How this calculator came about
This calculator was built on 12 June 2026 in a working session with Claude Fable 5 (Anthropic) in the Cowork mode of the Claude desktop app. The AI first compiled the current investment, user and revenue figures via web research (hyperscaler capex, venture funding, subscriber numbers), developed a simplified annuity model from them, cross-checked the calculations programmatically in Python and finally built this page in the School's design.
Version 2 added: separate depreciation by asset class, an electricity-price slider, a sensitivity analysis (tornado), a Monte Carlo simulation with an uncertainty band, a timeline view with break-even year and industry IRR, the comparison with historical megaprojects, plus shareable scenario links and a print export. The model deliberately stays simple: it is not meant to deliver a forecast, but to make orders of magnitude comprehensible. All sources and assumptions are disclosed below.
Your assumptions
Nine sliders in four areas – pick an area on the left, adjust on the right. Every change recalculates instantly and lands in the URL.
AI revenue arises mainly on a usage basis (API/enterprise), not through $20 subscriptions.
Blended price = 3:1 list price (output:input) of a flagship basket; our own compilation from public prices – order of magnitude, monthly snapshot. Context: a16z „LLMflation“, Epoch AI. Tangible assumptions: ~0.75 words/token, ~225 words/min reading, world population 8.1bn.
Simplified annuity model for putting orders of magnitude into perspective. Not a forecast, not investment advice. Data as of: 12.06.2026.
AI capex of the big tech companies
bn $/yr · Amazon, Microsoft, Alphabet, Meta, Oracle
Required vs. actual AI revenue
bn $/yr
Required monthly price by number of subscribers
$/month, logarithmic scale · curve follows your assumptions
Which slider tips the result? (Sensitivity analysis)
Range of the required monthly price when, in each case, one assumption is varied from "favorable" to "unfavorable" – all others stay at your values.
How uncertain is the result? (Monte Carlo simulation)
3,000 random scenarios around your settings – instead of a point estimate, an honest uncertainty band.
The timeline: when does the build-out pay off?
Instead of the static annual calculation: investment path against revenue ramp, discounted with your WACC. Break-even = the year in which the cumulative, discounted gross profit overtakes the cumulative, discounted investments.
Perspective: the AI build-out next to history's megaprojects
Total investments, roughly inflation-adjusted in bn $ of today – estimates for perspective, not exact statistics.
Cheaper tokens – the end of the story? Not necessarily.
Falling token prices sound like the death of margins. Three counterforces explain why capital providers are nonetheless optimistic – entirely without price-fixing:
App, agent and SaaS firms price by outcome or seat, not per token. If token costs fall, their cost of goods (COGS) drops while their selling price holds → their margin rises. This is exactly where the VC money sits. The price that doesn't fall as fast as the costs is that of the application – no collusion needed, but a different layer with differentiation and switching costs.
Cheaper means massively more usage. A thin margin times exploding volume can still let the total gross profit grow.
The newest models retain temporarily pricing power: „costs fall“ applies to yesterday's performance – today's frontier still costs. A moving target, not coordination.
The flip side – and the reason for this calculator: If value migrates into the capital-light application layer and the token price is competed away, it is precisely the capital-intensive infrastructure and model providers (who shoulder the trillion-dollar capex) who may not earn their return. It is exactly this tension that the calculator depicts.
The calculation path – step by step
| Step | Logic | Value |
|---|
Why this calculator is needed
AI is talked about in superlatives – but rarely in units a human can grasp. "$725 billion in capex" is a headline; "your subscription would have to cost over $1,000 a month if today's payers had to bear the bill alone" is a thought you can think through to the end. This is exactly the translation the calculator delivers: it breaks the most abstract figure of the present down to the most concrete – the price you yourself would pay.
This is more than a numbers game. Anyone who wants to understand whether the AI investment wave is a bubble or the build-out of a new base infrastructure has to see through exactly this mechanism: capital costs a return, hardware ages in a few years, and between revenue and profit lie power and compute costs. From these three sober facts it follows inevitably that somewhere in the world economy a triple-digit billion amount of new value must arise every year – through subscriptions, through enterprise solutions or through productivity. Whether that succeeds is perhaps the most important economic question of this decade.
And finally, the calculator is a piece of empowerment: instead of adopting a ready-made opinion ("bubble!" or "revolution!"), you can adjust the assumptions yourself and see when the math tips. Anyone who has once experienced firsthand that the result swings between $35 and $1,200 – depending on what you believe about subscriber numbers and margins – reads every AI headline differently afterwards. This is exactly the attitude we want to convey at the School: don't believe, calculate. Dare to think.
The assumptions in detail
Starting values of the base scenario – all adjustable via slider. The starting values are loaded from a separate data file and maintained quarterly.
| Assumption | Starting value | Rationale |
|---|---|---|
| Investment base | $2.0tn | Cumulative AI investments 2023–2026: AI share of hyperscaler capex (~$1.1tn: 2024 ≈ 192, 2025 ≈ 332, 2026e ≈ 544bn $), venture funding (~$0.6tn: 2024 = 114, 2025 = 211bn $), rest chips, energy and independent data-center operators. Deliberately conservative – Goldman Sachs expects a further $7.6tn in 2026–2031 alone. |
| Cost of capital (WACC) | 10 % | Typical return expectation for tech investments; venture capital expects considerably more, bonds less. |
| Useful life chips/servers | 5 years | Hyperscalers depreciate over 4–6 years; critics (among others on GPU devaluation through new generations) consider 2–3 years more realistic. |
| Useful life buildings/networks | 15 years | Data-center shells, power and cooling infrastructure are long-lived – that dampens the capital burden. |
| Hardware share | 60 % | The bulk of AI investments goes into chips and servers (short-lived), the rest into long-lived infrastructure. |
| Gross margin | 50 % | AI services have high variable costs (power, inference compute, operations) – well below classic software margins of 80–90%. |
| Electricity-price index | 100 % | Energy ≈ 15 revenue points in the base case. If the electricity price rises by 50%, the effective margin falls by ~7.5 points – power is the fastest-growing cost block of the build-out. |
| Subscription share | 40 % | Consumer subscriptions bear only a part; enterprise and API revenue grows faster (Anthropic: ~$47bn run-rate almost without consumer business). |
| Paying subscribers | 300m | Real today ~80m worldwide (ChatGPT: 50m paying out of 900m weekly users = ~5–6% conversion). 300m assumes almost a quadrupling. |
| Revenue growth (timeline) | 35%/yr | Between historical hypergrowth (Anthropic: 80× in two years) and mature software (~15–20%). Calculated as constant – a deliberate simplification. |
| Today's AI revenue | ~$175bn/yr | Rough sum mid-2026: OpenAI ~29, Anthropic ~47 (run-rate), plus AI revenue from Google, Microsoft, Meta & Co. |
Sources
IEEE ComSoc: Hyperscaler-Capex 2024–2026 · Goldman Sachs: Tracking Trillions – assumptions of the AI build-out · CNBC: Big Tech capex above $1tn from 2027 · AI Funding Tracker: $211bn venture funding 2025 · Crunchbase: funding of the foundation-model providers · Roic: ChatGPT – 900m users, 50m paying subscriptions · CNBC: Anthropic revenue trajectory · Bain & Company: $2tn annual revenue needed by 2030 · Tom's Hardware: Bain's $800bn gap · Al Jazeera: AI build-out vs. historical megaprojects · IMF World Economic Outlook: nominal GDP by country (2025)
Researched on 12.06.2026. The calculator's starting values reside in the file rechner-ai-breakeven-daten.json and are updated quarterly – the sliders let you insert your own values at any time. Methodological questions? Our sources & methodology page.