
    },
    {
     "name": "GE Vernova",
     "ticker": "GEV",
     "role": "gas turbines"
    },
    {
     "name": "Siemens Energy",
     "ticker": "ENR.DE",
     "role": "gas turbines"
    },
    {
     "name": "Mitsubishi Heavy",
     "ticker": "7011.T",
     "role": "large-frame turbines"
    },
    {
     "name": "Bloom Energy",
     "ticker": "BE",
     "role": "on-site fuel cells"
    },
    {
     "name": "Oklo",
     "ticker": "OKLO",
     "role": "SMR developer"
    },
    {
     "name": "NuScale",
     "ticker": "SMR",
     "role": "SMR developer"
    },
    {
     "name": "Fervo",
     "ticker": null,
     "private": true,
     "role": "enhanced geothermal"
    }
   ],
   "money_in": "Long-dated PPAs and capacity payments from operators (L12) and utilities (L02).",
   "money_out": "Turbines, fuel, reactors, drilling.",
   "desk_basket": "power",
   "explorer": "flow",
   "essay": "essay.html#power"
  },
  {
   "id": "L02",
   "order": 2,
   "name": "Grid & interconnection",
   "role": "Utilities, ISOs and the queue: getting a substation energized is now the critical path.",
   "hue": "#ffcf6b",
   "bottleneck": {
    "status": "choke",
    "flag": "3–5 yr substations",
    "flag_row": 4,
    "note": "High-voltage substation lead times of 3–5 years; PJM cleared its 2028/29 auction at the $325/MW-day cap and still fell 6.8 GW short (№38); ERCOT’s large-load queue is 474.7 GW, 90% data centers, and paused for a state audit (№39)."
   },
   "players": [
    {
     "name": "PJM",
     "ticker": null,
     "private": true,
     "role": "largest US ISO"
    },
    {
     "name": "ERCOT",
     "ticker": null,
     "private": true,
     "role": "Texas grid operator"
    },
    {
     "name": "Dominion",
     "ticker": "D",
     "role": "Northern Virginia utility"
    },
    {
     "name": "NextEra",
     "ticker": "NEE",
     "role": "largest US utility"
    },
    {
     "name": "Southern Co",
     "ticker": "SO",
     "role": "Georgia utility"
    },
    {
     "name": "AEP",
     "ticker": "AEP",
     "role": "Ohio/Texas transmission"
    },
    {
     "name": "Duke",
     "ticker": "DUK",
     "role": "Carolinas utility"
    },
    {
     "name": "Quanta",
     "ticker": "PWR",
     "role": "T&D construction"
    },
    {
     "name": "MasTec",
     "ticker": "MTZ",
     "role": "transmission EPC"
    }
   ],
   "money_in": "Interconnection deposits, large-load tariffs, transmission riders paid by operators (L12).",
   "money_out": "Transformers, switchgear and line crews (L03, L06).",
   "desk_basket": "power",
   "explorer": "flow",
   "essay": "essay.html#queue"
  },
  {
   "id": "L03",
   "order": 3,
   "name": "Electrical equipment",
   "role": "Transformers, switchgear, UPS and busway — the hardware queue.",
   "hue": "#ffcf6b",
   "bottleneck": {
    "status": "choke",
    "flag": "128 wk transformers",
    "flag_row": 3,
    "note": "Standard power transformers average 128 weeks; large units run 80–210 weeks. You cannot expedite the queue — factory capacity is the constraint."
   },
   "players": [
    {
     "name": "Eaton",
     "ticker": "ETN",
     "role": "switchgear, UPS, busway"
    },
    {
     "name": "Schneider Electric",
     "ticker": "SU.PA",
     "role": "power distribution"
    },
    {
     "name": "Siemens",
     "ticker": "SIE.DE",
     "role": "transformers, switchgear"
    },
    {
     "name": "Hitachi Energy",
     "ticker": "6501.T",
     "role": "large transformers"
    },
    {
     "name": "ABB",
     "ticker": "ABBN.SW",
     "role": "switchgear"
    },
    {
     "name": "Vertiv",
     "ticker": "VRT",
     "role": "UPS, busway, PDUs"
    },
    {
     "name": "Powell",
     "ticker": "POWL",
     "role": "switchgear"
    },
    {
     "name": "Hubbell",
     "ticker": "HUBB",
     "role": "grid components"
    },
    {
     "name": "Prolec/Xignux",
     "ticker": null,
     "private": true,
     "role": "transformers"
    }
   ],
   "money_in": "Electrical packages ~40–50% of construction cost, ordered by GCs and operators (L04, L12).",
   "money_out": "Grain-oriented steel, copper, skilled assembly.",
   "desk_basket": "build_cool",
   "explorer": "blueprint",
   "essay": "essay.html#queue"
  },
  {
   "id": "L04",
   "order": 4,
   "name": "Land, shells & construction",
   "role": "Developers, REITs, general contractors and EPCs — the buildings with power in them.",
   "hue": "#e6d27a",
   "bottleneck": {
    "status": "tight",
    "flag": "5 of 16 GW building",
    "flag_row": 14,
    "note": "~16 GW announced for 2026, ~5 GW actually under construction; Sightline expects 30–50% of the pipeline to slip."
   },
   "players": [
    {
     "name": "Equinix",
     "ticker": "EQIX",
     "role": "colocation REIT"
    },
    {
     "name": "Digital Realty",
     "ticker": "DLR",
     "role": "hyperscale REIT"
    },
    {
     "name": "QTS",
     "ticker": null,
     "private": true,
     "role": "Blackstone-owned developer"
    },
    {
     "name": "Vantage",
     "ticker": null,
     "private": true,
     "role": "hyperscale developer"
    },
    {
     "name": "Aligned",
     "ticker": null,
     "private": true,
     "role": "developer"
    },
    {
     "name": "Crusoe",
     "ticker": null,
     "private": true,
     "role": "energy-first developer (Abilene)"
    },
    {
     "name": "Turner",
     "ticker": null,
     "private": true,
     "role": "general contractor"
    },
    {
     "name": "DPR",
     "ticker": null,
     "private": true,
     "role": "general contractor"
    },
    {
     "name": "Mortenson",
     "ticker": null,
     "private": true,
     "role": "general contractor"
    },
    {
     "name": "Comfort Systems",
     "ticker": "FIX",
     "role": "mechanical/electrical"
    },
    {
     "name": "EMCOR",
     "ticker": "EME",
     "role": "electrical contractor"
    },
    {
     "name": "Sterling",
     "ticker": "STRL",
     "role": "site work"
    },
    {
     "name": "Jacobs",
     "ticker": "J",
     "role": "design/engineering"
    }
   ],
   "money_in": "Leases and build-to-suit contracts from operators (L12); construction loans and SPVs (L07).",
   "money_out": "Electrical gear (L03), cooling (L05), labor (L06), land and permits.",
   "desk_basket": "build_cool",
   "explorer": "types",
   "essay": "essay.html#slippage"
  },
  {
   "id": "L05",
   "order": 5,
   "name": "Cooling & thermal",
   "role": "Chillers, CDUs, liquid loops and the water bill — density decides the design.",
   "hue": "#43e0ff",
   "bottleneck": {
    "status": "loose",
    "flag": "PUE 1.52 · slow decline",
    "flag_row": 55,
    "note": "Uptime’s 2026 survey average is 1.52, capacity-weighted 1.36: big new sites pull the number down while the long tail stays flat. Liquid cooling is the lever as rack density passes 100 kW."
   },
   "players": [
    {
     "name": "Vertiv",
     "ticker": "VRT",
     "role": "CDUs, thermal management"
    },
    {
     "name": "Schneider Electric",
     "ticker": "SU.PA",
     "role": "cooling systems"
    },
    {
     "name": "Johnson Controls",
     "ticker": "JCI",
     "role": "chillers"
    },
    {
     "name": "Trane",
     "ticker": "TT",
     "role": "chillers"
    },
    {
     "name": "Modine",
     "ticker": "MOD",
     "role": "data-centre cooling"
    },
    {
     "name": "nVent",
     "ticker": "NVT",
     "role": "liquid cooling"
    },
    {
     "name": "CoolIT",
     "ticker": null,
     "private": true,
     "role": "direct-to-chip"
    },
    {
     "name": "LiquidStack",
     "ticker": null,
     "private": true,
     "role": "immersion"
    }
   ],
   "money_in": "Mechanical packages from GCs and operators (L04, L12).",
   "money_out": "Compressors, pumps, water.",
   "desk_basket": "build_cool",
   "explorer": "blueprint",
   "essay": null
  },
  {
   "id": "L06",
   "order": 6,
   "name": "Skilled labor",
   "role": "Electricians, pipefitters, linemen, commissioning engineers — the binding human constraint.",
   "hue": "#3ff0b0",
   "bottleneck": {
    "status": "choke",
    "flag": "300–500K short",
    "flag_row": 18,
    "note": "Microsoft estimates ~500K new electricians needed over the decade; BLS (Aug 2026) projects 821K → 897K by 2035 with ~72.7K openings a year; wages in data-center markets are rising 6.6–9.9% y/y vs 2.4% economy-wide (№49). A journeyman still takes 4–5 years to make."
   },
   "players": [
    {
     "name": "IBEW",
     "ticker": null,
     "private": true,
     "role": "electrical workers' union"
    },
    {
     "name": "NECA",
     "ticker": null,
     "private": true,
     "role": "electrical contractors"
    },
    {
     "name": "ABC",
     "ticker": null,
     "private": true,
     "role": "open-shop builders"
    },
    {
     "name": "Siemens USA training",
     "ticker": "SIE.DE",
     "role": "200K by 2030 pledge"
    },
    {
     "name": "Google / ETA",
     "ticker": "GOOGL",
     "role": "100K workers, 30K apprentices"
    },
    {
     "name": "Aerotek / Allegis",
     "ticker": null,
     "private": true,
     "role": "staffing"
    },
    {
     "name": "PE-rolled MEP firms",
     "ticker": null,
     "private": true,
     "role": "~800 acquired since 2022"
    }
   ],
   "money_in": "Wages with a 25–30% data-centre premium, paid by contractors (L04).",
   "money_out": "Training, apprenticeships.",
   "desk_basket": null,
   "explorer": "workforce",
   "essay": "essay.html#labor"
  },
  {
   "id": "L07",
   "order": 7,
   "name": "Financing & ownership",
   "role": "Who actually owns the buildings and the chips: private credit, SPVs, sovereigns, REITs.",
   "hue": "#9b8cff",
   "bottleneck": {
    "status": "loose",
    "flag": "$720–745B '26 capex",
    "flag_row": 34,
    "note": "Four companies guide $720–745B of 2026 capex and the bond market just absorbed ~$100B of IG AI paper in twelve weeks (№45) — capital is not the constraint. The new wrinkle is where it sits: Meta’s El Paso venture puts $12.5B of debt and a ~$13B residual-value guarantee off balance sheet (№47), and the neocloud credit ladder runs from Baa2 at SOFR+300 to 9.6% unsecured (№46)."
   },
   "players": [
    {
     "name": "Blackstone",
     "ticker": "BX",
     "role": "QTS, AirTrunk"
    },
    {
     "name": "Brookfield",
     "ticker": "BAM",
     "role": "infrastructure funds"
    },
    {
     "name": "Blue Owl",
     "ticker": "OWL",
     "role": "Hyperion SPV"
    },
    {
     "name": "KKR",
     "ticker": "KKR",
     "role": "infrastructure"
    },
    {
     "name": "DigitalBridge",
     "ticker": "DBRG",
     "role": "Vantage, Switch"
    },
    {
     "name": "Apollo",
     "ticker": "APO",
     "role": "private credit"
    },
    {
     "name": "MGX / G42",
     "ticker": null,
     "private": true,
     "role": "Abu Dhabi sovereign"
    },
    {
     "name": "PIF / Humain",
     "ticker": null,
     "private": true,
     "role": "Saudi sovereign"
    },
    {
     "name": "BlackRock",
     "ticker": "BLK",
     "role": "El Paso 1 GW venture (80/20 with Meta)"
    },
    {
     "name": "SpaceX (xAI)",
     "ticker": "SPCX",
     "role": "$85.7B IPO · $25B bonds"
    }
   ],
   "money_in": "LP capital, bonds, sovereign wealth.",
   "money_out": "Construction loans, GPU-backed debt, equity stakes in operators and labs (L12, L13).",
   "desk_basket": "hyperscale",
   "explorer": "markets",
   "essay": "essay.html#capex"
  },
  {
   "id": "L08",
   "order": 8,
   "name": "Semiconductor fabs & equipment",
   "role": "Leading-edge foundry, advanced packaging and the tools that make them.",
   "hue": "#43e0ff",
   "bottleneck": {
    "status": "tight",
    "flag": "CoWoS-bound",
    "flag_row": null,
    "note": "Advanced packaging (CoWoS) capacity at TSMC has been the reported gating factor on accelerator supply; figure not yet in the ledger — shown as REPORTED."
   },
   "players": [
    {
     "name": "TSMC",
     "ticker": "TSM",
     "role": "leading-edge foundry, CoWoS"
    },
    {
     "name": "Samsung Foundry",
     "ticker": "005930.KS",
     "role": "foundry"
    },
    {
     "name": "Intel Foundry",
     "ticker": "INTC",
     "role": "foundry"
    },
    {
     "name": "ASML",
     "ticker": "ASML",
     "role": "EUV lithography"
    },
    {
     "name": "Applied Materials",
     "ticker": "AMAT",
     "role": "deposition/etch"
    },
    {
     "name": "Lam Research",
     "ticker": "LRCX",
     "role": "etch"
    },
    {
     "name": "KLA",
     "ticker": "KLAC",
     "role": "inspection"
    },
    {
     "name": "Amkor",
     "ticker": "AMKR",
     "role": "OSAT packaging"
    }
   ],
   "money_in": "Wafer and packaging orders from chip designers (L10).",
   "money_out": "Tools (ASML, AMAT), fabs, energy.",
   "desk_basket": "compute",
   "explorer": "trends",
   "essay": null
  },
  {
   "id": "L09",
   "order": 9,
   "name": "Memory & storage",
   "role": "HBM, DRAM and NAND — the scarce input of 2026.",
   "hue": "#43e0ff",
   "bottleneck": {
    "status": "tight",
    "flag": "+13–18% QoQ DRAM",
    "flag_row": 33,
    "note": "TrendForce: conventional and server DRAM contract prices +13–18% QoQ in 3Q26; HBM rising to 30% of DRAM wafer input by 2027; Micron at an 85% gross margin."
   },
   "players": [
    {
     "name": "SK Hynix",
     "ticker": "000660.KS",
     "role": "HBM leader"
    },
    {
     "name": "Samsung",
     "ticker": "005930.KS",
     "role": "HBM, DRAM"
    },
    {
     "name": "Micron",
     "ticker": "MU",
     "role": "HBM3E/HBM4, DRAM"
    },
    {
     "name": "SanDisk",
     "ticker": "SNDK",
     "role": "NAND"
    },
    {
     "name": "Western Digital",
     "ticker": "WDC",
     "role": "HDD"
    },
    {
     "name": "Seagate",
     "ticker": "STX",
     "role": "HDD"
    },
    {
     "name": "Pure Storage",
     "ticker": "PSTG",
     "role": "flash arrays"
    }
   ],
   "money_in": "Component orders from server makers and hyperscalers (L11, L12).",
   "money_out": "Fab capex, wafers.",
   "desk_basket": "compute",
   "explorer": "blueprint",
   "essay": null
  },
  {
   "id": "L10",
   "order": 10,
   "name": "Accelerators & networking",
   "role": "GPUs, custom silicon, switches and optics — where most of the dollars land.",
   "hue": "#43e0ff",
   "bottleneck": {
    "status": "tight",
    "flag": ">90% NVIDIA",
    "flag_row": 5,
    "note": "NVIDIA holds more than 90% of data-centre AI GPUs (analyst range 80–94%) and allocates supply; custom silicon and AMD are the pressure valve."
   },
   "players": [
    {
     "name": "NVIDIA",
     "ticker": "NVDA",
     "role": "GPUs, NVLink, InfiniBand"
    },
    {
     "name": "AMD",
     "ticker": "AMD",
     "role": "MI-series GPUs"
    },
    {
     "name": "Broadcom",
     "ticker": "AVGO",
     "role": "custom ASICs (TPU), switches"
    },
    {
     "name": "Marvell",
     "ticker": "MRVL",
     "role": "custom silicon, optics"
    },
    {
     "name": "Arista",
     "ticker": "ANET",
     "role": "data-centre switching"
    },
    {
     "name": "Credo",
     "ticker": "CRDO",
     "role": "AECs, optical DSPs"
    },
    {
     "name": "Astera Labs",
     "ticker": "ALAB",
     "role": "connectivity"
    },
    {
     "name": "Coherent",
     "ticker": "COHR",
     "role": "optical transceivers"
    },
    {
     "name": "Google TPU",
     "ticker": "GOOGL",
     "role": "in-house accelerator"
    },
    {
     "name": "Amazon Trainium",
     "ticker": "AMZN",
     "role": "in-house accelerator"
    },
    {
     "name": "Cerebras",
     "ticker": "CBRS",
     "role": "wafer-scale (newly listed)"
    }
   ],
   "money_in": "Accelerator and network purchases — 40–60% of an AI build's cost (L11, L12).",
   "money_out": "Wafers and packaging (L08), HBM (L09).",
   "desk_basket": "compute",
   "explorer": "functions",
   "essay": null
  },
  {
   "id": "L11",
   "order": 11,
   "name": "Servers & racks",
   "role": "Integration: boards, racks, NVL72 systems, the ODMs behind the brands.",
   "hue": "#7fb8ff",
   "bottleneck": {
    "status": "loose",
    "flag": "40–60% of build",
    "flag_row": 6,
    "note": "Chips and servers are 40–60% of all-in AI build cost; integration itself is not the bottleneck — its inputs are."
   },
   "players": [
    {
     "name": "Supermicro",
     "ticker": "SMCI",
     "role": "AI servers"
    },
    {
     "name": "Dell",
     "ticker": "DELL",
     "role": "AI servers"
    },
    {
     "name": "HPE",
     "ticker": "HPE",
     "role": "AI servers, Cray"
    },
    {
     "name": "Foxconn",
     "ticker": "2317.TW",
     "role": "ODM, GB200 racks"
    },
    {
     "name": "Quanta",
     "ticker": "2382.TW",
     "role": "ODM"
    },
    {
     "name": "Wiwynn",
     "ticker": "6669.TW",
     "role": "ODM"
    },
    {
     "name": "Celestica",
     "ticker": "CLS",
     "role": "networking hardware"
    }
   ],
   "money_in": "System orders from operators (L12).",
   "money_out": "GPUs, memory, networking (L10, L09).",
   "desk_basket": "compute",
   "explorer": "blueprint",
   "essay": null
  },
  {
   "id": "L12",
   "order": 12,
   "name": "Operators — hyperscale, neocloud, colo",
   "role": "The buyers of everything below and the sellers of compute to everything above.",
   "hue": "#9b8cff",
   "bottleneck": {
    "status": "choke",
    "flag": "$678B RPO · $104B CRWV",
    "flag_row": 35,
    "note": "Microsoft commercial RPO $678B (+84%), AWS $496B, Google Cloud $514B, Oracle $664B (Q1 FY27, №51); CoreWeave’s backlog is $104B (+246%) with 1.5 GW active (№46). Hood in July: demand continues to exceed available capacity. Nscale’s S-1 (18 Sep) confirms Anthropic agreements worth up to ~$44.6B for Vera Rubin capacity at its Monarch, West Virginia campus (№53); the Microsoft 1.35 GW LOI story remains press-only (№48)."
   },
   "players": [
    {
     "name": "Microsoft",
     "ticker": "MSFT",
     "role": "Azure, Fairwater"
    },
    {
     "name": "Amazon",
     "ticker": "AMZN",
     "role": "AWS, Rainier"
    },
    {
     "name": "Google",
     "ticker": "GOOGL",
     "role": "Cloud, TPU fleet"
    },
    {
     "name": "Meta",
     "ticker": "META",
     "role": "Hyperion, Prometheus"
    },
    {
     "name": "Oracle",
     "ticker": "ORCL",
     "role": "OCI, Stargate"
    },
    {
     "name": "CoreWeave",
     "ticker": "CRWV",
     "role": "neocloud"
    },
    {
     "name": "Nebius",
     "ticker": "NBIS",
     "role": "neocloud"
    },
    {
     "name": "IREN",
     "ticker": "IREN",
     "role": "miner → AI hosting"
    },
    {
     "name": "Applied Digital",
     "ticker": "APLD",
     "role": "HPC hosting"
    },
    {
     "name": "Crusoe",
     "ticker": null,
     "private": true,
     "role": "neocloud, Abilene"
    },
    {
     "name": "Lambda",
     "ticker": null,
     "private": true,
     "role": "neocloud"
    },
    {
     "name": "xAI",
     "ticker": null,
     "private": true,
     "role": "Colossus (self-build)"
    },
    {
     "name": "Equinix",
     "ticker": "EQIX",
     "role": "colocation"
    },
    {
     "name": "Digital Realty",
     "ticker": "DLR",
     "role": "colocation"
    }
   ],
   "money_in": "Cloud and API revenue from labs and enterprises (L13); equity and debt (L07).",
   "money_out": "Capex $650–700B guided for 2026 — servers, chips, shells, power (L11, L10, L04, L01).",
   "desk_basket": "hyperscale",
   "explorer": "markets",
   "essay": "essay.html#the-wall"
  },
  {
   "id": "L13",
   "order": 13,
   "name": "Model labs & applications",
   "role": "The surface: frontier labs and the applications whose revenue has to pay for all of it.",
   "hue": "#c9a0ff",
   "bottleneck": {
    "status": "loose",
    "flag": "$0.96 /Mtok market",
    "flag_row": 52,
    "note": "Silicon Data’s usage-weighted token price was $0.96/Mtok on 22 Sep 2026 (from $1.62 in early July). The mainstream flagships were cut 20–50% that day while the top tier (GPT-6 Astra $10/$50, GPT-5.5-Pro $30/$180) kept climbing — demand is elastic, margins are not guaranteed."
   },
   "players": [
    {
     "name": "OpenAI",
     "ticker": null,
     "private": true,
     "role": "frontier lab (S-1 filed)"
    },
    {
     "name": "Anthropic",
     "ticker": null,
     "private": true,
     "role": "frontier lab (S-1 filed)"
    },
    {
     "name": "Google DeepMind",
     "ticker": "GOOGL",
     "role": "frontier lab"
    },
    {
     "name": "xAI",
     "ticker": null,
     "private": true,
     "role": "frontier lab (SpaceX)"
    },
    {
     "name": "Meta",
     "ticker": "META",
     "role": "open-weights lab"
    },
    {
     "name": "Mistral",
     "ticker": null,
     "private": true,
     "role": "European lab"
    },
    {
     "name": "DeepSeek",
     "ticker": null,
     "private": true,
     "role": "Chinese lab, price setter"
    },
    {
     "name": "Enterprise & consumer apps",
     "ticker": null,
     "private": true,
     "role": "the demand"
    }
   ],
   "money_in": "Subscriptions, API tokens, enterprise contracts.",
   "money_out": "Compute purchases from operators (L12); training runs.",
   "desk_basket": null,
   "explorer": "trends",
   "essay": "tokenomics.html#twotrack"
  }
 ]
}