AI Token Economics

Assumptions

Planning mix used to allocate the 1 trillion monthly token budget across AI use-case patterns, as per the reference model. Placement and per-pattern token volumes drive the on-premises vs. cloud split in all financial calculations.

Financial Model Assumptions

Key parameters used in the default model. All are editable in the Calculator tab.

CapEx & Facility Toggle — On-Prem GPU Guiding Rule

The Calculator tab includes an Include Fusion HCI CapEx & Facility Costs toggle. When unchecked, Fusion HCI CapEx, CapEx Amortisation Period and Power/Cooling/Data Centre are excluded entirely from the IBM Hybrid Rate and total cost — there is no on-prem GPU hardware to purchase, amortise, power or cool. Hybrid Networking Overhead is set to $0 and Operations Staff halves from $80K/yr to $40K/yr, both reflected directly in their edit boxes. To keep the model honest, the calculator adds back an On-Prem GPU Adjustment: at a sustained 640B tokens/month on-premises reference volume, the GPU compute cost differential between Fusion HCI and equivalent cloud capacity is estimated at $3–6M/month (avg $4.5M) in favour of on-premises HCI deployment. This differential — scaled to the actual on-prem token volume — is added to the IBM Hybrid OpEx when the toggle is off, so the Hybrid Rate rises to reflect the lost on-prem GPU cost advantage rather than appearing artificially cheaper.

IBM Software — Value & Benefit Assumptions

Each IBM software product's modelled financial benefit is gated on its licence cost being greater than $0. Setting any licence to $0 in the Calculator has two effects on the IBM Hybrid total: an upward impact (savings improve) because the licence cost drops out of Software Licences, and a downward impact (savings worsen) because the corresponding modelled benefit below is no longer captured. In most cases the lost benefit outweighs the licence saving.

References & Citations

Sources underpinning the financial models, technology assessments, and industry benchmarks in this analysis. IBM product documentation current as of June 2026.

Reference Architecture

Vendor-neutral hybrid AI infrastructure architecture derived from the token workload mix and placement model above. Functional capability layers are shown in place of specific products, so the design maps onto any compliant vendor stack.

User Manual

This calculator compares the total cost of running enterprise AI inference two ways: entirely through cloud LLM APIs (“Cloud-Only”) versus a hybrid deployment that blends on-premises IBM Fusion HCI infrastructure with cloud burst capacity (“IBM Hybrid”). Enter your token volume, workload split, and cost assumptions, and the tool calculates annual costs, a multi-year comparison, per-token rates, and a CapEx payback period — all updating live as you change inputs. This page walks through every tab and input.

Calculator Tab — Token Volume

The Calculator tab has three input groups — Token Volume, Cloud-Only Costs, and Hybrid Infrastructure Costs — followed by action buttons and, once you've run a calculation, the Results section.

Monthly Token Volume is a combined dropdown-and-text-field control. The dropdown offers four presets — 1 Trillion, 500 Billion, 100 Billion, and 50 Billion tokens/month, the last of which is selected by default — and the field next to it can also be typed into directly for any custom volume. A helper line below always shows the value in words (e.g. “= 250 Billion tokens”). Every volume-dependent cost input scales linearly with whatever value is in this field, relative to a fixed 1-Trillion-token reference point (not the same thing as the default — see below) — this applies whether you pick a preset or type any custom number directly: 1 Trillion sets fields to the full reference values, 500 Billion sets them to 50% of reference, 100 Billion to 10%, 50 Billion (the default) to 5%, and e.g. 2 Trillion would double them. Typing a value that doesn't exactly match a preset switches the dropdown to Custom, but the same scaling still applies. Two explicit exceptions to this linear rule: Fusion HCI CapEx is fixed at $1,000,000 specifically at the 50B preset (not the linearly-scaled $600,000), and Operations Staff plus all 6 IBM Software Licences are pinned to identical dollar figures at exactly 50B and exactly 100B (they don't double between those two presets the way every other field does) — both scale normally at every other volume point, including any custom value that isn't exactly 50B or 100B.

Power/Cooling/Data Centre and the IBM Software Licences total aren't in either column directly — they're derived fields that automatically recompute from Fusion HCI CapEx and the six licence inputs respectively, so they follow the scaling too.

* Fixed at $1,000,000 specifically at the 50B preset, overriding the linearly-scaled figure. † Pinned to the identical dollar figure at exactly 50B and exactly 100B — scales normally at every other volume point.

On-Premises Token Share is a slider (0–100%, default 60%) controlling what fraction of eligible tokens are served on-prem via Fusion HCI versus routed to the cloud as a “burst.” (“Eligible” excludes the Content/Code Generation workload, which is priced separately.) A live readout beneath the slider shows the actual token counts implied by the current percentage and token volume — for example, at the 50B default: “22,500,000,000 tokens/mo on-prem · 15,000,000,000 tokens/mo cloud burst.”

Calculator Tab — Cost Inputs

These four inputs define the “do-nothing-hybrid” Cloud-Only baseline every IBM Hybrid scenario is compared against.

Include Fusion HCI CapEx & Facility Costs is a toggle, checked by default. When checked, CapEx, its amortisation, and Power/Cooling costs all count toward the IBM Hybrid total. When unchecked, those three inputs are excluded entirely — modelling a no-CapEx, GPU-less deployment — and the tool instead adds an “On-Prem GPU Adjustment” cost line to reflect the lost cost advantage of not owning GPU hardware. Unchecking it also drops Hybrid Networking Overhead to $0 and halves Operations Staff from $80,000/yr to $40,000/yr.

IBM Software Licences is a sub-group of six individually priced products. Each has a default annual cost and a modelled financial benefit that only applies while its cost is above $0 (setting it to $0 removes both the cost and the benefit). The six feed an auto-summed, read-only IBM Software Licences (Annual) total, and all six scale with the token-volume preset — except that they're pinned to identical dollar figures at exactly the 50B and 100B presets (they don't double between those two the way every other volume-linked field does).

Calculate Savings runs the calculation immediately, though you rarely need it — every input change (typing, sliders, toggles, dropdowns) triggers a recalculation automatically after a brief pause. Reset to defaults restores every field, including the Token Volume preset, back to its original 50-Billion-token default.

Understanding the Results

Once you've entered inputs, the Results section appears below the form.

Disclaimer Popup & Consent

On every visit — a new tab, a refresh, coming back later — a disclaimer popup covers the whole page and everything is locked until you click Agree. This applies to every visitor, signed in or not; there's no "don't ask again." Agreeing logs a new record with the timestamp, your IP address, browser user agent, and the exact disclaimer text — plus your email and sign-in method if you happen to already be signed in. Anonymous visitors are logged too, just without an email.

Once past the popup, two sign-in options appear in the top navigation: Sign In with Google, the normal path for any user, and Sign in as Admin, which prompts for a masked admin code and is the only way to reveal a Consent Log tab in the page tab bar above — every consent event from every visitor, admin-only, with date/time, email (blank for anonymous visitors), sign-in method, and IP address. The tab button itself only exists for an admin session; it isn't just hidden content on another page.

Saving & Loading Scenarios

Once signed in (Google or Admin), a Save This Calculation button appears below the results, letting you save the current scenario with an optional label.

The Reports button in the top navigation opens your saved scenarios in a table (date, label, tokens/month, on-prem %, 3-year and 5-year saving, payback, notes). From there you can edit a label or notes inline, click Load to restore any saved scenario back into the calculator, Delete a scenario, Refresh the list, or Export CSV for use outside the tool. Saved calculations are private to the account that saved them — no other user can see them.

The Other Two Tabs

Architecture shows a vendor-neutral reference architecture diagram — consumption layer, orchestration/AgentOps control plane, the AI workload pattern mix, a placement policy engine, the on-premises and public cloud domains, and cross-cutting observability/FinOps/governance planes — plus a table mapping each IBM product in this tool's model to example alternative vendors.

References & Assumptions documents the two pillars the default numbers are built on: the AI Pattern Token Planning Mix (how the token budget splits across nine workload types and where each is placed) and the Financial Model Assumptions (a quick-reference grid of every default parameter, all editable in the Calculator tab), plus a sourced citation list.

Tips

All dollar-value fields accept typed numbers with or without commas (e.g. 12000000 or 12,000,000) and format themselves automatically as you type. Every figure in this tool is a modelled estimate for planning purposes — validate anything you intend to act on against current IBM commercial proposals and your own cloud billing data before making a decision.