Part 2 of Show Me the Market | By Vishal | October 2026
In my previous piece, I argued that AI cannot pay for itself at current pricing tiers. Roughly $3 trillion of debt and obligations sits behind the industry, and a subscription-based market, modelled generously, comes to tens of billions to low hundreds of billions of dollars a year. The gap is too large to close with seats at $20 a month. Neither is the expectation that each individual can sustain multiple 20$ subscriptions – that leaves the B2B revenue stream. That, while looking attractive, is, on deeper analysis, not quite as compelling a case for AI sustenance. The B2B industry operates on a concept of share of profits or percentage of revenue or a similar metric when it comes to technology spends – and they have to spread it across multiple solutions, with being but a share of it.
Clearly something has to give – either the models have to change, or the market has to crash – or even both. Current approaches are simply not sustainable. Early indicators suggest the industry is arriving at the same conclusion. OpenAI has cut the usage allowance on its $200 Pro plan and added a $500 tier above it. Does that change anything? Hard to say. We are in untested waters, in speculative territory. But it introduces some interesting paradigms.
The first price signal
In late September, OpenAI changed the plans for its heaviest ChatGPT users. The $200 Pro plan now includes half as much usage for coding and agent-style work, effective October 30: 10 times what the $20 Plus plan allows, down from 20 times. Weekly messages to its top reasoning model fall from 200 to 100. Above it sits a new $500 plan with 25 times the Plus allowance (reporting by TechCrunch and The Next Web).
Here is what that means for the price of each unit of usage. Treat what the $20 Plus plan allows as one unit:
- Plus: $20 for 1 unit, so $20 a unit.
- Old $200 Pro: 20 units, so $10 a unit.
- New $200 Pro: 10 units, so $20 a unit.
- New $500 plan: 25 units, so $20 a unit.
Before the change, heavy users got a bulk discount: half the price per unit. Now the discount is gone and every tier costs about $20 a unit. For the heaviest users, that doubles the effective price while the $200 sticker price stays the same. This covers only the coding and agent allowance, and a unit measures capacity, not value delivered. One vendor is not the whole industry. But it is the first visible move, and it points one way.
Untested waters
There is no price-volume curve for AI at scale. Nobody knows how adoption responds when prices rise, because prices have been subsidised since the start.
Here is why the unknown matters. In my first piece, the gap between the market and the debt came to roughly 10x: the industry would need about ten times today's revenue. Raising prices is the obvious lever, so the question is how far it can take us.
The logic is simple. Revenue is the price multiplied by the number of paying customers. When the price rises, some customers leave, and how many is the whole question. Take 1,000 customers paying $20 a month, which is $20,000 a month in total. To reach ten times that, $200,000 a month, here is what each case needs:
- Nobody leaves: the price must rise 10x, to $200 a month.
- Half leave (500 stay): the remaining customers must pay $400 a month each, a 20x rise.
- Nine in ten leave (100 stay): the remaining customers must pay $2,000 a month each, a 100x rise.
- Ninety-nine in a hundred leave (10 stay): the remaining customers must pay $20,000 a month each, a 1,000x rise.
The more customers leave, the less realistic the price becomes. So raising prices alone can only close the gap where customers stay, such as tools already built into a business's daily workflow. Nobody knows how many would leave at each price, because prices have been subsidised so far. That unknown is the point. These are illustrative examples, not a forecast.
Where the money has to come from: B2B
Most of the revenue will have to come from businesses, not consumers. That changes the per-customer arithmetic in my first piece, which assumed a 50:50 split.
Reported revenue mixes point the same way, though they differ by company. Press and analyst reports put about 80% of Anthropic's revenue at business and API customers. OpenAI has been consumer-led, with enterprise reported at around 40% of revenue and rising. ChatGPT has roughly 900 million weekly users, but only about 5% are reported to pay. On that evidence I assume a 70:30 split between B2B and B2C. It is an assumption, not a measured industry figure.
Take the $834B a year needed to service about $3 trillion of debt. At 70:30, businesses must supply about $584B and consumers about $250B. Spread across the same eight markets as before:
- Consumers: if 5%, 10% or 15% of people pay, each must spend about $1,800, $900 or $600 a year. That is $150, $75 or $50 a month, against a typical plan price of $20.
- Businesses: if 5%, 10% or 15% of firms adopt, each must spend about $354K, $177K or $118K a year.
Compared with a 50:50 split, the load on each consumer falls by about 40%, and the load on each firm rises by about 40%.
What this means:
- The business bill is the real problem. In my first piece I assumed about $5K a year per adopting firm, mostly small firms. At a 70% share, each adopting firm would need to spend roughly 24 to 70 times that.
- The money must come from labour and services budgets. Support, coding and back-office work are where those budgets sit, not in software budgets. My India map points the same way: enterprises are about 78% of a domestic AI tools market of roughly $5B in the mid case.
- Business revenue is concentrated, and loyalty is low. One lab claims more than 1,000 customers spending over $1 million a year. One analysis reports that 79% of firms paying for one leading lab also pay for another. Buying from several vendors gives businesses bargaining power, which caps price rises.
So the price shock lands hardest on businesses, and that is where the question of the market has to be answered first.
Six things that could change at once
Changing one parameter will not be enough. Too much is at stake, and the investment is too large. What is needed is closer to a new ecosystem. Here is where I think the shifts will come from.
1. Pricing tiers will evolve. Flat $20 subscriptions do not cover the cost of heavy users. Expect usage-based tiers, outcome-based pricing and ad-funded free tiers (ChatGPT's free and low-cost Go tiers already carry ads).
2. Debt structures will shift. Fixed-rate debt against uncertain revenue is dangerous for pure-play AI companies. Convertibles, debt-for-equity and revenue-linked instruments are legitimate alternatives. But they change who bears the risk and when. They do not change the underlying economics. They pay off when revenue grows, which is the case where they are needed least, and in a downturn they behave like debt.
3. Off-balance-sheet financing will grow. It is already visible. A reported $1.65 trillion of AI-related obligations sits off the balance sheets of the five largest hyperscalers, alongside about $1.35 trillion of on-balance-sheet debt. These structures are legal. They are also opaque, and they deepen the contagion channel I described in the first piece.
4. Insurance and risk will become central. Insurers have started excluding generative AI from standard liability policies, while specialist insurers add affirmative cover. Whoever can price AI risk, including correlated failures across many customers of the same model, becomes a gatekeeper for enterprise adoption.
5. Frugal engineering will matter. Smaller models, routing work to cheaper models, distillation and efficiency gains reduce the cost structure itself. India has a track record here, with UPI, Jio's cost structure and Mangalyaan as examples of doing more for much less.
6. Go-to-market will be rebuilt. Vertical, outcome-priced applications sold through the IT majors and specialist startups, rather than horizontal chat subscriptions.
Standard frameworks have no precedent here
I stress-tested the standard approaches and found nil battey sannata, a blank. The arithmetic of unit economics still binds, which is exactly why the numbers do not close. What lacks precedent is the answer. Standard frameworks do not have a ready one, and that calls for a leap of faith, or a different approach, with attendant risks. The off-balance-sheet moves of the biggest players are legal but risky, and they deserve attention alongside cost structures.
The chasm
Suppose current volumes are driven by early innovators and early adopters. At some point the gap between them and growth-phase adopters has to be crossed. That is the classic chasm of technology adoption, and AI has not crossed it at current prices.
Two things have to be sorted out before it can be crossed: the price-volume curve and the price-value equation. Standard frameworks will not settle them. It will take frugal engineering, innovative product design and flexible go-to-market, set within smart financial and risk management, to monetise and find the path to profit.
No technologist can solve this alone, and no marketer can either. It needs people who work in both. Price discovery for AI means willingness-to-pay research by segment, cohort tests of limits and prices, tracking downgrades and churn, anchoring price to the value delivered rather than to the cost to serve, and usage data to know what each customer costs.
What to watch
• Whether other AI vendors follow with similar tier changes, and whether the $20 tier moves.
• Downgrade and churn rates after tier changes, where they are disclosed.
• Refinancing terms and spreads on AI-linked debt.
• How insurers treat AI in professional liability cover.
• How much further off-balance-sheet obligations grow.
The company that works out the price-volume curve first, before its cash runs out, is the one likely to survive.
Caveats
• The elasticity figures are illustrative arithmetic, not estimates of real demand.
• The debt and obligation figures ($1.35T and $1.65T) are reported estimates, and the off-balance-sheet portion is commitments, not all borrowings.
• The OpenAI plan details are as reported by the press, and may have changed.
• This is analysis, not investment, legal or insurance advice.
Conclusion
That being said – the above is only a very simplified back of the envelope illustration meant to point out that the current pricing and business models do not seem to be sustainable basis the above calculation. The AI industry needs to demonstrate a higher value to the business so that a higher share of wallet in their spends and budgets can be justified – only then will businesses pay. And therein lies the biggest lacuna – the AI industry is stuck in proving how smart it can be, not in figuring out how to add value, actual real value, to real world businesses.
And that requires a marketing-led approach, not a technology centric approach. The industry will have to work, and work closely, with businesses across the board in order to understand their pain points, figure out how it can add value, and implement smoothly. And, in the process – some existing solutions will become obsolete; so disruption is anyway certain. Only question is will it be this positive disruption – or the mayhem-based disruption of a market collapse?
Comments
Post a Comment