Size AI infrastructure in the currency that matters.

Every other unit — racks, GPUs, capex dollars, lease months — is an input. Tokens are the output your customers actually pay for. A sizing exercise that minimises chip count or rack density but leaves you with $0.80/M-tok when competitors are at $0.40 is a losing plan.

We built Mintok because no existing tool ties the four axes — inference sizing, reference architecture, cluster economics, and contract structures (lease vs CapEx vs cloud) — into a single optimisation. People do it in spreadsheets that drift the moment a new chip drops.

• Three constraints, not one. Compute, memory, and bandwidth all bind at different points across model + workload combinations. A sizing tool that ignores any of them ships the wrong fleet.
• Contract is part of the chip. A B200 on a 36-month lease is different compute economics than the same B200 on a cloud hourly rate. Treating them as the same chip is a six-figure mistake.
• Spreadsheets are not infrastructure. Sizing, demand, supply, and economics should live in one source of truth that updates as chips, models, and contracts change.
• $/M-token is a portfolio metric. Optimising one workload doesn't move the needle. Optimising every workload across every chip across every contract — does.
• Analysis has to ship as artifacts. Whether you're committing to a chip order or briefing a customer, sizing math is only useful if it flows to the next decision without re-keying. Mintok closes that loop with engagement-level briefs that carry the math forward.

Mintok is in private alpha with two design-partner groups:

• Fleet operators — neoclouds, hyperscalers, and sovereign-AI builders sizing their own infrastructure and committing to orders.
• FDE consultants — forward-deployed engineers running sizing and cost engagements for customer fleets, delivering briefs that compare chips, contracts, and TCO without re-keying spreadsheets.

Both flows share the same chip, model, and economics engines — they differ in what they ship at the end: a procurement decision, or a customer-facing brief.

If you're running an AI inference platform — or advising someone who is — and the answer to “what's your cost per million tokens, by model, by chip, by contract?” is “let me get back to you,” we should talk.

General questions about Mintok? info@mintok.ai