What Research Models and Data Does SemiAnalysis Offer?

Updated

SemiAnalysis is an independent research firm covering the AI and semiconductor supply chain, read by 318,000+ subscribers. Beyond its Substack newsletter, it offers four institutional research products — an Accelerator & HBM Model, a Datacenter Industry Model, a Tokenomics Model, and open-source tooling and benchmarks — plus consulting and API access. The models are aimed at institutional users who need bottoms-up forecasts and supply-chain data rather than news commentary.

The Four Research Products

Accelerator & HBM Model

A bottoms-up demand forecast for AI accelerators and high-bandwidth memory, covering Nvidia, AMD, Google, AWS, Meta, Microsoft, Intel, and every major Chinese hyperscaler.

  • Granularity: SKU-level shipments and ASPs, with customer-level shipments and installed base for 60+ accounts.
  • HBM detail: tracked down to vendor, generation, layer count, and price — the same depth as compute.
  • Supply-chain link: shipments are built up from equipment requirements, wafer starts, and packaging orders, so revenue for supply-chain companies can be estimated directly from production.
  • Coverage: forecasts to 2030, history back to 2021.

This is the product to look at if your question is "how many accelerators ship to whom, and what does that imply for memory and upstream suppliers."

Datacenter Industry Model

A global datacenter capacity tracker built around structure-level megawatt modeling.

  • Scope: 6,500+ structures, including hyperscaler buildouts and colocation markets, with per-structure power draw, lease structures, and planned versus operational capacity.
  • Companies: 200+ covered, with the end customer identified behind each megawatt.
  • Demand side: accelerator shipments and installed base by operator, power drawn per deployment, and a supply-demand reconciliation showing which operators and regions run short of power, and when.
  • Delivery tracking: promised versus delivered MW, tracked quarterly, with regional site strategy.

Use this when the question is about power constraints, buildout pacing, or where capacity is actually landing.

Tokenomics Model

A unit-economics engine for AI serving: bottoms-up token throughput, ROI on AI spend, and translation of demand-side revenue into future accelerator demand by hyperscaler. It connects inference economics back to the hardware forecast — useful if you need to test whether serving revenue can justify projected compute purchases.

OSS Tooling & Benchmarks

Open-source tooling and benchmarks published under a transparency-oriented approach. These are the lowest-commitment entry point: you can inspect methodology and outputs before engaging with the commercial models.

Choosing Between Them

If your question is about… Start with
Accelerator or HBM shipments, ASPs, customer-level demand Accelerator & HBM Model
Datacenter capacity, power constraints, buildout timing Datacenter Industry Model
Inference cost, serving ROI, revenue-to-compute linkage Tokenomics Model
Methodology and benchmark transparency OSS Tooling & Benchmarks
Custom or programmatic needs Consulting or API Access

Access and Practical Notes

The site lists a newsletter subscription link, institutional login, and "Talk to Sales" / "View model" paths for each product, along with consulting and API access offerings. The available material does not state prices or whether any tier is free, so treat access terms as something to confirm directly — the newsletter is the public-facing entry point, while the models appear to be positioned as institutional offerings.

A reasonable sequence: read the newsletter and OSS benchmarks to judge research quality, then request model access or a sales conversation for the specific product matching your question.