Every loss-leader has an expiry. AI subscription pricing is no exception.
Image: Artem Beliaikin / Unsplash.
Last week I left a comment under Melissa Rosenthal's LinkedIn post on AI subscription economics, asking (for a friend *wink*) whether the subsidised rates we are all currently enjoying might persist for super-users outside the enterprise pipeline.
Rosenthal, a former Global vice president of creative at BuzzFeed and chief creative officer at ClickUp who co-founded influence agency Outlever, had laid out the numbers in unsettling detail.
Her answer to my question was a qualified yes. Rosenthal’s sense is that the freebies might survive in some capacity, but not in a way that affects earnings or Wall Street expectations once OpenAI and Anthropic head for their IPOs.
Her State of Brand article pulls together a series of admissions that have leaked out of the AI labs in recent months.
OpenAI's VP of Product Nick Turley told the Bg2 Pod podcast that his company "stumbled into" its current subscription pricing, and likened unlimited plans to "unlimited electricity".
GitHub has announced that Copilot will move to usage-based billing on 1 June, citing agentic workloads as the reason the flat-rate model collapsed.
A 50-person team on Claude Pro looks like a USD 1,000-a-month commitment on paper. At actual token rates, the same usage costs between USD 15,000 and USD 40,000.
Some of this terrain is familiar.
I wrote in February about Jason Calacanis and Chamath Palihapitiya facing similar arithmetic out loud on the All-In podcast – Calacanis tabling a USD 100,000-a-year-per-agent burn rate at 10–20% utilisation, and Chamath concluding that AI needed to be at least twice as productive as an equivalent employee to justify itself.
The Trajectory Africa’s Tayo Akinyemi flagged a related observation in our LinkedIn thread.
She pointed to thinking in the construction enterprise software space about matching AI to probabilistic tasks (and leaving deterministic work, like compliance, elsewhere).
I put it to Akinyemi, half-joking and half not, that going human might prove the economical choice in plenty of instances, at least in the short to medium term.
It is a slightly odd thing to say out loud. But consider it seriously for a moment. A senior analyst, a careful sub-editor, a junior compliance officer, a developer who reads the codebase.
Stack their fully-loaded cost against five-figure monthly token bills for an agentic system whose outputs still need human review, and the comparison stops being obvious in the AI tool's favour.
The implications land particularly awkwardly for organisations in Corporate Africa being sold AI tools that depend on clean data most firms do not yet have, on subsidised rates that probably will not last.
Meanwhile, the consultancy bill to make those tools functional is real. The human bill to produce the same outputs properly may be smaller than the all-in AI bill once the subsidies come off… for now.
One commenter on Rosenthal's post added a sharp third dimension. Non-enterprise super-users (such as myself) are easier to subsidise because they feed the machine training data and can be monetised through ads. I am clearly the product and all that… *hides* Enterprises tolerate neither.
Which means the bill, when it lands, will land hardest where the workflows are deepest.
Andile Masuku is co-founder and executive producer at African Tech Roundup. He serves as executive editor of Future in the Humanities (FITH), powered by the SA–UK Chair in the Digital Humanities at Wits University. Connect and engage with Andile on X (@MasukuAndile) and via LinkedIn.
Image: Supplied.
Andile Masuku is co-founder and executive producer at African Tech Roundup. He serves as executive editor of Future in the Humanities (FITH), powered by the SA–UK Chair in the Digital Humanities at Wits University. Connect and engage with Andile on X (@MasukuAndile) and via LinkedIn.
Follow Business Report on Facebook, X and on LinkedIn for the latest Business and tech news.