{
 "name": "AI Lattice",
 "description": "A sourced map of who depends on whom in the AI economy, from power plants to model labs, with the chokepoints marked.",
 "version": "0.1.0",
 "as_of": "2026-10-01",
 "status": "published",
 "publisher": {
  "name": "exmxc.ai",
  "url": "https://www.exmxc.ai",
  "founder": "Mike Ye"
 },
 "framework": {
  "name": "The Four Forces of AI Power",
  "url": "https://www.exmxc.ai/frameworks/four-forces-of-ai-power"
 },
 "companion_datasets": [
  {
   "name": "Strategic Signal",
   "url": "https://strategicsignal.ai/"
  }
 ],
 "rules": [
  "Every link between two entities cites a dated public source. Links the source implies but does not state are marked basis=inferred.",
  "Observed facts and interpretation are stored in separate fields and never mixed.",
  "Each quarterly version is kept; links are added, changed or retired, never silently edited."
 ],
 "license": "CC BY 4.0, cite as: AI Lattice (exmxc.ai), version and date",
 "forces": [
  {
   "id": "compute",
   "name": "Compute",
   "definition": "The physical capacity to train and deploy intelligence.",
   "state_2026": "Consolidating into competing, vertically integrated camps: Nvidia's, Google's TPU camp, Amazon's Trainium camp, and China's Huawei stack.",
   "binding_constraint": "Packaging and memory, not chip design. TSMC's packaging lines and three memory makers decide how many AI chips ship."
  },
  {
   "id": "interface",
   "name": "Interface",
   "definition": "The layer that controls human interaction and belief.",
   "state_2026": "Expanding from chat windows into operating systems, coding tools and devices.",
   "binding_constraint": "Default placement. Google now powers Siri on top of its own search and Android, and ChatGPT holds the largest standalone audience."
  },
  {
   "id": "alignment",
   "name": "Alignment",
   "definition": "The governance system that decides permissible behavior.",
   "state_2026": "Now includes permission to sell. A US model was switched off worldwide for 18 days in June 2026 by a Commerce letter.",
   "binding_constraint": "Government permission. Export licensing, Pentagon procurement status and EU enforcement now decide market access for frontier models."
  },
  {
   "id": "energy",
   "name": "Energy",
   "definition": "The resource that powers the exponential.",
   "state_2026": "Moving from grid access toward owned, firm power: on-site gas, restarted nuclear plants, fuel cells.",
   "binding_constraint": "Turbines, transformers and grid connections. Chips are arriving faster than buildings can be powered."
  }
 ],
 "layers": [
  {
   "id": "energy",
   "name": "Power",
   "force": "energy",
   "order": 0,
   "description": "Electricity to run the machines: power plants, turbines, grid connections and the electrical gear between them."
  },
  {
   "id": "tools",
   "name": "Chipmaking tools",
   "force": "compute",
   "order": 1,
   "description": "The machines that make chips. A few suppliers, years of lead time."
  },
  {
   "id": "fab",
   "name": "Chip factories and packaging",
   "force": "compute",
   "order": 2,
   "description": "Foundries that print the chips, and the packaging lines that stitch chips and memory together."
  },
  {
   "id": "memory",
   "name": "Memory",
   "force": "compute",
   "order": 3,
   "description": "High-bandwidth memory (HBM) stacked beside each AI chip. Sold out in advance."
  },
  {
   "id": "chips",
   "name": "AI chips and networking",
   "force": "compute",
   "order": 4,
   "description": "The accelerators that do the math, and the switches and lasers that connect them."
  },
  {
   "id": "cloud",
   "name": "Data centers and cloud",
   "force": "compute",
   "order": 5,
   "description": "The companies that buy the chips, build the buildings and rent out the compute."
  },
  {
   "id": "models",
   "name": "Model labs",
   "force": "interface",
   "order": 6,
   "description": "The labs that train frontier models and sell access to them."
  },
  {
   "id": "interface",
   "name": "Distribution and data",
   "force": "interface",
   "order": 7,
   "description": "The screens, devices and content that put models in front of people."
  },
  {
   "id": "rules",
   "name": "Rules and permissions",
   "force": "alignment",
   "order": 8,
   "description": "The governments and regulators that decide what can be sold, to whom, and where it can be built."
  }
 ],
 "entities": [
  {
   "id": "ge-vernova",
   "name": "GE Vernova",
   "layer": "energy",
   "kind": "company",
   "ticker": "GEV",
   "role": "Largest gas turbine maker; 116 GW backlog including reservations.",
   "forces": [
    "energy"
   ],
   "url": "https://ailattice.ai/e/ge-vernova/"
  },
  {
   "id": "siemens-energy",
   "name": "Siemens Energy",
   "layer": "energy",
   "kind": "company",
   "ticker": "ENR",
   "role": "Gas turbines and grid equipment; 69 GW turbine backlog, lead times three years or more.",
   "forces": [
    "energy"
   ],
   "url": "https://ailattice.ai/e/siemens-energy/"
  },
  {
   "id": "mitsubishi-power",
   "name": "Mitsubishi Heavy (Power)",
   "layer": "energy",
   "kind": "company",
   "ticker": "7011",
   "role": "Gas turbines; 35 GW backlog, new orders deliver 2028 to 2030.",
   "forces": [
    "energy"
   ],
   "url": "https://ailattice.ai/e/mitsubishi-power/"
  },
  {
   "id": "constellation",
   "name": "Constellation Energy",
   "layer": "energy",
   "kind": "company",
   "ticker": "CEG",
   "role": "Largest US nuclear fleet operator; 20-year contracts with Microsoft, Meta and Amazon.",
   "forces": [
    "energy"
   ],
   "url": "https://ailattice.ai/e/constellation/"
  },
  {
   "id": "vistra",
   "name": "Vistra",
   "layer": "energy",
   "kind": "company",
   "ticker": "VST",
   "role": "Nuclear and gas generator; 20-year nuclear contracts with Meta and Amazon.",
   "forces": [
    "energy"
   ],
   "url": "https://ailattice.ai/e/vistra/"
  },
  {
   "id": "talen",
   "name": "Talen Energy",
   "layer": "energy",
   "kind": "company",
   "ticker": "TLN",
   "role": "Owns Susquehanna nuclear plant; 1.92 GW contract with Amazon through 2042.",
   "forces": [
    "energy"
   ],
   "url": "https://ailattice.ai/e/talen/"
  },
  {
   "id": "terrapower",
   "name": "TerraPower",
   "layer": "energy",
   "kind": "company",
   "ticker": null,
   "role": "Advanced reactor developer (Natrium); up to 2.8 GW option with Meta.",
   "forces": [
    "energy"
   ],
   "url": "https://ailattice.ai/e/terrapower/"
  },
  {
   "id": "oklo",
   "name": "Oklo",
   "layer": "energy",
   "kind": "company",
   "ticker": "OKLO",
   "role": "Small reactor developer; 1.2 GW campus plan with Meta, first phase around 2030.",
   "forces": [
    "energy"
   ],
   "url": "https://ailattice.ai/e/oklo/"
  },
  {
   "id": "entergy",
   "name": "Entergy",
   "layer": "energy",
   "kind": "company",
   "ticker": "ETR",
   "role": "Louisiana utility building gas plants to serve Meta's Hyperion campus.",
   "forces": [
    "energy"
   ],
   "url": "https://ailattice.ai/e/entergy/"
  },
  {
   "id": "bloom",
   "name": "Bloom Energy",
   "layer": "energy",
   "kind": "company",
   "ticker": "BE",
   "role": "On-site fuel cells; up to 2.8 GW agreement with Oracle.",
   "forces": [
    "energy"
   ],
   "url": "https://ailattice.ai/e/bloom/"
  },
  {
   "id": "eaton",
   "name": "Eaton",
   "layer": "energy",
   "kind": "company",
   "ticker": "ETN",
   "role": "Switchgear, power distribution and liquid cooling for data centers.",
   "forces": [
    "energy"
   ],
   "url": "https://ailattice.ai/e/eaton/"
  },
  {
   "id": "pjm",
   "name": "PJM grid",
   "layer": "energy",
   "kind": "grid",
   "ticker": null,
   "role": "Largest US grid region (13 states plus DC); 2028/29 capacity auction cleared at its price cap and still fell short.",
   "forces": [
    "energy"
   ],
   "url": "https://ailattice.ai/e/pjm/"
  },
  {
   "id": "ercot",
   "name": "ERCOT (Texas grid)",
   "layer": "energy",
   "kind": "grid",
   "ticker": null,
   "role": "Texas grid; about 474 GW of connection requests, roughly 90% from data centers; large-load study paused.",
   "forces": [
    "energy"
   ],
   "url": "https://ailattice.ai/e/ercot/"
  },
  {
   "id": "asml",
   "name": "ASML",
   "layer": "tools",
   "kind": "company",
   "ticker": "ASML",
   "role": "Sole maker of EUV lithography machines needed for leading-edge chips.",
   "forces": [
    "compute"
   ],
   "url": "https://ailattice.ai/e/asml/"
  },
  {
   "id": "applied-materials",
   "name": "Applied Materials",
   "layer": "tools",
   "kind": "company",
   "ticker": "AMAT",
   "role": "Deposition, etch and packaging equipment for logic and memory makers.",
   "forces": [
    "compute"
   ],
   "url": "https://ailattice.ai/e/applied-materials/"
  },
  {
   "id": "lam",
   "name": "Lam Research",
   "layer": "tools",
   "kind": "company",
   "ticker": "LRCX",
   "role": "Etch and deposition tools; packaging tools up more than 70% in 2026.",
   "forces": [
    "compute"
   ],
   "url": "https://ailattice.ai/e/lam/"
  },
  {
   "id": "tokyo-electron",
   "name": "Tokyo Electron",
   "layer": "tools",
   "kind": "company",
   "ticker": "8035",
   "role": "Etch, coating and packaging tools; China share of revenue fell to 26.8%.",
   "forces": [
    "compute"
   ],
   "url": "https://ailattice.ai/e/tokyo-electron/"
  },
  {
   "id": "tsmc",
   "name": "TSMC",
   "layer": "fab",
   "kind": "company",
   "ticker": "TSM",
   "role": "Makes nearly every leading AI accelerator and runs CoWoS, the dominant AI chip packaging line.",
   "forces": [
    "compute"
   ],
   "url": "https://ailattice.ai/e/tsmc/"
  },
  {
   "id": "samsung-foundry",
   "name": "Samsung Foundry",
   "layer": "fab",
   "kind": "company",
   "ticker": "005930",
   "role": "Second leading-edge foundry; Tesla AI5 on 2nm at Taylor, Texas.",
   "forces": [
    "compute"
   ],
   "url": "https://ailattice.ai/e/samsung-foundry/"
  },
  {
   "id": "intel",
   "name": "Intel",
   "layer": "fab",
   "kind": "company",
   "ticker": "INTC",
   "role": "Chipmaker and foundry; 18A in production with ASML High-NA; Nvidia owns a $5B stake.",
   "forces": [
    "compute"
   ],
   "url": "https://ailattice.ai/e/intel/"
  },
  {
   "id": "smic",
   "name": "SMIC",
   "layer": "fab",
   "kind": "company",
   "ticker": "0981",
   "role": "China's leading foundry; makes Huawei's Ascend chips; ran at 93.7% utilization in Q2 2026.",
   "forces": [
    "compute"
   ],
   "url": "https://ailattice.ai/e/smic/"
  },
  {
   "id": "amkor",
   "name": "Amkor",
   "layer": "fab",
   "kind": "company",
   "ticker": "AMKR",
   "role": "Chip packaging; building Nvidia-funded capacity in Peoria, Arizona.",
   "forces": [
    "compute"
   ],
   "url": "https://ailattice.ai/e/amkor/"
  },
  {
   "id": "ase",
   "name": "ASE / SPIL",
   "layer": "fab",
   "kind": "company",
   "ticker": "ASX",
   "role": "Largest outside packaging house; now taking part of TSMC's CoWoS work.",
   "forces": [
    "compute"
   ],
   "url": "https://ailattice.ai/e/ase/"
  },
  {
   "id": "sk-hynix",
   "name": "SK hynix",
   "layer": "memory",
   "kind": "company",
   "ticker": "000660",
   "role": "Leading HBM supplier; about 60 to 70% of Nvidia's HBM4 volume; 76% operating margin.",
   "forces": [
    "compute"
   ],
   "url": "https://ailattice.ai/e/sk-hynix/"
  },
  {
   "id": "samsung-memory",
   "name": "Samsung Memory",
   "layer": "memory",
   "kind": "company",
   "ticker": "005930",
   "role": "HBM4 qualified at Nvidia and AMD in 2026; sole HBM4 source for AMD's MI455X.",
   "forces": [
    "compute"
   ],
   "url": "https://ailattice.ai/e/samsung-memory/"
  },
  {
   "id": "micron",
   "name": "Micron",
   "layer": "memory",
   "kind": "company",
   "ticker": "MU",
   "role": "US memory maker; $54.2B quarter; $100B of minimum memory commitments through 2030.",
   "forces": [
    "compute"
   ],
   "url": "https://ailattice.ai/e/micron/"
  },
  {
   "id": "cxmt",
   "name": "CXMT",
   "layer": "memory",
   "kind": "company",
   "ticker": null,
   "role": "China's DRAM maker, about 10% share; sampling HBM3E to Chinese chip designers.",
   "forces": [
    "compute"
   ],
   "url": "https://ailattice.ai/e/cxmt/"
  },
  {
   "id": "nvidia",
   "name": "Nvidia",
   "layer": "chips",
   "kind": "company",
   "ticker": "NVDA",
   "role": "Dominant AI accelerator maker; Vera Rubin in full production; Q2 FY27 revenue $96.2B.",
   "forces": [
    "compute"
   ],
   "url": "https://ailattice.ai/e/nvidia/"
  },
  {
   "id": "amd",
   "name": "AMD",
   "layer": "chips",
   "kind": "company",
   "ticker": "AMD",
   "role": "Second GPU supplier; 6 GW deals each with OpenAI and Meta; Helios racks shipping.",
   "forces": [
    "compute"
   ],
   "url": "https://ailattice.ai/e/amd/"
  },
  {
   "id": "broadcom",
   "name": "Broadcom",
   "layer": "chips",
   "kind": "company",
   "ticker": "AVGO",
   "role": "Designs custom AI chips for Google, Meta, OpenAI and Anthropic; AI revenue $16.7B a quarter.",
   "forces": [
    "compute"
   ],
   "url": "https://ailattice.ai/e/broadcom/"
  },
  {
   "id": "marvell",
   "name": "Marvell",
   "layer": "chips",
   "kind": "company",
   "ticker": "MRVL",
   "role": "Custom chip partner across Google, AWS and Microsoft programs.",
   "forces": [
    "compute"
   ],
   "url": "https://ailattice.ai/e/marvell/"
  },
  {
   "id": "huawei",
   "name": "Huawei (Ascend)",
   "layer": "chips",
   "kind": "company",
   "ticker": null,
   "role": "China's main AI chip supplier; about 1.5M Ascend chips expected in 2026.",
   "forces": [
    "compute"
   ],
   "url": "https://ailattice.ai/e/huawei/"
  },
  {
   "id": "cerebras",
   "name": "Cerebras",
   "layer": "chips",
   "kind": "company",
   "ticker": null,
   "role": "Wafer-scale inference chips; 750 MW deal with OpenAI through 2028.",
   "forces": [
    "compute"
   ],
   "url": "https://ailattice.ai/e/cerebras/"
  },
  {
   "id": "lumentum",
   "name": "Lumentum",
   "layer": "chips",
   "kind": "company",
   "ticker": "LITE",
   "role": "Lasers for optical links; only volume shipper of 200G-per-lane lasers.",
   "forces": [
    "compute"
   ],
   "url": "https://ailattice.ai/e/lumentum/"
  },
  {
   "id": "coherent",
   "name": "Coherent",
   "layer": "chips",
   "kind": "company",
   "ticker": "COHR",
   "role": "Lasers and optical modules; took about $2B from Nvidia for priority capacity.",
   "forces": [
    "compute"
   ],
   "url": "https://ailattice.ai/e/coherent/"
  },
  {
   "id": "microsoft",
   "name": "Microsoft",
   "layer": "cloud",
   "kind": "company",
   "ticker": "MSFT",
   "role": "Azure passed $100B a year; owns about 27% of OpenAI; $678B contracted backlog.",
   "forces": [
    "compute",
    "interface"
   ],
   "url": "https://ailattice.ai/e/microsoft/"
  },
  {
   "id": "amazon",
   "name": "Amazon (AWS)",
   "layer": "cloud",
   "kind": "company",
   "ticker": "AMZN",
   "role": "Largest cloud; Trainium chips; investor in both Anthropic and OpenAI.",
   "forces": [
    "compute"
   ],
   "url": "https://ailattice.ai/e/amazon/"
  },
  {
   "id": "google",
   "name": "Google (Alphabet)",
   "layer": "cloud",
   "kind": "company",
   "ticker": "GOOGL",
   "role": "Cloud plus its own TPU chips, now rented to Anthropic and Meta; Gemini app at 950M monthly users.",
   "forces": [
    "compute",
    "interface"
   ],
   "url": "https://ailattice.ai/e/google/"
  },
  {
   "id": "oracle",
   "name": "Oracle",
   "layer": "cloud",
   "kind": "company",
   "ticker": "ORCL",
   "role": "Builds Stargate capacity for OpenAI; $664B backlog, negative free cash flow.",
   "forces": [
    "compute"
   ],
   "url": "https://ailattice.ai/e/oracle/"
  },
  {
   "id": "coreweave",
   "name": "CoreWeave",
   "layer": "cloud",
   "kind": "company",
   "ticker": "CRWV",
   "role": "Largest neocloud; $104B backlog; customers include Meta and OpenAI.",
   "forces": [
    "compute"
   ],
   "url": "https://ailattice.ai/e/coreweave/"
  },
  {
   "id": "nebius",
   "name": "Nebius",
   "layer": "cloud",
   "kind": "company",
   "ticker": "NBIS",
   "role": "Neocloud with multi-year contracts from Microsoft and Meta.",
   "forces": [
    "compute"
   ],
   "url": "https://ailattice.ai/e/nebius/"
  },
  {
   "id": "crusoe",
   "name": "Crusoe",
   "layer": "cloud",
   "kind": "company",
   "ticker": null,
   "role": "Builds and runs the Abilene, Texas Stargate campus; valued at $30B.",
   "forces": [
    "compute",
    "energy"
   ],
   "url": "https://ailattice.ai/e/crusoe/"
  },
  {
   "id": "fluidstack",
   "name": "Fluidstack",
   "layer": "cloud",
   "kind": "company",
   "ticker": null,
   "role": "Partner in Anthropic's $50B data center build-out; Google backstops its power-site leases.",
   "forces": [
    "compute"
   ],
   "url": "https://ailattice.ai/e/fluidstack/"
  },
  {
   "id": "alibaba-cloud",
   "name": "Alibaba Cloud",
   "layer": "cloud",
   "kind": "company",
   "ticker": "BABA",
   "role": "China's largest cloud; trains Qwen; capex up 75% in its latest quarter.",
   "forces": [
    "compute"
   ],
   "url": "https://ailattice.ai/e/alibaba-cloud/"
  },
  {
   "id": "openai",
   "name": "OpenAI",
   "layer": "models",
   "kind": "lab",
   "ticker": null,
   "role": "Makes ChatGPT; raised $110B in February 2026; buys compute and chips from Microsoft, Oracle, Amazon, CoreWeave, AMD and Cerebras.",
   "forces": [
    "interface",
    "compute"
   ],
   "url": "https://ailattice.ai/e/openai/"
  },
  {
   "id": "anthropic",
   "name": "Anthropic",
   "layer": "models",
   "kind": "lab",
   "ticker": null,
   "role": "Claude models; $965B valuation at its May 2026 raise; compute from Google, Amazon, Microsoft and SpaceX.",
   "forces": [
    "interface",
    "alignment"
   ],
   "url": "https://ailattice.ai/e/anthropic/"
  },
  {
   "id": "meta",
   "name": "Meta",
   "layer": "models",
   "kind": "lab",
   "ticker": "META",
   "role": "Builds its own campuses and rents from Google, CoreWeave and Nebius; 2026 capex $135 to 145B.",
   "forces": [
    "interface",
    "compute",
    "energy"
   ],
   "url": "https://ailattice.ai/e/meta/"
  },
  {
   "id": "spacexai",
   "name": "SpaceXAI (xAI)",
   "layer": "models",
   "kind": "lab",
   "ticker": null,
   "role": "SpaceX merged with xAI in February; owns Colossus and, since August, Cursor.",
   "forces": [
    "compute",
    "interface"
   ],
   "url": "https://ailattice.ai/e/spacexai/"
  },
  {
   "id": "deepseek",
   "name": "DeepSeek",
   "layer": "models",
   "kind": "lab",
   "ticker": null,
   "role": "Chinese open-weight lab; serving on Huawei Ascend chips.",
   "forces": [
    "interface"
   ],
   "url": "https://ailattice.ai/e/deepseek/"
  },
  {
   "id": "moonshot",
   "name": "Moonshot (Kimi)",
   "layer": "models",
   "kind": "lab",
   "ticker": null,
   "role": "Chinese lab; raised $2B at $20B; Kimi is the base of Cursor's own model.",
   "forces": [
    "interface"
   ],
   "url": "https://ailattice.ai/e/moonshot/"
  },
  {
   "id": "mistral",
   "name": "Mistral",
   "layer": "models",
   "kind": "lab",
   "ticker": null,
   "role": "European lab; raised €3B at €21B to build 1 GW of European compute.",
   "forces": [
    "interface",
    "alignment"
   ],
   "url": "https://ailattice.ai/e/mistral/"
  },
  {
   "id": "apple",
   "name": "Apple",
   "layer": "interface",
   "kind": "company",
   "ticker": "AAPL",
   "role": "Siri in iOS 27 runs on a custom Google Gemini model.",
   "forces": [
    "interface"
   ],
   "url": "https://ailattice.ai/e/apple/"
  },
  {
   "id": "tesla",
   "name": "Tesla",
   "layer": "interface",
   "kind": "company",
   "ticker": "TSLA",
   "role": "Designs its own AI5 chip for cars and robots, made by Samsung and TSMC.",
   "forces": [
    "compute",
    "interface"
   ],
   "url": "https://ailattice.ai/e/tesla/"
  },
  {
   "id": "cursor",
   "name": "Cursor",
   "layer": "interface",
   "kind": "company",
   "ticker": null,
   "role": "AI coding tool; acquired by SpaceX for $60B, closed August 2026.",
   "forces": [
    "interface"
   ],
   "url": "https://ailattice.ai/e/cursor/"
  },
  {
   "id": "reddit",
   "name": "Reddit",
   "layer": "interface",
   "kind": "company",
   "ticker": "RDDT",
   "role": "Licenses its posts to Google and OpenAI for model training.",
   "forces": [
    "interface"
   ],
   "url": "https://ailattice.ai/e/reddit/"
  },
  {
   "id": "news-corp",
   "name": "News Corp",
   "layer": "interface",
   "kind": "company",
   "ticker": "NWSA",
   "role": "Licenses WSJ and New York Post archives to Meta.",
   "forces": [
    "interface"
   ],
   "url": "https://ailattice.ai/e/news-corp/"
  },
  {
   "id": "us-commerce",
   "name": "US Commerce (BIS)",
   "layer": "rules",
   "kind": "government",
   "ticker": null,
   "role": "Controls chip and model exports; required licenses for Anthropic's Fable 5 and Mythos 5 from June 12 until June 30.",
   "forces": [
    "alignment"
   ],
   "url": "https://ailattice.ai/e/us-commerce/"
  },
  {
   "id": "pentagon",
   "name": "Pentagon",
   "layer": "rules",
   "kind": "government",
   "ticker": null,
   "role": "Labeled Anthropic a supply chain risk; label upheld 2-1 on appeal Sept 25.",
   "forces": [
    "alignment"
   ],
   "url": "https://ailattice.ai/e/pentagon/"
  },
  {
   "id": "ferc",
   "name": "FERC",
   "layer": "rules",
   "kind": "government",
   "ticker": null,
   "role": "Federal power regulator; ordered six grid operators to rewrite large-load rules.",
   "forces": [
    "alignment",
    "energy"
   ],
   "url": "https://ailattice.ai/e/ferc/"
  },
  {
   "id": "new-york",
   "name": "New York State",
   "layer": "rules",
   "kind": "government",
   "ticker": null,
   "role": "First statewide moratorium on new hyperscale data centers, up to one year.",
   "forces": [
    "alignment",
    "energy"
   ],
   "url": "https://ailattice.ai/e/new-york/"
  },
  {
   "id": "eu-ai-office",
   "name": "EU AI Office",
   "layer": "rules",
   "kind": "government",
   "ticker": null,
   "role": "Enforcement against general-purpose model providers began Aug 2, 2026.",
   "forces": [
    "alignment"
   ],
   "url": "https://ailattice.ai/e/eu-ai-office/"
  },
  {
   "id": "china-gov",
   "name": "China regulators",
   "layer": "rules",
   "kind": "government",
   "ticker": null,
   "role": "Blocked Nvidia H200 imports at customs despite US approval.",
   "forces": [
    "alignment"
   ],
   "url": "https://ailattice.ai/e/china-gov/"
  }
 ],
 "links": [
  {
   "id": "L001",
   "from": "asml",
   "to": "tsmc",
   "kind": "tools",
   "what": "EUV lithography",
   "evidence": "Taiwan took 30% of ASML's Q2 2026 machine shipments.",
   "basis": "reported",
   "source": {
    "key": "asml_q2",
    "title": "ASML Q2 2026 results slides",
    "publisher": "Investing.com",
    "url": "https://www.investing.com/news/company-news/asml-q2-2026-slides-outlook-raised-on-ai-demand-high-na-milestone-93CH-4792166",
    "date": "2026-07-15",
    "tier": "press"
   }
  },
  {
   "id": "L002",
   "from": "asml",
   "to": "sk-hynix",
   "kind": "tools",
   "what": "EUV lithography for memory",
   "evidence": "South Korea took 43% of Q2 shipments; ASML guides memory revenue up 75%.",
   "basis": "reported",
   "source": {
    "key": "asml_q2",
    "title": "ASML Q2 2026 results slides",
    "publisher": "Investing.com",
    "url": "https://www.investing.com/news/company-news/asml-q2-2026-slides-outlook-raised-on-ai-demand-high-na-milestone-93CH-4792166",
    "date": "2026-07-15",
    "tier": "press"
   }
  },
  {
   "id": "L003",
   "from": "asml",
   "to": "samsung-memory",
   "kind": "tools",
   "what": "EUV lithography for memory",
   "evidence": "South Korea took 43% of Q2 shipments; ASML guides memory revenue up 75%.",
   "basis": "reported",
   "source": {
    "key": "asml_q2",
    "title": "ASML Q2 2026 results slides",
    "publisher": "Investing.com",
    "url": "https://www.investing.com/news/company-news/asml-q2-2026-slides-outlook-raised-on-ai-demand-high-na-milestone-93CH-4792166",
    "date": "2026-07-15",
    "tier": "press"
   }
  },
  {
   "id": "L004",
   "from": "asml",
   "to": "intel",
   "kind": "tools",
   "what": "High-NA EUV",
   "evidence": "High-NA EUV now in production use on Intel's 18A process.",
   "basis": "reported",
   "source": {
    "key": "asml_q2",
    "title": "ASML Q2 2026 results slides",
    "publisher": "Investing.com",
    "url": "https://www.investing.com/news/company-news/asml-q2-2026-slides-outlook-raised-on-ai-demand-high-na-milestone-93CH-4792166",
    "date": "2026-07-15",
    "tier": "press"
   }
  },
  {
   "id": "L005",
   "from": "asml",
   "to": "smic",
   "kind": "tools",
   "what": "Older DUV lithography",
   "evidence": "China was 14% of Q2 shipments and about 20% of sales.",
   "basis": "reported",
   "source": {
    "key": "asml_q2",
    "title": "ASML Q2 2026 results slides",
    "publisher": "Investing.com",
    "url": "https://www.investing.com/news/company-news/asml-q2-2026-slides-outlook-raised-on-ai-demand-high-na-milestone-93CH-4792166",
    "date": "2026-07-15",
    "tier": "press"
   }
  },
  {
   "id": "L006",
   "from": "applied-materials",
   "to": "tsmc",
   "kind": "tools",
   "what": "Deposition and packaging tools",
   "evidence": "Q4 FY26 revenue guide about $10.3B, driven by AI logic and memory.",
   "basis": "inferred",
   "source": {
    "key": "amat_q",
    "title": "Applied Materials Q3 FY26",
    "publisher": "BigGo Finance",
    "url": "https://finance.biggo.com/news/80dc4fdb-2a02-4112-8b43-9eab6aa47b69",
    "date": "2026-08-14",
    "tier": "secondary"
   }
  },
  {
   "id": "L007",
   "from": "lam",
   "to": "micron",
   "kind": "tools",
   "what": "Etch and deposition tools",
   "evidence": "Lam's NAND tool revenue doubled quarter on quarter; packaging revenue guided up more than 70%.",
   "basis": "inferred",
   "source": {
    "key": "lam_q",
    "title": "Lam Research quarterly results",
    "publisher": "BigGo Finance",
    "url": "https://finance.biggo.com/news/US_LRCX_2026-07-29",
    "date": "2026-07-29",
    "tier": "secondary"
   }
  },
  {
   "id": "L008",
   "from": "tokyo-electron",
   "to": "sk-hynix",
   "kind": "tools",
   "what": "Etch, coating and packaging tools",
   "evidence": "Packaging tool sales guided up 60%; China share fell to 26.8%.",
   "basis": "inferred",
   "source": {
    "key": "tel_fy",
    "title": "Tokyo Electron FY2026 results",
    "publisher": "NineScrolls",
    "url": "https://ninescrolls.com/news/tokyo-electron-posts-record-jpy-2-44-trillion-fy2026-guides-first-half-fy2027",
    "date": "2026-04-30",
    "tier": "secondary"
   }
  },
  {
   "id": "L009",
   "from": "tsmc",
   "to": "nvidia",
   "kind": "manufacturing",
   "what": "Blackwell and Rubin chips",
   "evidence": "Analysts estimate Nvidia passed Apple as TSMC's largest customer in 2026, about 22% of revenue.",
   "basis": "reported",
   "source": {
    "key": "nvda_tsmc_top",
    "title": "Nvidia replaces Apple as TSMC's biggest customer",
    "publisher": "MacRumors",
    "url": "https://www.macrumors.com/2026/01/28/nvidia-replaces-apple-as-biggest-tsmc-customer/",
    "date": "2026-01-28",
    "tier": "press"
   }
  },
  {
   "id": "L010",
   "from": "tsmc",
   "to": "nvidia",
   "kind": "packaging",
   "what": "CoWoS packaging capacity",
   "evidence": "Nvidia reportedly holds more than half of TSMC's 2026 CoWoS capacity.",
   "basis": "reported",
   "source": {
    "key": "cowos_osat",
    "title": "TSMC expands outsourcing of CoWoS front-end step to OSATs",
    "publisher": "TrendForce",
    "url": "https://www.trendforce.com/news/2026/08/05/news-tsmc-reportedly-expands-outsourcing-of-key-cowos-front-end-step-to-osats-amid-rising-nvidia-asic-demand/",
    "date": "2026-08-05",
    "tier": "press"
   }
  },
  {
   "id": "L011",
   "from": "tsmc",
   "to": "amd",
   "kind": "manufacturing",
   "what": "2nm server CPUs and MI455X",
   "evidence": "AMD Venice is the first x86 server CPU on TSMC 2nm.",
   "basis": "reported",
   "source": {
    "key": "amd_moor",
    "title": "AMD Advancing AI 2026: Helios ships",
    "publisher": "Moor Insights & Strategy",
    "url": "https://moorinsightsstrategy.com/field-notes/amd-advancing-ai-2026-helios-ships-venice-swings-and-the-software-moat-narrows/",
    "date": "2026-07-24",
    "tier": "press"
   }
  },
  {
   "id": "L012",
   "from": "tsmc",
   "to": "broadcom",
   "kind": "manufacturing",
   "what": "Custom AI chips on 3nm",
   "evidence": "Google's TPU v7 is made on TSMC 3nm; Meta's MTIA is reportedly moving to 3nm.",
   "basis": "reported",
   "source": {
    "key": "asic_tsmc",
    "title": "Custom AI ASICs examined, from Broadcom to MTIA",
    "publisher": "Tom's Hardware",
    "url": "https://www.tomshardware.com/tech-industry/semiconductors/custom-ai-asics-examined-from-broadcom-to-mtia",
    "date": "2026-05",
    "tier": "press"
   }
  },
  {
   "id": "L013",
   "from": "tsmc",
   "to": "google",
   "kind": "manufacturing",
   "what": "8th-generation TPUs on 2nm",
   "evidence": "Both 8th-gen TPU variants target late 2027 on TSMC 2nm.",
   "basis": "reported",
   "source": {
    "key": "tpu8",
    "title": "Google Ironwood and 8th-gen TPUs",
    "publisher": "The Next Web",
    "url": "https://thenextweb.com/news/google-ironwood-tpu-inference-cloud-next",
    "date": "2026-04-22",
    "tier": "press"
   }
  },
  {
   "id": "L014",
   "from": "samsung-foundry",
   "to": "tesla",
   "kind": "manufacturing",
   "what": "Tesla AI5 chip on 2nm",
   "evidence": "Tesla AI5 taped out at Samsung's Taylor, Texas fab; volume in 2H27.",
   "basis": "reported",
   "source": {
    "key": "tesla_ai5",
    "title": "Samsung Taylor fab tapes out Tesla AI5 on 2nm",
    "publisher": "Electrek",
    "url": "https://electrek.co/2026/07/13/samsung-taylor-fab-tesla-ai5-chip-2nm/",
    "date": "2026-07-13",
    "tier": "press"
   }
  },
  {
   "id": "L015",
   "from": "smic",
   "to": "huawei",
   "kind": "manufacturing",
   "what": "Ascend 950 chips",
   "evidence": "SMIC's N+3 process makes the Ascend 950 dies; Huawei plans about 1.5M AI chips in 2026.",
   "basis": "reported",
   "source": {
    "key": "huawei_960",
    "title": "Huawei pulls Ascend 960DT forward; demand outstrips capacity",
    "publisher": "Traders Agency",
    "url": "https://tradersagency.com/blog/huawei-pulls-ascend-960dt-forward-to-q1-2027-says-ai-chip-demand-outstrips-its-capacity",
    "date": "2026-09-17",
    "tier": "secondary"
   }
  },
  {
   "id": "L016",
   "from": "nvidia",
   "to": "intel",
   "kind": "capital",
   "what": "$5B equity stake",
   "evidence": "Nvidia bought $5B of Intel stock; Intel will design x86 CPUs linked to NVLink.",
   "basis": "reported",
   "source": {
    "key": "nvda_intel",
    "title": "Intel to design CPUs with Nvidia NVLink for $5B investment",
    "publisher": "Network World",
    "url": "https://www.networkworld.com/article/4059448/intel-will-design-cpus-with-nvidia-nvlink-in-return-for-5-billion-investment.html",
    "date": "2025-09",
    "tier": "press"
   }
  },
  {
   "id": "L017",
   "from": "tsmc",
   "to": "ase",
   "kind": "packaging",
   "what": "Outsourced CoWoS front-end step",
   "evidence": "TSMC reportedly outsources the chip-on-wafer step to ASE.",
   "basis": "reported",
   "source": {
    "key": "cowos_osat",
    "title": "TSMC expands outsourcing of CoWoS front-end step to OSATs",
    "publisher": "TrendForce",
    "url": "https://www.trendforce.com/news/2026/08/05/news-tsmc-reportedly-expands-outsourcing-of-key-cowos-front-end-step-to-osats-amid-rising-nvidia-asic-demand/",
    "date": "2026-08-05",
    "tier": "press"
   }
  },
  {
   "id": "L018",
   "from": "ase",
   "to": "amd",
   "kind": "packaging",
   "what": "Packaging capacity",
   "evidence": "AMD providing strategic backing to ASE and SPIL capacity.",
   "basis": "reported",
   "source": {
    "key": "cowos_osat",
    "title": "TSMC expands outsourcing of CoWoS front-end step to OSATs",
    "publisher": "TrendForce",
    "url": "https://www.trendforce.com/news/2026/08/05/news-tsmc-reportedly-expands-outsourcing-of-key-cowos-front-end-step-to-osats-amid-rising-nvidia-asic-demand/",
    "date": "2026-08-05",
    "tier": "press"
   }
  },
  {
   "id": "L019",
   "from": "amkor",
   "to": "nvidia",
   "kind": "packaging",
   "what": "US packaging capacity",
   "evidence": "About $1.5B multi-year prepayment to expand Amkor's Peoria, Arizona site.",
   "basis": "reported",
   "source": {
    "key": "amkor_az",
    "title": "Nvidia GB300 at TSMC Arizona; Amkor packaging",
    "publisher": "ExplainX",
    "url": "https://explainx.ai/blog/nvidia-gb300-tsmc-arizona-amkor-packaging-july-2026",
    "date": "2026-07-27",
    "tier": "secondary"
   }
  },
  {
   "id": "L020",
   "from": "sk-hynix",
   "to": "nvidia",
   "kind": "memory",
   "what": "HBM4 for Vera Rubin",
   "evidence": "Roughly 60 to 70% of Nvidia's HBM4 volume.",
   "basis": "reported",
   "source": {
    "key": "hbm4_suppliers",
    "title": "Vera Rubin enters full production; HBM4 suppliers named",
    "publisher": "Tech Times",
    "url": "https://www.techtimes.com/articles/317539/20260602/nvidia-vera-rubin-enters-full-production-samsung-sk-hynix-micron-named-hbm4-suppliers.htm",
    "date": "2026-06-02",
    "tier": "press"
   }
  },
  {
   "id": "L021",
   "from": "samsung-memory",
   "to": "nvidia",
   "kind": "memory",
   "what": "HBM4 for Vera Rubin",
   "evidence": "An estimated 25 to 30% of volume.",
   "basis": "reported",
   "source": {
    "key": "hbm4_suppliers",
    "title": "Vera Rubin enters full production; HBM4 suppliers named",
    "publisher": "Tech Times",
    "url": "https://www.techtimes.com/articles/317539/20260602/nvidia-vera-rubin-enters-full-production-samsung-sk-hynix-micron-named-hbm4-suppliers.htm",
    "date": "2026-06-02",
    "tier": "press"
   }
  },
  {
   "id": "L022",
   "from": "micron",
   "to": "nvidia",
   "kind": "memory",
   "what": "HBM4 for Vera Rubin",
   "evidence": "Micron says HBM4 is shipping in volume for Vera Rubin.",
   "basis": "reported",
   "source": {
    "key": "micron_q4",
    "title": "Micron posts record $54B quarter, locks $100B memory contracts",
    "publisher": "Tech Times",
    "url": "https://www.techtimes.com/articles/328376/20261001/micron-posts-record-54b-quarter-locks-hyperscalers-100b-memory-contracts.htm",
    "date": "2026-10-01",
    "tier": "press"
   }
  },
  {
   "id": "L023",
   "from": "samsung-memory",
   "to": "amd",
   "kind": "memory",
   "what": "HBM4 for MI455X",
   "evidence": "AMD secures Samsung HBM4 for the MI455X.",
   "basis": "reported",
   "source": {
    "key": "amd_samsung_hbm",
    "title": "AMD secures Samsung HBM4 for MI455X",
    "publisher": "TrendForce",
    "url": "https://www.trendforce.com/news/2026/03/19/news-amd-secures-samsung-hbm4-for-mi455x-deal-may-tie-partial-ai-chip-shift-to-samsung-foundry/",
    "date": "2026-03-19",
    "tier": "press"
   }
  },
  {
   "id": "L024",
   "from": "cxmt",
   "to": "huawei",
   "kind": "memory",
   "what": "Domestic HBM",
   "evidence": "Only about 240k of 1.5M Ascend chips use domestic HBM.",
   "basis": "reported",
   "source": {
    "key": "huawei_960",
    "title": "Huawei pulls Ascend 960DT forward; demand outstrips capacity",
    "publisher": "Traders Agency",
    "url": "https://tradersagency.com/blog/huawei-pulls-ascend-960dt-forward-to-q1-2027-says-ai-chip-demand-outstrips-its-capacity",
    "date": "2026-09-17",
    "tier": "secondary"
   }
  },
  {
   "id": "L025",
   "from": "cxmt",
   "to": "alibaba-cloud",
   "kind": "memory",
   "what": "HBM3E samples",
   "evidence": "Sampling HBM3E to Alibaba's T-Head chip unit.",
   "basis": "reported",
   "source": {
    "key": "cxmt",
    "title": "CXMT hits 10% DRAM share",
    "publisher": "Tech Times",
    "url": "https://www.techtimes.com/articles/326406/20260903/cxmt-hits-10-dram-share-samsungs-hbm-pivot-handed-china-market-vacancy.htm",
    "date": "2026-09-03",
    "tier": "press"
   }
  },
  {
   "id": "L026",
   "from": "broadcom",
   "to": "google",
   "kind": "chips",
   "what": "TPU design and supply",
   "evidence": "Design agreement for Google's TPUs runs through 2031.",
   "basis": "reported",
   "source": {
    "key": "marvell_google",
    "title": "Marvell attaches across Google's TPU stack",
    "publisher": "Futurum Group",
    "url": "https://futurumgroup.com/insights/marvell-attaches-across-googles-tpu-stack-with-a-warrant-vesting-toward-120b/",
    "date": "2026-08",
    "tier": "press"
   }
  },
  {
   "id": "L027",
   "from": "marvell",
   "to": "google",
   "kind": "chips",
   "what": "Custom chips across the TPU system",
   "evidence": "Google holds a warrant for about 7% of Marvell, vesting with custom revenue.",
   "basis": "reported",
   "source": {
    "key": "marvell_google",
    "title": "Marvell attaches across Google's TPU stack",
    "publisher": "Futurum Group",
    "url": "https://futurumgroup.com/insights/marvell-attaches-across-googles-tpu-stack-with-a-warrant-vesting-toward-120b/",
    "date": "2026-08",
    "tier": "press"
   }
  },
  {
   "id": "L028",
   "from": "marvell",
   "to": "amazon",
   "kind": "chips",
   "what": "Trainium programs",
   "evidence": "Marvell says it has signed programs at all three big US clouds.",
   "basis": "reported",
   "source": {
    "key": "marvell_google",
    "title": "Marvell attaches across Google's TPU stack",
    "publisher": "Futurum Group",
    "url": "https://futurumgroup.com/insights/marvell-attaches-across-googles-tpu-stack-with-a-warrant-vesting-toward-120b/",
    "date": "2026-08",
    "tier": "press"
   }
  },
  {
   "id": "L029",
   "from": "marvell",
   "to": "microsoft",
   "kind": "chips",
   "what": "Maia programs",
   "evidence": "Marvell says it has signed programs at all three big US clouds.",
   "basis": "reported",
   "source": {
    "key": "marvell_google",
    "title": "Marvell attaches across Google's TPU stack",
    "publisher": "Futurum Group",
    "url": "https://futurumgroup.com/insights/marvell-attaches-across-googles-tpu-stack-with-a-warrant-vesting-toward-120b/",
    "date": "2026-08",
    "tier": "press"
   }
  },
  {
   "id": "L030",
   "from": "broadcom",
   "to": "openai",
   "kind": "chips",
   "what": "Jalapeño inference chip",
   "evidence": "OpenAI's first custom inference chip, rolling into data centers by year-end.",
   "basis": "reported",
   "source": {
    "key": "jalapeno",
    "title": "OpenAI unveils Jalapeño inference chip with Broadcom",
    "publisher": "VentureBeat",
    "url": "https://venturebeat.com/infrastructure/openai-unveils-first-custom-ai-inference-chip-jalapeno-with-broadcom-and-its-development-was-sped-up-with-openais-own-models",
    "date": "2026-06-24",
    "tier": "press"
   }
  },
  {
   "id": "L031",
   "from": "broadcom",
   "to": "meta",
   "kind": "chips",
   "what": "MTIA custom chips",
   "evidence": "Meta named as a custom-chip customer on Broadcom's Q3 FY26 call.",
   "basis": "reported",
   "source": {
    "key": "avgo_q3",
    "title": "Broadcom Q3 FY2026: AI revenue $16.7B",
    "publisher": "Tech Insider",
    "url": "https://tech-insider.org/broadcom-q3-fy2026-ai-chip-revenue-16-7-billion/",
    "date": "2026-09",
    "tier": "secondary"
   }
  },
  {
   "id": "L032",
   "from": "broadcom",
   "to": "anthropic",
   "kind": "capital",
   "what": "Up to $42B of chip financing",
   "evidence": "Convertible financing against a five-year TPU lease.",
   "basis": "reported",
   "source": {
    "key": "avgo_anth_fin",
    "title": "Broadcom lines up $42B in chip financing for Anthropic",
    "publisher": "Tradingpedia",
    "url": "https://www.tradingpedia.com/2026/10/01/broadcom-lines-up-42b-in-chip-financing-for-anthropic/",
    "date": "2026-10-01",
    "tier": "secondary"
   }
  },
  {
   "id": "L033",
   "from": "amd",
   "to": "openai",
   "kind": "chips",
   "what": "MI450 / Helios racks, 6 GW",
   "evidence": "Revenue shipments start end of Q3 2026.",
   "basis": "reported",
   "source": {
    "key": "amd_moor",
    "title": "AMD Advancing AI 2026: Helios ships",
    "publisher": "Moor Insights & Strategy",
    "url": "https://moorinsightsstrategy.com/field-notes/amd-advancing-ai-2026-helios-ships-venice-swings-and-the-software-moat-narrows/",
    "date": "2026-07-24",
    "tier": "press"
   }
  },
  {
   "id": "L034",
   "from": "amd",
   "to": "meta",
   "kind": "chips",
   "what": "GPUs, up to 6 GW",
   "evidence": "Expanded partnership announced Feb 24, 2026.",
   "basis": "reported",
   "source": {
    "key": "amd_meta",
    "title": "AMD and Meta announce expanded strategic partnership",
    "publisher": "AMD",
    "url": "https://www.amd.com/en/newsroom/press-releases/2026-2-24-amd-and-meta-announce-expanded-strategic-partnersh.html",
    "date": "2026-02-24",
    "tier": "primary"
   }
  },
  {
   "id": "L035",
   "from": "cerebras",
   "to": "openai",
   "kind": "compute",
   "what": "750 MW of inference compute",
   "evidence": "More than $10B through 2028.",
   "basis": "reported",
   "source": {
    "key": "cerebras_oai",
    "title": "OpenAI signs compute deal with Cerebras",
    "publisher": "TechCrunch",
    "url": "https://techcrunch.com/2026/01/14/openai-signs-deal-reportedly-worth-10-billion-for-compute-from-cerebras/",
    "date": "2026-01-14",
    "tier": "press"
   }
  },
  {
   "id": "L036",
   "from": "nvidia",
   "to": "coreweave",
   "kind": "chips",
   "what": "Vera Rubin systems",
   "evidence": "Named as a Vera Rubin ramping partner.",
   "basis": "reported",
   "source": {
    "key": "nvda_q2fy27",
    "title": "Nvidia Q2 fiscal 2027 results",
    "publisher": "Nvidia",
    "url": "https://nvidianews.nvidia.com/news/nvidia-announces-financial-results-for-second-quarter-fiscal-2027",
    "date": "2026-08-26",
    "tier": "primary"
   }
  },
  {
   "id": "L037",
   "from": "nvidia",
   "to": "microsoft",
   "kind": "chips",
   "what": "Vera Rubin systems",
   "evidence": "Azure named as a Vera Rubin ramping partner.",
   "basis": "reported",
   "source": {
    "key": "nvda_q2fy27",
    "title": "Nvidia Q2 fiscal 2027 results",
    "publisher": "Nvidia",
    "url": "https://nvidianews.nvidia.com/news/nvidia-announces-financial-results-for-second-quarter-fiscal-2027",
    "date": "2026-08-26",
    "tier": "primary"
   }
  },
  {
   "id": "L038",
   "from": "nvidia",
   "to": "oracle",
   "kind": "chips",
   "what": "Vera Rubin systems",
   "evidence": "Oracle named as a Vera Rubin ramping partner.",
   "basis": "reported",
   "source": {
    "key": "nvda_q2fy27",
    "title": "Nvidia Q2 fiscal 2027 results",
    "publisher": "Nvidia",
    "url": "https://nvidianews.nvidia.com/news/nvidia-announces-financial-results-for-second-quarter-fiscal-2027",
    "date": "2026-08-26",
    "tier": "primary"
   }
  },
  {
   "id": "L039",
   "from": "nvidia",
   "to": "google",
   "kind": "chips",
   "what": "Vera Rubin systems",
   "evidence": "Google Cloud named as a Vera Rubin ramping partner.",
   "basis": "reported",
   "source": {
    "key": "nvda_q2fy27",
    "title": "Nvidia Q2 fiscal 2027 results",
    "publisher": "Nvidia",
    "url": "https://nvidianews.nvidia.com/news/nvidia-announces-financial-results-for-second-quarter-fiscal-2027",
    "date": "2026-08-26",
    "tier": "primary"
   }
  },
  {
   "id": "L040",
   "from": "nvidia",
   "to": "nebius",
   "kind": "chips",
   "what": "Vera Rubin systems",
   "evidence": "Nebius named as a Vera Rubin ramping partner.",
   "basis": "reported",
   "source": {
    "key": "nvda_q2fy27",
    "title": "Nvidia Q2 fiscal 2027 results",
    "publisher": "Nvidia",
    "url": "https://nvidianews.nvidia.com/news/nvidia-announces-financial-results-for-second-quarter-fiscal-2027",
    "date": "2026-08-26",
    "tier": "primary"
   }
  },
  {
   "id": "L041",
   "from": "nvidia",
   "to": "spacexai",
   "kind": "chips",
   "what": "GPUs for Colossus",
   "evidence": "About 555,000 GPUs across Colossus 1 and 2; 2 GW planned.",
   "basis": "reported",
   "source": {
    "key": "colossus",
    "title": "xAI Colossus 2 GW expansion, 555k GPUs",
    "publisher": "Introl",
    "url": "https://introl.com/blog/xai-colossus-2-gigawatt-expansion-555k-gpus-january-2026",
    "date": "2026-01",
    "tier": "secondary"
   }
  },
  {
   "id": "L042",
   "from": "nvidia",
   "to": "openai",
   "kind": "capital",
   "what": "$30B equity plus dedicated capacity",
   "evidence": "Part of OpenAI's $110B February round; 3 GW inference plus 2 GW training on Vera Rubin.",
   "basis": "reported",
   "source": {
    "key": "oai_110b",
    "title": "OpenAI raises $110B in one of the largest private rounds",
    "publisher": "TechCrunch",
    "url": "https://techcrunch.com/2026/02/27/openai-raises-110b-in-one-of-the-largest-private-funding-rounds-in-history/",
    "date": "2026-02-27",
    "tier": "press"
   }
  },
  {
   "id": "L043",
   "from": "lumentum",
   "to": "nvidia",
   "kind": "chips",
   "what": "Lasers for optical links",
   "evidence": "Nvidia paid about $2B each to Lumentum and Coherent for priority capacity.",
   "basis": "reported",
   "source": {
    "key": "lasers",
    "title": "Nvidia's $4B laser lockup pushes rivals past 2027",
    "publisher": "Tech Times",
    "url": "https://www.techtimes.com/articles/317281/20260527/ai-data-center-optical-component-shortage-nvidias-4b-laser-lockup-pushes-rivals-past-2027.htm",
    "date": "2026-05-27",
    "tier": "press"
   }
  },
  {
   "id": "L044",
   "from": "coherent",
   "to": "nvidia",
   "kind": "chips",
   "what": "Lasers for optical links",
   "evidence": "Nvidia paid about $2B each to Lumentum and Coherent for priority capacity.",
   "basis": "reported",
   "source": {
    "key": "lasers",
    "title": "Nvidia's $4B laser lockup pushes rivals past 2027",
    "publisher": "Tech Times",
    "url": "https://www.techtimes.com/articles/317281/20260527/ai-data-center-optical-component-shortage-nvidias-4b-laser-lockup-pushes-rivals-past-2027.htm",
    "date": "2026-05-27",
    "tier": "press"
   }
  },
  {
   "id": "L045",
   "from": "huawei",
   "to": "deepseek",
   "kind": "chips",
   "what": "Ascend 950DT chips (planned)",
   "evidence": "Plans at least 160,000 chips for an Inner Mongolia cluster to serve models.",
   "basis": "reported",
   "source": {
    "key": "deepseek_ascend",
    "title": "DeepSeek to use Huawei Ascend 950DT in Inner Mongolia",
    "publisher": "TechNode",
    "url": "https://technode.com/2026/09/07/deepseek-huawei-ascend-950dt-chips-inner-mongolia/",
    "date": "2026-09-07",
    "tier": "press"
   }
  },
  {
   "id": "L046",
   "from": "google",
   "to": "anthropic",
   "kind": "compute",
   "what": "5 GW of TPUs over five years",
   "evidence": "Plus 3.5 GW via Broadcom from 2027.",
   "basis": "reported",
   "source": {
    "key": "goog_anth_40",
    "title": "Google to invest up to $40B in Anthropic",
    "publisher": "TechCrunch",
    "url": "https://techcrunch.com/2026/04/24/google-to-invest-up-to-40b-in-anthropic-in-cash-and-compute/",
    "date": "2026-04-24",
    "tier": "press"
   }
  },
  {
   "id": "L047",
   "from": "google",
   "to": "anthropic",
   "kind": "capital",
   "what": "Up to $40B",
   "evidence": "$10B cash at a $350B valuation plus $30B tied to targets.",
   "basis": "reported",
   "source": {
    "key": "goog_anth_40",
    "title": "Google to invest up to $40B in Anthropic",
    "publisher": "TechCrunch",
    "url": "https://techcrunch.com/2026/04/24/google-to-invest-up-to-40b-in-anthropic-in-cash-and-compute/",
    "date": "2026-04-24",
    "tier": "press"
   }
  },
  {
   "id": "L048",
   "from": "google",
   "to": "meta",
   "kind": "compute",
   "what": "TPU rental, multibillion-dollar",
   "evidence": "Meta may buy TPUs outright from 2027.",
   "basis": "reported",
   "source": {
    "key": "meta_tpu",
    "title": "Meta signs multibillion deal to rent Google TPUs",
    "publisher": "The Decoder",
    "url": "https://the-decoder.com/meta-signs-multi-billion-dollar-deal-to-rent-googles-tpus-in-a-direct-challenge-to-nvidias-ai-chip-dominance/",
    "date": "2026-02-27",
    "tier": "press"
   }
  },
  {
   "id": "L049",
   "from": "amazon",
   "to": "anthropic",
   "kind": "compute",
   "what": "Trainium capacity, up to 5 GW",
   "evidence": "About 1 GW by end-2026; Anthropic commits $100B+ over ten years.",
   "basis": "reported",
   "source": {
    "key": "anth_aws",
    "title": "Anthropic and Amazon expand compute partnership",
    "publisher": "Anthropic",
    "url": "https://www.anthropic.com/news/anthropic-amazon-compute",
    "date": "2026-04-20",
    "tier": "primary"
   }
  },
  {
   "id": "L050",
   "from": "amazon",
   "to": "anthropic",
   "kind": "capital",
   "what": "$5B plus up to $20B",
   "evidence": "On top of the earlier $8B.",
   "basis": "reported",
   "source": {
    "key": "amzn_anth_5",
    "title": "Amazon invests additional $5B in Anthropic",
    "publisher": "Amazon",
    "url": "https://www.aboutamazon.com/news/company-news/amazon-invests-additional-5-billion-anthropic-ai",
    "date": "2026-04-20",
    "tier": "primary"
   }
  },
  {
   "id": "L051",
   "from": "amazon",
   "to": "openai",
   "kind": "capital",
   "what": "$50B investment",
   "evidence": "Part of the $110B February round; $35B reportedly conditional.",
   "basis": "reported",
   "source": {
    "key": "oai_110b",
    "title": "OpenAI raises $110B in one of the largest private rounds",
    "publisher": "TechCrunch",
    "url": "https://techcrunch.com/2026/02/27/openai-raises-110b-in-one-of-the-largest-private-funding-rounds-in-history/",
    "date": "2026-02-27",
    "tier": "press"
   }
  },
  {
   "id": "L052",
   "from": "amazon",
   "to": "openai",
   "kind": "compute",
   "what": "Trainium capacity, 2 GW+",
   "evidence": "$38B deal expanded by $100B.",
   "basis": "reported",
   "source": {
    "key": "oai_110b",
    "title": "OpenAI raises $110B in one of the largest private rounds",
    "publisher": "TechCrunch",
    "url": "https://techcrunch.com/2026/02/27/openai-raises-110b-in-one-of-the-largest-private-funding-rounds-in-history/",
    "date": "2026-02-27",
    "tier": "press"
   }
  },
  {
   "id": "L053",
   "from": "microsoft",
   "to": "openai",
   "kind": "capital",
   "what": "About 27% ownership",
   "evidence": "Exclusivity ended April 27; Microsoft remains primary cloud.",
   "basis": "reported",
   "source": {
    "key": "msft_oai_reset",
    "title": "Microsoft and OpenAI gut their exclusive deal",
    "publisher": "VentureBeat",
    "url": "https://venturebeat.com/technology/microsoft-and-openai-gut-their-exclusive-deal-freeing-openai-to-sell-on-aws-and-google-cloud",
    "date": "2026-04-27",
    "tier": "press"
   }
  },
  {
   "id": "L054",
   "from": "microsoft",
   "to": "anthropic",
   "kind": "compute",
   "what": "$30B of Azure capacity",
   "evidence": "Microsoft and Nvidia committed up to $15B combined.",
   "basis": "reported",
   "source": {
    "key": "msft_nvda_anth",
    "title": "Microsoft, Nvidia and Anthropic announce strategic partnerships",
    "publisher": "Microsoft",
    "url": "https://blogs.microsoft.com/blog/2025/11/18/microsoft-nvidia-and-anthropic-announce-strategic-partnerships/",
    "date": "2025-11-18",
    "tier": "primary"
   }
  },
  {
   "id": "L055",
   "from": "oracle",
   "to": "openai",
   "kind": "compute",
   "what": "Stargate cloud capacity",
   "evidence": "Roughly half of Oracle's $664B contracted backlog is tied to OpenAI.",
   "basis": "reported",
   "source": {
    "key": "orcl_q1",
    "title": "Oracle Q1 FY27: $664B backlog",
    "publisher": "ERP Today",
    "url": "https://erp.today/oracle-q1-fy27-results-664b-backlog-ai-contracts/",
    "date": "2026-09-11",
    "tier": "press"
   }
  },
  {
   "id": "L056",
   "from": "coreweave",
   "to": "meta",
   "kind": "compute",
   "what": "GPU cloud, $35B+ total",
   "evidence": "$14.2B (2025) plus $21B more (April 2026).",
   "basis": "reported",
   "source": {
    "key": "meta_cw",
    "title": "Meta and CoreWeave $21B AI cloud deal",
    "publisher": "The Next Web",
    "url": "https://thenextweb.com/news/meta-coreweave-21-billion-ai-cloud-deal",
    "date": "2026-04",
    "tier": "press"
   }
  },
  {
   "id": "L057",
   "from": "coreweave",
   "to": "openai",
   "kind": "compute",
   "what": "GPU cloud, $22.4B total",
   "evidence": "Contracts expanded through 2025.",
   "basis": "reported",
   "source": {
    "key": "cw_oai",
    "title": "CoreWeave expands agreement with OpenAI",
    "publisher": "CoreWeave",
    "url": "https://www.coreweave.com/news/coreweave-expands-agreement-with-openai-by-up-to-6-5b",
    "date": "2025-09",
    "tier": "primary"
   }
  },
  {
   "id": "L058",
   "from": "nebius",
   "to": "microsoft",
   "kind": "compute",
   "what": "GPU cloud, $17.4 to 19.4B",
   "evidence": "Five-year agreement.",
   "basis": "reported",
   "source": {
    "key": "nebius_deals",
    "title": "Nebius signs $46B in AI cloud deals",
    "publisher": "The Motley Fool",
    "url": "https://www.fool.com/investing/2026/04/02/nebius-just-signed-46-billion-in-ai-cloud-deals-wi/",
    "date": "2026-04-02",
    "tier": "press"
   }
  },
  {
   "id": "L059",
   "from": "nebius",
   "to": "meta",
   "kind": "compute",
   "what": "GPU cloud, $12B firm plus options",
   "evidence": "Up to $15B more optional.",
   "basis": "reported",
   "source": {
    "key": "nebius_meta",
    "title": "Nebius signs new AI infrastructure agreement with Meta",
    "publisher": "Nebius",
    "url": "https://nebius.com/newsroom/nebius-signs-new-ai-infrastructure-agreement-with-meta",
    "date": "2026-04",
    "tier": "primary"
   }
  },
  {
   "id": "L060",
   "from": "crusoe",
   "to": "oracle",
   "kind": "compute",
   "what": "Abilene campus build and operation",
   "evidence": "4 of 8 buildings live; expansion beyond 1.2 GW dropped.",
   "basis": "reported",
   "source": {
    "key": "stargate_sites",
    "title": "OpenAI Stargate: where the US sites stand",
    "publisher": "Epoch AI",
    "url": "https://epoch.ai/publications/openai-stargate-where-the-us-sites-stand",
    "date": "2026-04-17",
    "tier": "press"
   }
  },
  {
   "id": "L061",
   "from": "fluidstack",
   "to": "anthropic",
   "kind": "compute",
   "what": "Data center build-out",
   "evidence": "Part of Anthropic's $50B build-out with Fluidstack.",
   "basis": "reported",
   "source": {
    "key": "anth_colossus",
    "title": "Higher limits: Anthropic and SpaceX compute agreement",
    "publisher": "Anthropic",
    "url": "https://www.anthropic.com/news/higher-limits-spacex",
    "date": "2026-05-06",
    "tier": "primary"
   }
  },
  {
   "id": "L062",
   "from": "google",
   "to": "fluidstack",
   "kind": "capital",
   "what": "Backstop on power-site leases",
   "evidence": "Google guarantees payments on about 1 GW of former crypto-mining sites.",
   "basis": "reported",
   "source": {
    "key": "fluidstack",
    "title": "A tiny startup helping Google take on Nvidia is now worth $18B",
    "publisher": "Forbes",
    "url": "https://www.forbes.com/sites/iainmartin/2026/09/03/a-tiny-startup-helping-google-take-on-nvidia-is-now-worth-18-billion/",
    "date": "2026-09-03",
    "tier": "press"
   }
  },
  {
   "id": "L063",
   "from": "spacexai",
   "to": "anthropic",
   "kind": "compute",
   "what": "All of Colossus 1",
   "evidence": "More than 300 MW and 220,000+ Nvidia GPUs.",
   "basis": "reported",
   "source": {
    "key": "anth_colossus",
    "title": "Higher limits: Anthropic and SpaceX compute agreement",
    "publisher": "Anthropic",
    "url": "https://www.anthropic.com/news/higher-limits-spacex",
    "date": "2026-05-06",
    "tier": "primary"
   }
  },
  {
   "id": "L064",
   "from": "constellation",
   "to": "microsoft",
   "kind": "power",
   "what": "Crane (Three Mile Island) restart, ~835 MW",
   "evidence": "Expected 2027.",
   "basis": "reported",
   "source": {
    "key": "ceg_meta",
    "title": "Constellation's biggest AI power deal starts June 2027",
    "publisher": "The Motley Fool",
    "url": "https://www.fool.com/investing/2026/09/08/constellation-energy-s-biggest-ai-power-deal-doesn-t-start-paying-until-june-2027/",
    "date": "2026-09-08",
    "tier": "press"
   }
  },
  {
   "id": "L065",
   "from": "constellation",
   "to": "meta",
   "kind": "power",
   "what": "Clinton plant, 1,121 MW",
   "evidence": "20 years starting June 2027.",
   "basis": "reported",
   "source": {
    "key": "ceg_meta",
    "title": "Constellation's biggest AI power deal starts June 2027",
    "publisher": "The Motley Fool",
    "url": "https://www.fool.com/investing/2026/09/08/constellation-energy-s-biggest-ai-power-deal-doesn-t-start-paying-until-june-2027/",
    "date": "2026-09-08",
    "tier": "press"
   }
  },
  {
   "id": "L066",
   "from": "constellation",
   "to": "amazon",
   "kind": "power",
   "what": "Calvert Cliffs, 690 MW",
   "evidence": "20 years plus a $3B upgrade.",
   "basis": "reported",
   "source": {
    "key": "calvert",
    "title": "Calvert Cliffs power purchase with Amazon",
    "publisher": "Maryland Matters",
    "url": "https://marylandmatters.org/2026/09/30/calvert-cliffs-power-purchase-amazon",
    "date": "2026-09-30",
    "tier": "press"
   }
  },
  {
   "id": "L067",
   "from": "vistra",
   "to": "meta",
   "kind": "power",
   "what": "2,176 MW nuclear",
   "evidence": "Perry, Davis-Besse and Beaver Valley; 2,609 MW by 2034.",
   "basis": "reported",
   "source": {
    "key": "meta_nuclear",
    "title": "Meta announces landmark agreements for new nuclear",
    "publisher": "World Nuclear News",
    "url": "https://www.world-nuclear-news.org/articles/meta-announces-landmark-agreements-for-new-nuclear",
    "date": "2026-01-09",
    "tier": "press"
   }
  },
  {
   "id": "L068",
   "from": "vistra",
   "to": "amazon",
   "kind": "power",
   "what": "Comanche Peak, 1.2 GW",
   "evidence": "20 years starting 2027.",
   "basis": "reported",
   "source": {
    "key": "vistra_amzn",
    "title": "Vistra Comanche Peak deal with Amazon",
    "publisher": "Utility Dive",
    "url": "https://www.utilitydive.com/news/vistra-texas-data-center-pause-comanche-amazon-earnings/827434/",
    "date": "2026-08",
    "tier": "press"
   }
  },
  {
   "id": "L069",
   "from": "talen",
   "to": "amazon",
   "kind": "power",
   "what": "Susquehanna, 1.92 GW",
   "evidence": "Through 2042.",
   "basis": "reported",
   "source": {
    "key": "talen_amzn",
    "title": "Talen and Amazon expand partnership",
    "publisher": "World Nuclear News",
    "url": "https://www.world-nuclear-news.org/articles/new-supply-agreement-expands-talen-amazon-partnership",
    "date": "2025-06",
    "tier": "press"
   }
  },
  {
   "id": "L070",
   "from": "terrapower",
   "to": "meta",
   "kind": "power",
   "what": "Natrium reactors, up to 2.8 GW",
   "evidence": "Two reactors plus rights to six more.",
   "basis": "reported",
   "source": {
    "key": "meta_nuclear",
    "title": "Meta announces landmark agreements for new nuclear",
    "publisher": "World Nuclear News",
    "url": "https://www.world-nuclear-news.org/articles/meta-announces-landmark-agreements-for-new-nuclear",
    "date": "2026-01-09",
    "tier": "press"
   }
  },
  {
   "id": "L071",
   "from": "oklo",
   "to": "meta",
   "kind": "power",
   "what": "1.2 GW campus, Ohio",
   "evidence": "First phase around 2030.",
   "basis": "reported",
   "source": {
    "key": "meta_nuclear",
    "title": "Meta announces landmark agreements for new nuclear",
    "publisher": "World Nuclear News",
    "url": "https://www.world-nuclear-news.org/articles/meta-announces-landmark-agreements-for-new-nuclear",
    "date": "2026-01-09",
    "tier": "press"
   }
  },
  {
   "id": "L072",
   "from": "entergy",
   "to": "meta",
   "kind": "power",
   "what": "Gas plants for Hyperion",
   "evidence": "Seven new plants, ten in total, over 7 GW.",
   "basis": "reported",
   "source": {
    "key": "hyperion_gas",
    "title": "Meta funds seven gas plants for Hyperion",
    "publisher": "WinBuzzer",
    "url": "https://winbuzzer.com/2026/03/31/meta-funds-seven-gas-plants-largest-ai-data-center-xcxwbn/",
    "date": "2026-03-31",
    "tier": "secondary"
   }
  },
  {
   "id": "L073",
   "from": "siemens-energy",
   "to": "meta",
   "kind": "power",
   "what": "Turbines for on-site gas in Ohio",
   "evidence": "Siemens Energy, Solar Turbines and Caterpillar supply Meta's off-grid Socrates South plant.",
   "basis": "reported",
   "source": {
    "key": "meta_socrates",
    "title": "Meta deploys tent data centers backed by 400MW off-grid gas",
    "publisher": "MLQ",
    "url": "https://mlq.ai/news/v2/meta-deploys-tent-based-data-centers-in-ohio-and-tennessee-backed-by-400mw-of-off-grid-gas-power/",
    "date": "2026",
    "tier": "secondary"
   }
  },
  {
   "id": "L074",
   "from": "bloom",
   "to": "oracle",
   "kind": "power",
   "what": "Fuel cells, up to 2.8 GW",
   "evidence": "On-site generation agreement.",
   "basis": "reported",
   "source": {
    "key": "bloom_orcl",
    "title": "Bloom Energy 2.8GW Oracle fuel cell deal",
    "publisher": "EnkiAI",
    "url": "https://enkiai.com/solid-oxide-fuel-cells/bloom-energy-fuel-cell-2026-2-8gw-oracle-deal/",
    "date": "2026-04",
    "tier": "secondary"
   }
  },
  {
   "id": "L075",
   "from": "talen",
   "to": "pjm",
   "kind": "power",
   "what": "Susquehanna output into the PJM grid",
   "evidence": "Amazon contract shifting to grid-delivered power in 2026.",
   "basis": "reported",
   "source": {
    "key": "talen_amzn",
    "title": "Talen and Amazon expand partnership",
    "publisher": "World Nuclear News",
    "url": "https://www.world-nuclear-news.org/articles/new-supply-agreement-expands-talen-amazon-partnership",
    "date": "2025-06",
    "tier": "press"
   }
  },
  {
   "id": "L076",
   "from": "ercot",
   "to": "crusoe",
   "kind": "power",
   "what": "Texas grid connection for Abilene",
   "evidence": "Texas delayed its first large-load study in August after a governor-ordered audit.",
   "basis": "inferred",
   "source": {
    "key": "ercot_pause",
    "title": "Texas hits pause on data center interconnections",
    "publisher": "Utility Dive",
    "url": "https://www.utilitydive.com/news/texas-hits-pause-data-center-interconnections/827046/",
    "date": "2026-08",
    "tier": "press"
   }
  },
  {
   "id": "L077",
   "from": "google",
   "to": "apple",
   "kind": "distribution",
   "what": "Gemini model behind Siri",
   "evidence": "iOS 27 shipped a fully Gemini-based Siri; reported about $1B a year.",
   "basis": "reported",
   "source": {
    "key": "apple_gemini",
    "title": "Apple and Google Gemini Siri deal",
    "publisher": "Tech Insider",
    "url": "https://tech-insider.org/apple-google-gemini-siri-deal-1-billion-2026/",
    "date": "2026-08-15",
    "tier": "secondary"
   }
  },
  {
   "id": "L078",
   "from": "openai",
   "to": "amazon",
   "kind": "distribution",
   "what": "Models on AWS Bedrock",
   "evidence": "GPT models and Codex generally available on Bedrock after exclusivity ended.",
   "basis": "reported",
   "source": {
    "key": "bedrock_oai",
    "title": "OpenAI models on Amazon Bedrock",
    "publisher": "Amazon",
    "url": "https://www.aboutamazon.com/news/aws/bedrock-openai-models",
    "date": "2026-09-08",
    "tier": "primary"
   }
  },
  {
   "id": "L079",
   "from": "moonshot",
   "to": "cursor",
   "kind": "distribution",
   "what": "Base model for Cursor's Composer 2",
   "evidence": "Cursor confirmed Composer 2 is built on Kimi 2.5.",
   "basis": "reported",
   "source": {
    "key": "cursor_kimi",
    "title": "Cursor admits Composer 2 is based on Moonshot AI's Kimi 2.5",
    "publisher": "Dataconomy",
    "url": "https://dataconomy.com/2026/03/23/cursor-admits-composer-2-based-on-moonshot-ais-kimi-2-5/",
    "date": "2026-03-23",
    "tier": "press"
   }
  },
  {
   "id": "L080",
   "from": "spacexai",
   "to": "cursor",
   "kind": "capital",
   "what": "$60B acquisition",
   "evidence": "Announced June 16, closed August 14, 2026.",
   "basis": "reported",
   "source": {
    "key": "cursor_close",
    "title": "SpaceX closes $60B Cursor deal",
    "publisher": "AI Weekly",
    "url": "https://aiweekly.co/alerts/spacex-closes-60b-cursor-deal-folds-it-into-spacexai-unit",
    "date": "2026-08-14",
    "tier": "secondary"
   }
  },
  {
   "id": "L081",
   "from": "reddit",
   "to": "google",
   "kind": "data",
   "what": "Training data license",
   "evidence": "Google and OpenAI are Reddit's two largest licensing buyers.",
   "basis": "reported",
   "source": {
    "key": "reddit_data",
    "title": "Reddit's data revenue: OpenAI and Google are the buyers",
    "publisher": "The Motley Fool",
    "url": "https://www.fool.com/investing/2026/08/17/reddits-data-revenue-openai-google-buyers/",
    "date": "2026-08-17",
    "tier": "press"
   }
  },
  {
   "id": "L082",
   "from": "reddit",
   "to": "openai",
   "kind": "data",
   "what": "Training data license",
   "evidence": "Google and OpenAI are Reddit's two largest licensing buyers.",
   "basis": "reported",
   "source": {
    "key": "reddit_data",
    "title": "Reddit's data revenue: OpenAI and Google are the buyers",
    "publisher": "The Motley Fool",
    "url": "https://www.fool.com/investing/2026/08/17/reddits-data-revenue-openai-google-buyers/",
    "date": "2026-08-17",
    "tier": "press"
   }
  },
  {
   "id": "L083",
   "from": "news-corp",
   "to": "meta",
   "kind": "data",
   "what": "News archive license",
   "evidence": "Up to $50M a year for at least three years.",
   "basis": "reported",
   "source": {
    "key": "pressgazette",
    "title": "News publisher AI deals and lawsuits tracker",
    "publisher": "Press Gazette",
    "url": "https://pressgazette.co.uk/platforms/news-publisher-ai-deals-lawsuits-openai-google/",
    "date": "2026-09-23",
    "tier": "press"
   }
  },
  {
   "id": "L084",
   "from": "us-commerce",
   "to": "anthropic",
   "kind": "rules",
   "what": "Export license required for Fable 5 and Mythos 5",
   "evidence": "Anthropic disabled both models for all customers June 12; controls lifted June 30, with Mythos 5 restored only for approved US organizations.",
   "basis": "reported",
   "source": {
    "key": "bis_fable",
    "title": "Commerce restricted access to Anthropic's latest models: what comes next",
    "publisher": "CSIS",
    "url": "https://www.csis.org/analysis/department-commerce-restricted-access-anthropics-latest-models-what-comes-next",
    "date": "2026-06-16",
    "tier": "press"
   }
  },
  {
   "id": "L085",
   "from": "us-commerce",
   "to": "nvidia",
   "kind": "rules",
   "what": "China export limits",
   "evidence": "H200 allowed case by case from January; Blackwell and Rubin stay banned.",
   "basis": "reported",
   "source": {
    "key": "h200_allow",
    "title": "US eases AI chip export rules for China",
    "publisher": "Interesting Engineering",
    "url": "https://interestingengineering.com/culture/us-eases-ai-chip-export-rules-china-nvidia-h200",
    "date": "2026-01-13",
    "tier": "press"
   }
  },
  {
   "id": "L086",
   "from": "pentagon",
   "to": "anthropic",
   "kind": "rules",
   "what": "Supply chain risk designation",
   "evidence": "Appeals court upheld the label 2-1 on Sept 25.",
   "basis": "reported",
   "source": {
    "key": "dod_anth",
    "title": "Appeals court declines to block Pentagon blacklisting of Anthropic",
    "publisher": "ABC News",
    "url": "https://abcnews.com/Business/anthropic-appeals-court-declines-block-pentagon-blacklisting/story?id=136755690",
    "date": "2026-09-25",
    "tier": "press"
   }
  },
  {
   "id": "L087",
   "from": "china-gov",
   "to": "nvidia",
   "kind": "rules",
   "what": "H200 import block",
   "evidence": "Customs stopped H200 shipments despite US approval.",
   "basis": "reported",
   "source": {
    "key": "h200_block",
    "title": "China customs blocks Nvidia H200",
    "publisher": "Taipei Times",
    "url": "https://www.taipeitimes.com/News/biz/archives/2026/01/15/2003850605",
    "date": "2026-01-15",
    "tier": "press"
   }
  },
  {
   "id": "L088",
   "from": "eu-ai-office",
   "to": "openai",
   "kind": "rules",
   "what": "General-purpose model enforcement",
   "evidence": "From Aug 2, 2026, fines up to 3% of worldwide turnover or €15M, whichever is higher.",
   "basis": "inferred",
   "source": {
    "key": "eu_gpai",
    "title": "GPAI enforcement has started",
    "publisher": "Praxikon",
    "url": "https://www.praxikon.com/en/posts/gpai-enforcement-has-started-powers-2026",
    "date": "2026-08",
    "tier": "secondary"
   }
  },
  {
   "id": "L089",
   "from": "eu-ai-office",
   "to": "mistral",
   "kind": "rules",
   "what": "General-purpose model enforcement",
   "evidence": "From Aug 2, 2026, fines up to 3% of worldwide turnover or €15M, whichever is higher.",
   "basis": "inferred",
   "source": {
    "key": "eu_gpai",
    "title": "GPAI enforcement has started",
    "publisher": "Praxikon",
    "url": "https://www.praxikon.com/en/posts/gpai-enforcement-has-started-powers-2026",
    "date": "2026-08",
    "tier": "secondary"
   }
  },
  {
   "id": "L090",
   "from": "ferc",
   "to": "pjm",
   "kind": "rules",
   "what": "Large-load connection rules",
   "evidence": "Ordered six grid operators to justify or rewrite their rules.",
   "basis": "reported",
   "source": {
    "key": "ferc_large_load",
    "title": "FERC acts to advance data center large-load integration",
    "publisher": "Duane Morris",
    "url": "https://www.duanemorris.com/alerts/ferc_acts_advance_data_center_large_load_integration_six_rto_regions_0626.html",
    "date": "2026-06-18",
    "tier": "press"
   }
  }
 ],
 "chokepoints": [
  {
   "id": "cowos",
   "name": "Advanced packaging (CoWoS)",
   "force": "compute",
   "layer": "fab",
   "status": "binding",
   "controllers": [
    "tsmc",
    "ase",
    "amkor"
   ],
   "observed": [
    {
     "fact": "Nvidia reportedly holds more than half of TSMC's 2026 CoWoS capacity.",
     "source": {
      "key": "cowos_osat",
      "title": "TSMC expands outsourcing of CoWoS front-end step to OSATs",
      "publisher": "TrendForce",
      "url": "https://www.trendforce.com/news/2026/08/05/news-tsmc-reportedly-expands-outsourcing-of-key-cowos-front-end-step-to-osats-amid-rising-nvidia-asic-demand/",
      "date": "2026-08-05",
      "tier": "press"
     }
    },
    {
     "fact": "TSMC is reportedly outsourcing part of CoWoS to ASE; ASE and SPIL are adding their own lines with AMD's backing.",
     "source": {
      "key": "cowos_osat",
      "title": "TSMC expands outsourcing of CoWoS front-end step to OSATs",
      "publisher": "TrendForce",
      "url": "https://www.trendforce.com/news/2026/08/05/news-tsmc-reportedly-expands-outsourcing-of-key-cowos-front-end-step-to-osats-amid-rising-nvidia-asic-demand/",
      "date": "2026-08-05",
      "tier": "press"
     }
    },
    {
     "fact": "CoWoS and 3nm lead times run 52 to 78 weeks; 2nm runs 78 to 156 weeks.",
     "source": {
      "key": "foundry_alloc",
      "title": "Foundry allocation status 2026",
      "publisher": "Silicon Analysts",
      "url": "https://siliconanalysts.com/analysis/foundry-allocation-status-q1-2026",
      "date": "2026-06-16",
      "tier": "secondary"
     }
    }
   ],
   "interpretation": {
    "text": "The tightest gate in the stack. Whoever holds packaging allocation decides who ships AI chips in 2027, and today that is Nvidia. Custom-chip programs from Google, Amazon and Meta compete for the remainder. TSMC outsourcing part of the process tells you it cannot add capacity fast enough on its own.",
    "author": "exmxc.ai",
    "type": "interpretation"
   }
  },
  {
   "id": "hbm",
   "name": "High-bandwidth memory",
   "force": "compute",
   "layer": "memory",
   "status": "binding",
   "controllers": [
    "sk-hynix",
    "samsung-memory",
    "micron"
   ],
   "observed": [
    {
     "fact": "Micron can meet only 50 to 67% of key customers' medium-term demand and signed $100B of minimum commitments through 2030.",
     "source": {
      "key": "micron_q4",
      "title": "Micron posts record $54B quarter, locks $100B memory contracts",
      "publisher": "Tech Times",
      "url": "https://www.techtimes.com/articles/328376/20261001/micron-posts-record-54b-quarter-locks-hyperscalers-100b-memory-contracts.htm",
      "date": "2026-10-01",
      "tier": "press"
     }
    },
    {
     "fact": "SK hynix reported a 76% operating margin in Q2 2026.",
     "source": {
      "key": "skh_q2",
      "title": "SK hynix Q2 2026 business results",
      "publisher": "SK hynix",
      "url": "https://news.skhynix.com/en/q2-2026-business-results/",
      "date": "2026-07-29",
      "tier": "primary"
     }
    }
   ],
   "interpretation": {
    "text": "Three companies, sold out through 2027, with customers prepaying billions to secure supply. That is pricing power memory makers have never had. Samsung's return as a qualified supplier is the main thing that could loosen it, and it has not loosened yet.",
    "author": "exmxc.ai",
    "type": "interpretation"
   }
  },
  {
   "id": "leading-edge",
   "name": "Leading-edge chipmaking (3nm, 2nm)",
   "force": "compute",
   "layer": "fab",
   "status": "tight",
   "controllers": [
    "tsmc"
   ],
   "observed": [
    {
     "fact": "TSMC 2nm is reported booked into 2028.",
     "source": {
      "key": "foundry_alloc",
      "title": "Foundry allocation status 2026",
      "publisher": "Silicon Analysts",
      "url": "https://siliconanalysts.com/analysis/foundry-allocation-status-q1-2026",
      "date": "2026-06-16",
      "tier": "secondary"
     }
    },
    {
     "fact": "TSMC raised 2026 capital spending to as much as $64B.",
     "source": {
      "key": "tsmc_q2",
      "title": "TSMC Q2 2026 results",
      "publisher": "BigGo Finance",
      "url": "https://finance.biggo.com/news/US_TSM_2026-07-16",
      "date": "2026-07-16",
      "tier": "secondary"
     }
    }
   ],
   "interpretation": {
    "text": "TSMC is the single point of failure for nearly every AI chip, Nvidia's and the custom ones alike. Samsung's Tesla win and Intel's 18A are real but small. A second source at scale is a 2028 story at the earliest.",
    "author": "exmxc.ai",
    "type": "interpretation"
   }
  },
  {
   "id": "euv",
   "name": "EUV lithography",
   "force": "compute",
   "layer": "tools",
   "status": "tight",
   "controllers": [
    "asml"
   ],
   "observed": [
    {
     "fact": "ASML is the only maker; it targets 60+ EUV machines in 2026.",
     "source": {
      "key": "asml_euv60",
      "title": "ASML EUV shipments: 60 units in 2026",
      "publisher": "Silicon Analysts",
      "url": "https://siliconanalysts.com/analysis/asml-euv-shipments-60-units-2026-memory-demand",
      "date": "2026",
      "tier": "secondary"
     }
    },
    {
     "fact": "2026 sales guided at €43 to 45B, raised in July.",
     "source": {
      "key": "asml_q2",
      "title": "ASML Q2 2026 results slides",
      "publisher": "Investing.com",
      "url": "https://www.investing.com/news/company-news/asml-q2-2026-slides-outlook-raised-on-ai-demand-high-na-milestone-93CH-4792166",
      "date": "2026-07-15",
      "tier": "press"
     }
    }
   ],
   "interpretation": {
    "text": "A monopoly with a long queue. Memory makers are now its fastest-growing buyers, which links this chokepoint directly to the HBM one.",
    "author": "exmxc.ai",
    "type": "interpretation"
   }
  },
  {
   "id": "lasers",
   "name": "Optical lasers",
   "force": "compute",
   "layer": "chips",
   "status": "binding",
   "controllers": [
    "lumentum",
    "coherent"
   ],
   "observed": [
    {
     "fact": "Nvidia paid about $4B to lock up laser capacity, pushing rivals' lead times past 2027.",
     "source": {
      "key": "lasers",
      "title": "Nvidia's $4B laser lockup pushes rivals past 2027",
      "publisher": "Tech Times",
      "url": "https://www.techtimes.com/articles/317281/20260527/ai-data-center-optical-component-shortage-nvidias-4b-laser-lockup-pushes-rivals-past-2027.htm",
      "date": "2026-05-27",
      "tier": "press"
     }
    },
    {
     "fact": "Nvidia's co-packaged optics switch is in full production with a five-vendor supply chain.",
     "source": {
      "key": "cpo",
      "title": "Spectrum-X Photonics enters full production",
      "publisher": "StorageReview",
      "url": "https://www.storagereview.com/news/nvidia-spectrum-x-ethernet-photonics-enters-full-production-with-4x-fewer-lasers-and-a-five-vendor-cpo-supply-chain",
      "date": "2026-08-15",
      "tier": "press"
     }
    }
   ],
   "interpretation": {
    "text": "The least-watched bottleneck. Connecting tens of thousands of chips needs lasers, and about five companies make them at scale. Nvidia locking up supply is a quiet way of making rival chip clusters harder to build.",
    "author": "exmxc.ai",
    "type": "interpretation"
   }
  },
  {
   "id": "turbines",
   "name": "Gas turbines",
   "force": "energy",
   "layer": "energy",
   "status": "binding",
   "controllers": [
    "ge-vernova",
    "siemens-energy",
    "mitsubishi-power"
   ],
   "observed": [
    {
     "fact": "GE Vernova's backlog is 116 GW including slot reservations (53 GW firm orders), against output rising to about 20 GW a year.",
     "source": {
      "key": "gev_backlog",
      "title": "GE Vernova gas turbine backlog hits 116 GW",
      "publisher": "Turbomachinery Magazine",
      "url": "https://www.turbomachinerymag.com/view/ge-vernova-gas-turbine-backlog-hits-116-gw-as-power-orders-more-than-double",
      "date": "2026-07",
      "tier": "press"
     }
    },
    {
     "fact": "Siemens Energy quotes lead times of three years or more.",
     "source": {
      "key": "siemens_backlog",
      "title": "Siemens gas turbine backlog nears 70 GW",
      "publisher": "Utility Dive",
      "url": "https://www.utilitydive.com/news/siemens-gas-turbine-backlog-nears-70-gw-as-company-expands-manufacturing/827390/",
      "date": "2026-08",
      "tier": "press"
     }
    },
    {
     "fact": "Mitsubishi's new orders deliver 2028 to 2030.",
     "source": {
      "key": "mhi_backlog",
      "title": "Mitsubishi gas turbine backlog",
      "publisher": "Utility Dive",
      "url": "https://www.utilitydive.com/news/mitsubishi-gas-turbine-backlog-earnings/827761/",
      "date": "2026-08",
      "tier": "press"
     }
    }
   ],
   "interpretation": {
    "text": "Gas is the near-term answer to AI power, and three companies make the turbines. Their order books now run years ahead of their factories. This is the clearest case of the stack shifting rent from chips to heavy equipment.",
    "author": "exmxc.ai",
    "type": "interpretation"
   }
  },
  {
   "id": "grid",
   "name": "Grid connections and capacity",
   "force": "energy",
   "layer": "energy",
   "status": "binding",
   "controllers": [
    "pjm",
    "ercot",
    "ferc"
   ],
   "observed": [
    {
     "fact": "PJM's 2028/29 auction cleared at its price cap and still came up 6,831 MW short.",
     "source": {
      "key": "pjm_auction",
      "title": "PJM capacity auction clears at $325 cap, short of reliability requirement",
      "publisher": "RTO Insider",
      "url": "https://www.rtoinsider.com/136577-pjm-capacity-auction-clears-325-cap-short-reliability-requirement/",
      "date": "2026-07-15",
      "tier": "press"
     }
    },
    {
     "fact": "About 474 GW is queued in Texas, 90% of it data centers; ERCOT delayed its first large-load study after the governor ordered an audit.",
     "source": {
      "key": "ercot_pause",
      "title": "Texas hits pause on data center interconnections",
      "publisher": "Utility Dive",
      "url": "https://www.utilitydive.com/news/texas-hits-pause-data-center-interconnections/827046/",
      "date": "2026-08",
      "tier": "press"
     }
    },
    {
     "fact": "New York paused new hyperscale data center permits statewide for up to a year.",
     "source": {
      "key": "ny_moratorium",
      "title": "Governor Hochul launches first statewide moratorium on new hyperscale data centers",
      "publisher": "New York State",
      "url": "https://www.governor.ny.gov/news/first-statewide-moratorium-new-hyperscale-data-centers-launched-governor-kathy-hochul",
      "date": "2026-07-14",
      "tier": "primary"
     }
    }
   ],
   "interpretation": {
    "text": "The grid is now the binding constraint on how fast AI capacity comes online. Expect more sites built with their own power plants, and more political fights over who pays.",
    "author": "exmxc.ai",
    "type": "interpretation"
   }
  },
  {
   "id": "electrical",
   "name": "Transformers and switchgear",
   "force": "energy",
   "layer": "energy",
   "status": "tight",
   "controllers": [
    "eaton",
    "siemens-energy"
   ],
   "observed": [
    {
     "fact": "Large power transformers average about 128 weeks to arrive; about 80% are imported.",
     "source": {
      "key": "transformers",
      "title": "Power transformer lead times and the US grid shortage",
      "publisher": "Industrial Sage",
      "url": "https://www.industrialsage.com/power-transformer-lead-times-us-grid-shortage/",
      "date": "2026",
      "tier": "secondary"
     }
    },
    {
     "fact": "Eaton puts the US data center backlog at 307 GW, about 15 years of building at 2025 rates.",
     "source": {
      "key": "eaton",
      "title": "Eaton and the data center supercycle",
      "publisher": "Electrical Trends",
      "url": "https://electricaltrends.com/2026/08/26/eaton-delivers-again-a-north-american-powerhouse-fueled-by-the-data-center-supercycle/",
      "date": "2026-08-26",
      "tier": "secondary"
     }
    }
   ],
   "interpretation": {
    "text": "Unglamorous and slow to fix. Even a site with power promised cannot use it without this gear.",
    "author": "exmxc.ai",
    "type": "interpretation"
   }
  },
  {
   "id": "firm-power",
   "name": "Firm nuclear power",
   "force": "energy",
   "layer": "energy",
   "status": "tight",
   "controllers": [
    "constellation",
    "vistra",
    "talen"
   ],
   "observed": [
    {
     "fact": "Meta (Clinton, 20 years) and Microsoft (Crane restart) have signed with Constellation.",
     "source": {
      "key": "ceg_meta",
      "title": "Constellation's biggest AI power deal starts June 2027",
      "publisher": "The Motley Fool",
      "url": "https://www.fool.com/investing/2026/09/08/constellation-energy-s-biggest-ai-power-deal-doesn-t-start-paying-until-june-2027/",
      "date": "2026-09-08",
      "tier": "press"
     }
    },
    {
     "fact": "Amazon has signed for Calvert Cliffs (20 years) and, with Talen, Susquehanna through 2042.",
     "source": {
      "key": "calvert",
      "title": "Calvert Cliffs power purchase with Amazon",
      "publisher": "Maryland Matters",
      "url": "https://marylandmatters.org/2026/09/30/calvert-cliffs-power-purchase-amazon",
      "date": "2026-09-30",
      "tier": "press"
     }
    },
    {
     "fact": "Meta's new-reactor deals with Oklo and TerraPower deliver from about 2030 to 2032.",
     "source": {
      "key": "meta_nuclear",
      "title": "Meta announces landmark agreements for new nuclear",
      "publisher": "World Nuclear News",
      "url": "https://www.world-nuclear-news.org/articles/meta-announces-landmark-agreements-for-new-nuclear",
      "date": "2026-01-09",
      "tier": "press"
     }
    }
   ],
   "interpretation": {
    "text": "The existing fleet is almost fully spoken for, and new nuclear arrives after 2030. Owners of operating plants hold scarce, long-dated contracts. Nuclear is the long-term answer; gas is the answer now.",
    "author": "exmxc.ai",
    "type": "interpretation"
   }
  },
  {
   "id": "model-permission",
   "name": "Government permission to sell a model",
   "force": "alignment",
   "layer": "rules",
   "status": "binding",
   "controllers": [
    "us-commerce",
    "pentagon",
    "eu-ai-office"
   ],
   "observed": [
    {
     "fact": "On June 12 Commerce required export licenses for foreign access to Fable 5 and Mythos 5, and Anthropic disabled both for all customers. Controls were lifted June 30.",
     "source": {
      "key": "bis_fable",
      "title": "Commerce restricted access to Anthropic's latest models: what comes next",
      "publisher": "CSIS",
      "url": "https://www.csis.org/analysis/department-commerce-restricted-access-anthropics-latest-models-what-comes-next",
      "date": "2026-06-16",
      "tier": "press"
     }
    },
    {
     "fact": "Access was restored July 1; Mythos 5 returned only for approved US organizations.",
     "source": {
      "key": "anth_access",
      "title": "Statement on Fable and Mythos access",
      "publisher": "Anthropic",
      "url": "https://www.anthropic.com/news/fable-mythos-access",
      "date": "2026-07-01",
      "tier": "primary"
     }
    },
    {
     "fact": "The Pentagon's supply chain risk label on Anthropic was upheld on appeal.",
     "source": {
      "key": "dod_anth",
      "title": "Appeals court declines to block Pentagon blacklisting of Anthropic",
      "publisher": "ABC News",
      "url": "https://abcnews.com/Business/anthropic-appeals-court-declines-block-pentagon-blacklisting/story?id=136755690",
      "date": "2026-09-25",
      "tier": "press"
     }
    },
    {
     "fact": "EU enforcement against general-purpose model providers began Aug 2, 2026.",
     "source": {
      "key": "eu_gpai",
      "title": "GPAI enforcement has started",
      "publisher": "Praxikon",
      "url": "https://www.praxikon.com/en/posts/gpai-enforcement-has-started-powers-2026",
      "date": "2026-08",
      "tier": "secondary"
     }
    }
   ],
   "interpretation": {
    "text": "New in 2026. A frontier model can now be switched off for every customer by a government letter. That makes regulatory standing a real input to a lab's revenue, and it hands foreign buyers an argument for local or open-weight models.",
    "author": "exmxc.ai",
    "type": "interpretation"
   }
  },
  {
   "id": "default-assistant",
   "name": "The default assistant",
   "force": "interface",
   "layer": "interface",
   "status": "tight",
   "controllers": [
    "google",
    "apple",
    "openai"
   ],
   "observed": [
    {
     "fact": "Siri in iOS 27 runs on a custom Google Gemini model.",
     "source": {
      "key": "apple_gemini",
      "title": "Apple and Google Gemini Siri deal",
      "publisher": "Tech Insider",
      "url": "https://tech-insider.org/apple-google-gemini-siri-deal-1-billion-2026/",
      "date": "2026-08-15",
      "tier": "secondary"
     }
    },
    {
     "fact": "Gemini app at 950M monthly users; AI search mode above 1B.",
     "source": {
      "key": "goog_q2",
      "title": "Alphabet Q2 2026 earnings: CEO remarks",
      "publisher": "Google",
      "url": "https://blog.google/company-news/inside-google/message-ceo/alphabet-earnings-q2-2026/",
      "date": "2026-07-22",
      "tier": "primary"
     }
    }
   ],
   "interpretation": {
    "text": "Google now sits behind search, Android and Siri at once. OpenAI has the largest standalone audience but no device. Whoever is the default on the phone wins the next round of distribution.",
    "author": "exmxc.ai",
    "type": "interpretation"
   }
  },
  {
   "id": "china-stack",
   "name": "China's domestic chip supply",
   "force": "compute",
   "layer": "chips",
   "status": "binding",
   "controllers": [
    "smic",
    "huawei",
    "cxmt"
   ],
   "observed": [
    {
     "fact": "SMIC ran at 93.7% utilization in Q2 2026.",
     "source": {
      "key": "smic_q2",
      "title": "SMIC Q2 2026 revenue surpasses $3B",
      "publisher": "AnySilicon",
      "url": "https://anysilicon.com/news/smic-q2-2026-revenue-surpasses-3-billion-as-ai-demand-tightens-foundry-capacity/",
      "date": "2026-08-15",
      "tier": "secondary"
     }
    },
    {
     "fact": "Only about 240k of 1.5M Ascend chips use domestic HBM.",
     "source": {
      "key": "huawei_960",
      "title": "Huawei pulls Ascend 960DT forward; demand outstrips capacity",
      "publisher": "Traders Agency",
      "url": "https://tradersagency.com/blog/huawei-pulls-ascend-960dt-forward-to-q1-2027-says-ai-chip-demand-outstrips-its-capacity",
      "date": "2026-09-17",
      "tier": "secondary"
     }
    },
    {
     "fact": "China blocked Nvidia H200 imports even after the US allowed them.",
     "source": {
      "key": "h200_block",
      "title": "China customs blocks Nvidia H200",
      "publisher": "Taipei Times",
      "url": "https://www.taipeitimes.com/News/biz/archives/2026/01/15/2003850605",
      "date": "2026-01-15",
      "tier": "press"
     }
    }
   ],
   "interpretation": {
    "text": "China chose self-reliance over Nvidia, and the price is a hard ceiling on its compute. Memory, not chip design, is the weak link. Watch CXMT's HBM yields.",
    "author": "exmxc.ai",
    "type": "interpretation"
   }
  }
 ],
 "changes_2026": [
  {
   "date": "2026-01-09",
   "event": "Meta signs about 6.6 GW of nuclear deals with Vistra, TerraPower and Oklo.",
   "source": {
    "key": "meta_nuclear",
    "title": "Meta announces landmark agreements for new nuclear",
    "publisher": "World Nuclear News",
    "url": "https://www.world-nuclear-news.org/articles/meta-announces-landmark-agreements-for-new-nuclear",
    "date": "2026-01-09",
    "tier": "press"
   }
  },
  {
   "date": "2026-01-15",
   "event": "China's customs blocks Nvidia H200 imports two days after the US allowed them.",
   "source": {
    "key": "h200_block",
    "title": "China customs blocks Nvidia H200",
    "publisher": "Taipei Times",
    "url": "https://www.taipeitimes.com/News/biz/archives/2026/01/15/2003850605",
    "date": "2026-01-15",
    "tier": "press"
   }
  },
  {
   "date": "2026-02-27",
   "event": "OpenAI raises $110B from Amazon, Nvidia and SoftBank.",
   "source": {
    "key": "oai_110b",
    "title": "OpenAI raises $110B in one of the largest private rounds",
    "publisher": "TechCrunch",
    "url": "https://techcrunch.com/2026/02/27/openai-raises-110b-in-one-of-the-largest-private-funding-rounds-in-history/",
    "date": "2026-02-27",
    "tier": "press"
   }
  },
  {
   "date": "2026-02-27",
   "event": "Meta agrees to rent Google's TPUs, the first big outside buyer after Anthropic.",
   "source": {
    "key": "meta_tpu",
    "title": "Meta signs multibillion deal to rent Google TPUs",
    "publisher": "The Decoder",
    "url": "https://the-decoder.com/meta-signs-multi-billion-dollar-deal-to-rent-googles-tpus-in-a-direct-challenge-to-nvidias-ai-chip-dominance/",
    "date": "2026-02-27",
    "tier": "press"
   }
  },
  {
   "date": "2026-03-06",
   "event": "Oracle and OpenAI drop the Abilene Stargate expansion.",
   "source": {
    "key": "abilene_drop",
    "title": "Oracle/OpenAI drop Abilene expansion; Meta in talks for Crusoe capacity",
    "publisher": "DatacenterDynamics",
    "url": "https://www.datacenterdynamics.com/en/news/oracleopenai-drop-plans-to-expand-flagship-abilene-stargate-site-meta-in-talks-to-pick-up-crusoe-capacity-with-nvidias-help/",
    "date": "2026-03-06",
    "tier": "press"
   }
  },
  {
   "date": "2026-04-07",
   "event": "Anthropic, Google and Broadcom sign a 3.5 GW TPU deal starting 2027.",
   "source": {
    "key": "anth_goog_avgo",
    "title": "Anthropic, Google and Broadcom announce 3.5GW TPU deal",
    "publisher": "Silicon Republic",
    "url": "https://www.siliconrepublic.com/machines/anthropic-google-broadcom-announce-3-5gw-tpu-deal",
    "date": "2026-04-07",
    "tier": "press"
   }
  },
  {
   "date": "2026-04-20",
   "event": "Amazon and Anthropic expand to up to 5 GW of Trainium.",
   "source": {
    "key": "anth_aws",
    "title": "Anthropic and Amazon expand compute partnership",
    "publisher": "Anthropic",
    "url": "https://www.anthropic.com/news/anthropic-amazon-compute",
    "date": "2026-04-20",
    "tier": "primary"
   }
  },
  {
   "date": "2026-04-27",
   "event": "Microsoft and OpenAI end exclusivity; OpenAI models reach AWS.",
   "source": {
    "key": "msft_oai_reset",
    "title": "Microsoft and OpenAI gut their exclusive deal",
    "publisher": "VentureBeat",
    "url": "https://venturebeat.com/technology/microsoft-and-openai-gut-their-exclusive-deal-freeing-openai-to-sell-on-aws-and-google-cloud",
    "date": "2026-04-27",
    "tier": "press"
   }
  },
  {
   "date": "2026-05-06",
   "event": "Anthropic rents all of SpaceX's Colossus 1 data center.",
   "source": {
    "key": "anth_colossus",
    "title": "Higher limits: Anthropic and SpaceX compute agreement",
    "publisher": "Anthropic",
    "url": "https://www.anthropic.com/news/higher-limits-spacex",
    "date": "2026-05-06",
    "tier": "primary"
   }
  },
  {
   "date": "2026-06-02",
   "event": "Nvidia's Vera Rubin enters full production with three HBM4 suppliers.",
   "source": {
    "key": "hbm4_suppliers",
    "title": "Vera Rubin enters full production; HBM4 suppliers named",
    "publisher": "Tech Times",
    "url": "https://www.techtimes.com/articles/317539/20260602/nvidia-vera-rubin-enters-full-production-samsung-sk-hynix-micron-named-hbm4-suppliers.htm",
    "date": "2026-06-02",
    "tier": "press"
   }
  },
  {
   "date": "2026-06-12",
   "event": "Commerce requires licenses for Anthropic's Fable 5 and Mythos 5; Anthropic disables both for all customers until controls lift June 30.",
   "source": {
    "key": "bis_fable",
    "title": "Commerce restricted access to Anthropic's latest models: what comes next",
    "publisher": "CSIS",
    "url": "https://www.csis.org/analysis/department-commerce-restricted-access-anthropics-latest-models-what-comes-next",
    "date": "2026-06-16",
    "tier": "press"
   }
  },
  {
   "date": "2026-06-18",
   "event": "FERC orders six grid operators to rewrite large-load connection rules.",
   "source": {
    "key": "ferc_large_load",
    "title": "FERC acts to advance data center large-load integration",
    "publisher": "Duane Morris",
    "url": "https://www.duanemorris.com/alerts/ferc_acts_advance_data_center_large_load_integration_six_rto_regions_0626.html",
    "date": "2026-06-18",
    "tier": "press"
   }
  },
  {
   "date": "2026-07-14",
   "event": "New York imposes the first statewide hyperscale data center moratorium.",
   "source": {
    "key": "ny_moratorium",
    "title": "Governor Hochul launches first statewide moratorium on new hyperscale data centers",
    "publisher": "New York State",
    "url": "https://www.governor.ny.gov/news/first-statewide-moratorium-new-hyperscale-data-centers-launched-governor-kathy-hochul",
    "date": "2026-07-14",
    "tier": "primary"
   }
  },
  {
   "date": "2026-07-15",
   "event": "PJM capacity auction clears at its cap and still falls short.",
   "source": {
    "key": "pjm_auction",
    "title": "PJM capacity auction clears at $325 cap, short of reliability requirement",
    "publisher": "RTO Insider",
    "url": "https://www.rtoinsider.com/136577-pjm-capacity-auction-clears-325-cap-short-reliability-requirement/",
    "date": "2026-07-15",
    "tier": "press"
   }
  },
  {
   "date": "2026-08-02",
   "event": "EU begins enforcing rules on general-purpose AI models.",
   "source": {
    "key": "eu_gpai",
    "title": "GPAI enforcement has started",
    "publisher": "Praxikon",
    "url": "https://www.praxikon.com/en/posts/gpai-enforcement-has-started-powers-2026",
    "date": "2026-08",
    "tier": "secondary"
   }
  },
  {
   "date": "2026-08-05",
   "event": "TSMC reportedly outsources part of CoWoS packaging to ASE.",
   "source": {
    "key": "cowos_osat",
    "title": "TSMC expands outsourcing of CoWoS front-end step to OSATs",
    "publisher": "TrendForce",
    "url": "https://www.trendforce.com/news/2026/08/05/news-tsmc-reportedly-expands-outsourcing-of-key-cowos-front-end-step-to-osats-amid-rising-nvidia-asic-demand/",
    "date": "2026-08-05",
    "tier": "press"
   }
  },
  {
   "date": "2026-08-14",
   "event": "SpaceX closes its $60B purchase of Cursor.",
   "source": {
    "key": "cursor_spacex",
    "title": "SpaceX to acquire Cursor in $60B deal",
    "publisher": "DevOps.com",
    "url": "https://devops.com/spacex-to-acquire-ai-coding-leader-cursor-in-60-billion-blockbuster-deal/",
    "date": "2026-06-16",
    "tier": "press"
   }
  },
  {
   "date": "2026-09-11",
   "event": "Oracle reports a $664B backlog with negative free cash flow.",
   "source": {
    "key": "orcl_q1",
    "title": "Oracle Q1 FY27: $664B backlog",
    "publisher": "ERP Today",
    "url": "https://erp.today/oracle-q1-fy27-results-664b-backlog-ai-contracts/",
    "date": "2026-09-11",
    "tier": "press"
   }
  },
  {
   "date": "2026-09-25",
   "event": "Appeals court upholds the Pentagon's supply chain risk label on Anthropic.",
   "source": {
    "key": "dod_anth",
    "title": "Appeals court declines to block Pentagon blacklisting of Anthropic",
    "publisher": "ABC News",
    "url": "https://abcnews.com/Business/anthropic-appeals-court-declines-block-pentagon-blacklisting/story?id=136755690",
    "date": "2026-09-25",
    "tier": "press"
   }
  },
  {
   "date": "2026-10-01",
   "event": "Micron signs $100B of multi-year memory commitments; says it can meet only half to two-thirds of demand.",
   "source": {
    "key": "micron_q4",
    "title": "Micron posts record $54B quarter, locks $100B memory contracts",
    "publisher": "Tech Times",
    "url": "https://www.techtimes.com/articles/328376/20261001/micron-posts-record-54b-quarter-locks-hyperscalers-100b-memory-contracts.htm",
    "date": "2026-10-01",
    "tier": "press"
   }
  }
 ]
}