Global Compute Spread Intelligence

Compute is made of electricity. We track industrial power prices x GPU cloud rates x LLM token prices in one margin engine, refreshed daily, so you can see where tokens are cheapest to produce and dearest to sell.

PRO yearly ¥12,990 (daily snapshots + arbitrage table) →　Monthly report ¥1,299 →　Request custom study (USD invoice) →

What you get

Deliverable	Detail
Margin matrix (monthly report)	1344 GPU x region x model rows: electricity cost and rental cost per 1M tokens vs market token price
Buy-low / sell-high arbitrage table	224 rows: lowest vs highest public GPU listing per model, cross-vendor spread, self-operate gross multiple
Daily auto-collection archive	499 price history rows over 1 days (public listings, sourced & dated per row)
PRO extras	Full history export (CSV) + one custom study (your region / vendor / model mix)
Token margin ranking	450 LLM price points auto-collected daily (OpenRouter public catalog), ranked against our cost engine for overseas resale

Sample arbitrage table (H100, full table for subscribers)

GPU	Model	Lowest buy	Highest sell	Spread	Cost /1M tok	Token price	Gross multiple
H100	Claude Opus 4	$2.72	$10.98	4.0x	$0.3778	$75.00	198x
H100	o3	$2.72	$10.98	4.0x	$0.3778	$40.00	106x
H100	Claude Sonnet 4	$2.72	$10.98	4.0x	$0.3778	$15.00	40x
H100	Grok 4	$2.72	$10.98	4.0x	$0.3778	$15.00	40x
H100	GPT-4o	$2.72	$10.98	4.0x	$0.3778	$10.00	26x
H100	Gemini 2.5 Pro	$2.72	$10.98	4.0x	$0.3778	$10.00	26x
H100	GPT-4.1	$2.72	$10.98	4.0x	$0.3778	$8.00	21x
H100	Mistral Large	$2.72	$10.98	4.0x	$0.3778	$6.00	16x

Figures are snapshots of public price listings and common engineering assumptions (throughput / PUE marked as assumptions, not measured); they exclude network, staffing and depreciation and are not investment advice.

Token margin ranking (overseas resale, TOP 8)

Model	Output price /1M	Power cost /1M	Lowest rent cost /1M	Power margin	Suggested role
openai/gpt-5.5-pro:batch	$90.00	$0.0103	$0.3778	8713x	走量
openai/gpt-5.4-pro:batch	$90.00	$0.0103	$0.3778	8713x	走量
openai/gpt-5.2-pro:batch	$84.00	$0.0103	$0.3778	8132x	走量
openai/o3-pro	$80.00	$0.0103	$0.3778	7744x	走量
anthropic/claude-opus-4.1	$75.00	$0.0103	$0.3778	7260x	走量
Claude Opus 4	$75.00	$0.0103	$0.3778	7260x	走量
openai/gpt-5-pro:batch	$60.00	$0.0103	$0.3778	5808x	走量
openai/o1	$60.00	$0.0103	$0.3778	5808x	走量

Costs assume H100 hosting in a low-power-cost region (snapshot prices; PUE and throughput are assumptions, not measured). Roles are rule-based suggestions from live data: flagship = top margin, traffic = lowest price, volume = the rest.

Who buys this

Buyer	Use case
AI startups & inference API providers	Pick hosting region and vendor; price your API margin before you commit
GPU brokers & resellers	Cross-vendor spread table = daily deal sheet
Data-center site selectors	Power price vs margin ranking for siting
Funds & researchers	Reproducible, dated, sourced price series

FAQ

Where does the data come from?

Public price pages and public marketplace APIs only, every row carries source and date; no private or scraped personal data.

How do I pay from outside China?

We invoice in CNY or USD by bank transfer - submit an inquiry and we will issue a proforma invoice with your order number.

Can we get a custom cut?

Yes - PRO yearly includes one custom study (regions, vendors, models of your choice); standalone custom studies start at ¥4,800.