Tokenomics Model
The hardware inputs of AI connected to the software outputs built on top of it. The first demand-side model of AI translates revenue and usage growth into future hardware demand across Nvidia, AMD, Google TPU, and Amazon Trainium fleets, for everyone following the money trail.
- Scope
- Demand side
- Build
- Bottoms-up
- Workloads
- 4 profiled
- Revenue lines
- 3 modeled
The multi-trillion dollar question, made calculable.
The SemiAnalysis Tokenomics Model connects the hardware inputs of AI to the software outputs of the services built on top of AI models. It completes our end-to-end coverage of the AI and chip industry with the tools and metrics to calculate the ROI on AI spend, adoption of AI by use case, and the growth of new business models enabled by AI.
Investors, corporates, and policy makers following the money trail of AI can disentangle the economic relationships between the players in the AI value chain and answer the question the whole buildout hangs on: what is the ROI and profitability of AI business models?
Who uses it, and what it decides.
The buyers of this model, and the calls they make with it.
- Public & private market investors
- Positioning across the AI value chain with a demand-side view: which business models actually earn a return on the compute they buy, and whose revenue holds up as token economics shift.
- Corporate strategy & finance teams
- ROI on AI spend before the budget commits: adoption by use case, the unit economics of token consumption, and where AI spending actually pays back.
- Policy makers & economists
- The token economy sized and followed: where AI value accrues, what the money trail funds, and how usage growth turns into infrastructure demand.
- Software vendors & AI-native builders
- The unit economics of the new software model: what serving costs per token, what the market bears, and how token consumption pricing disrupts the seat-based incumbents.
From installed base to income statement.
The fleet is counted, its token output is modeled from first principles, the token economy is sized, and the loop closes back into hardware demand.
The fleet, counted
A detailed forecast of the AI hardware installed base and future investments, buyer by buyer:
Hyperscalers
Microsoft, Google, Amazon, Meta, and Oracle, with installed base and future investment tracked per buyer
Foundation labs
OpenAI, Anthropic, and DeepSeek as compute demand sources, tracked alongside Thinking Machines and peers
Neoclouds
Coreweave, Nebius, Crusoe, and the quality neocloud supply base as measured by our ClusterMAX ratings
Tokens per second, from first principles
Bottoms-up token throughput forecasts built from the three things that set them:
Hardware systems
GB200 NVL72, VR144, CPX, TPU v7, Trainium 3, and the systems that follow them
Model architectures
GPT 5, Sonnet 4, DeepSeek V3, Kimi K2, and the architecture choices that move throughput
User workloads
Coding, chat, document analysis, and agentic behavior, each with its own token profile
The token economy, sized
Addressable market analysis across everywhere tokens are consumed:
In-app usage
Existing application token usage across Google AI Overviews, ChatGPT, Grok, and Meta AI
API endpoints
Inference endpoints for ChatGPT, Claude, Qwen, DeepSeek, Llama, and the rest of the serving market
Token-native software
Cursor, Windsurf, Harvey, Perplexity, and the emerging startups whose COGS is tokens
The money trail, closed
Usage and revenue growth translate back into aggregate hardware demand: inference demand from adoption, training demand from architecture development and expected returns, ROIC by deployer, and revenue and profit forecasts across rental, model, and software lines, including the disruption of seat-based SaaS by token consumption pricing at companies like Salesforce, Workday, Adobe, SAP, ServiceNow, and Atlassian.
Follow the money trail.
AI economics is a loop. Capex buys compute, compute produces tokens, tokens earn revenue, and the returns decide the next order of hardware. This model is the first to close that loop from the demand side.
The loop · Capex
Fleets get bought
Hyperscalers, foundation labs, and neoclouds put capital into accelerator fleets.
In the model: installed base and investment forecasts, buyer by buyer
The loop · Compute
Fleets get scheduled
GB200 NVL72 to TPU v7 turn capex into serving capacity, set by hardware and architecture.
In the model: bottoms-up tokens per second by system and architecture
The loop · Tokens
Tokens get consumed
Chat, coding, document analysis, and agentic workloads burn tokens through apps, APIs, and token-native software.
In the model: the addressable market of the token economy, channel by channel
The loop · Revenue
Returns close the loop
Rental, model, and software lines split the take, and ROIC decides the next order of hardware.
In the model: revenue, profit, and ROIC by business model, translated into future GPU demand
The ring is a map, not a chart. In the model, every turn of the loop carries numbers: installed base by buyer, tokens per second by system, consumption by channel, and ROIC by business model.
How the model is built.
Supply of compute, production of tokens, and the demand that pays for both, forced to agree.
-
Count the compute
Installed base and investment forecasts across hyperscalers, foundation labs, and neoclouds set the supply of AI compute, with quality graded by ClusterMAX.
-
Model the tokens
Bottoms-up throughput per hardware system, model architecture, and user workload turns compute into token supply, and usage tracking turns applications into token demand.
-
Close the loop
Adoption, revenue, and ROIC translate back into future hardware demand, split into inference and training, so the money trail runs end to end.
Research that ships with the model.
Model subscribers receive the update notes, webinars, and analysis published against each release. A sample of recent coverage:
Common questions.
Anything not covered here, ask the team directly through the form below.
What does the Tokenomics Model include?
Installed base and investment forecasts for hyperscalers, foundation labs, and neoclouds, bottoms-up token throughput by hardware system, model architecture, and workload, addressable market analysis of the token economy, tracking of SaaS disruption by token consumption pricing, ROIC of AI deployments, and aggregate inference and training hardware demand with revenue and profit forecasts across rental, model, and software lines.
How is the model built?
Demand side first. Token consumption is tracked across applications, API endpoints, and token-native software, and set against bottoms-up token production per hardware system and model architecture. Usage and revenue growth then translate into future hardware demand and ROIC by deployer, closing the loop between AI spending and AI income.
How is the model delivered?
As an Excel workbook with dashboard access, including one year of quarterly updates, an onboarding call with the team to walk through the model and methodologies, and ad-hoc calls for questions that come up in use.
Is it part of the SemiAnalysis newsletter subscription?
No. Industry models are separate institutional offerings and are not included with the annual newsletter membership.
Can it translate AI usage into GPU demand?
Yes. It is the first demand-side model of AI, built to translate revenue and usage growth into future hardware demand across Nvidia GPUs, AMD GPUs, Google TPUs, Amazon Trainium, and more, split into inference demand from adoption and training demand from model development.
Models that pair with this one.
Tokenomics is the demand side of a chain the other models cover link by link: the silicon, the facilities, and the cost of running both.
AI Cloud TCO Model
Supply-side twinThe cost of producing a token: what a GPU-hour costs to rent and run, cluster by cluster.
View model
Accelerator & HBM Model
The siliconThe accelerator shipments that the hardware demand this model forecasts ultimately orders.
View model
Datacenter Industry Model
The facilitiesWhere the fleets serving these tokens get built, powered, and energized, site by site.
View modelGet the Tokenomics Model.
Start with the sales team. They come back with scoping, licensing, and pricing for your mandate.
- Scoped to your use case
- Onboarding and ad-hoc analyst calls included
- Custom research engagements available
