Latest across your models
The AI buildout, quantified.
Fourteen institutional models covering compute end to end: accelerator shipments, cluster networking, cloud economics, datacenters, power, and the silicon that feeds them all. Built from the supply chain up, and reconciled against each other.
AI Infrastructure
3 models
Accelerator & HBM Model
Bottom-up demand forecast for AI accelerators and high-bandwidth memory across Nvidia, AMD, Google, AWS, Meta, Microsoft, Intel, and every major Chinese hyperscaler, with customer-level shipments and installed base for 60+ accounts.
Hyperscalers, semiconductor companies, public & private investors
View modelAI Networking Model
Cluster networking
Quantitative model of AI cluster networking spend: switches, transceivers, AECs, and DACs across scale-up, scale-out, and front-end networks, with vendor market share and next-generation optical interconnects.
View modelMemory Model
DRAM & NAND
DRAM and NAND cycle tracker: wafer capacity, bit supply, HBM pricing, inventory levels, and node transition timelines, covering Samsung, SK hynix, and Micron with fab-level output.
View modelFinance
3 models
AI Cloud TCO Model
Comprehensive total cost of ownership for AI clouds: GPU rental economics, residual value curves, three-statement financials for every major cloud operator, and training and inference cost per token.
Cloud operators, investors & finance teams
View modelSpace DC TCO Model
Orbital compute
Scenario-based total cost model for orbital compute: launch economics, thermal rejection, GPU-level TCO, and cost crossover against terrestrial datacenters through 2050.
View modelTokenomics Model
Serving economics
Unit economics engine for AI serving: bottoms-up token throughput, ROI on AI spend, and demand-side translation of revenue into future accelerator demand by hyperscaler, lab, and neocloud.
View modelDatacenter
3 models
Datacenter Industry Model
Global datacenter capacity tracker covering 5,000+ facilities: hyperscaler buildouts, colocation markets, power constraints, and regional deployment trends through 2030, with per-campus power draw, lease structures, and planned versus operational capacity.
Infrastructure, energy, investors
View modelEnergy Model
Grid demand
Translates AI datacenter buildout into power infrastructure demand: utility-level supply and demand gaps, generator inventory, renewable PPAs, gas turbine lead times, and where the grid will fall short.
View modelIndustrials Model
Supply-chain map
Datacenter supply-chain map: 500+ suppliers, top-down TAM, capex per megawatt, and bottom-up revenue build across power, cooling, and facilities, from switchgear and transformers to liquid cooling loops.
View modelSemiconductor Supply Chain
4 models
Wafer Fab Equipment Model
Bottoms-up forecast of semiconductor equipment sales: lithography, deposition, etch, and metrology, broken down by chipmaker, node, and vendor, tracking ASML, Applied Materials, Lam, KLA, and Tokyo Electron with fab-level tool install projections.
Investors & supply chain teams
View modelFoundry Industry Model
Wafer economics
Competitive capacity model covering TSMC, Samsung, Intel Foundry, SMIC, and emerging players: wafer shipments, ASPs, utilization, and node mix through 2030.
View modelChipBook
Industry tracker
Monthly AI and semiconductor industry tracker: 125+ datasets covering the full value chain. Core industry trackers, memory trackers, and hand-picked timely datasets distilled from our proprietary models.
View modelVR NVL72 Component BoM & Power Model
Platform BoM
System-level bill of materials and integrated power view for the NVL72 platform: compute modules, NVLink switching, tray infrastructure, thermal and cooling, PCB and substrate, and power delivery.
View modelCore Research
Three configurations · fourth coming soon
Core Research
Institutional research tracking the inflections that move the AI and semiconductor stack, from packaging and memory to power, networking, and model economics, with clear reads on who wins and who loses. Available in three configurations, with a fourth on the way.
Institutional teams across investing, semiconductors & AI infrastructure
View modelThe newsletter, read by 284,000+ every month.
Long-form public analysis on Nvidia, TSMC, HBM, hyperscaler capex, and the rest of the AI supply chain. The newsletter is a separate publication: a newsletter subscription does not include the institutional models or Core Research on this page.
Aug 16, 2026
$12B of US ratepayers' money wasted on a modeling mistake and PJM wants to do it again
American Grid design needs an overhaul, Why it is good to be full of cold air.
Read more
Aug 10, 2026
Ultra-High Interactivity on NVIDIA GPUs? – TileRT InferenceX
Can TileRT software on NVIDIA GPU compete with Cerebras, Groq LPU, SambaNova? Batch Size 1, Disaggregated engine, high throughput engine Prefill, high intera…
Read more
Aug 7, 2026
SpaceX 10GW in 2027 – Why It’s Real, Will Drive $300B ARR for SpaceX, and Why Microsoft Will Be the Largest Offtaker
Inference at 100B/GW/year, SpaceX's stellar pace, Microsoft's 10GW 2026 Awakening, Azure Can Grow Tiple-Digits
Read more
Aug 7, 2026
Gemini is Cooked but GCP is Cooking
or why DeepMind's long term failure is GCP's short term gain
Read more
Aug 3, 2026
Kimi K3, The Manos, The Mythos, The Legendos
Kimi K3’s architecture: compressed memory, attention across depth, latent expert routing, and the inference performance
Read more
Jul 29, 2026
The Wild Wild West Of LEGO Datacenters
The Labor Problem and Modularization to the Rescue
Read more
Get any SemiAnalysis model.
Start with the sales team. They come back with scoping, licensing, and pricing for your mandate.
- Scoped to your use case
- Dashboard and API access for your team
- Custom research engagements available
-
Model Releases: OpenAI, Google, xAI
OpenAI 4.1, o3, and o4-mini; Tool Use, Gemini 2.5 Flash, xAI Grok 3-mini









