GenAI economics are about to get real

We've modelled GenAI unit economics, and what we found should give any CFO pause: every big company is busy redesigning their operating model around AI (or so they claim) to chase performance and efficiency. But most are banking their gains on AI pricing that almost certainly won't last. Neither economically nor strategically.

Today's $20/month per-user AI subscriptions are deeply loss-making for providers such as OpenAI, Anthropic and Google. With the inference cost of AI, we are not in "zero marginal cost to serve" software territory anymore.

BC Strategy's Semyon Mitrofanov, CFA unpacked the P&L and capex bills of AI firms to estimate the true cost of tokens (the atomic unit of AI: small chunks of text the model processes with every query). At $6.50–$8.00 per million tokens, consumer-grade flat-fee plans are running at roughly -275% gross margin. Nobody cares if this is sustainable — it's an adoption and market-share grab.

You can already see where the real economics sit. API pricing (what developers and enterprises pay per token) clusters around ~$10 per million, close to a sustainable 50% margin for most use cases bar the lowest-latency ones, which will undoubtedly grow in usage over time.

So what happens when the "Token Honeymoon" ends?

Two forces compound. First, repricing: when prices normalise to true token economics, per-seat costs jump ~5–7×. Second, agentic AI: autonomous workflows that chain multiple AI calls to complete a single task drive another 2–5× in token consumption. A 100-seat organisation paying ~$24K/year today could face $170K at true cost, or $840K once agentic workloads scale.

And here's what transformation programs miss: by the time repricing arrives, you probably can't walk away. Your workflows have been re-engineered around a specific platform. Memory, connectors, integrations, retrained teams: these are switching costs by design. Alex Kowatsch said it before: At that point you're not a customer, you're a price-taker.

Key practical implications: reflect (higher) token pricing in your business case, segment workloads and route them efficiently, negotiate harder with AI suppliers, preserve portability and in-house capabilities, test alternatives such as open-source or private deployments, and establish usage governance.

The strategic window to build cost optionality is now — while the subsidies last.

If this is relevant to your organisation, get in touch for access to the full version of our report and our views on the strategic implications.