Subsidised pricing is not a conspiracy theory, it is a well-documented growth strategy, and we have watched it run before. In the year to September 2015, Uber passengers "were paying only 41% of the actual cost of their trips", with investor money covering the rest, according to transport analyst Hubert Horan's breakdown of the company's financials (Naked Capitalism, Can Uber Ever Deliver?). By the time the habit was locked in, the cost of a ride-hailing trip had risen 92% between January 2018 and July 2021, per Rakuten Intelligence data reported by CNBC.
AI is being funded on a similar shape of money. Anthropic signed a term sheet in January 2026 for a $10 billion round at a $350 billion valuation (CNBC), and reporting on its internal projections has it reaching positive cash flow around 2028 rather than today (TechCrunch). None of that means prices are about to double next Tuesday. It means the sensible move is to know your exposure, not to guess at it.
The audit is only as good as what you feed it, and almost nobody knows their true AI spend from memory. Spend ten minutes collecting these five things first. Every one of them is somewhere you already have access to.
/usage. If you are nowhere near your limits you may be on too big a plan; if you hit them constantly, that is a real signal too (Anthropic, manage usage credits).This is the main one. Paste your list underneath it. It works in any Claude, on any plan, and it is worth turning web search on for so it can check current list prices rather than recalling them.
Act as a sceptical finance analyst reviewing my personal AI spend. I am going to paste a list of every AI subscription, app, and metered API account I pay for, with what I use each one for and what it costs me.
Build me a single table with one row per tool and these columns: tool, what I actually use it for, billing type (fixed plan or metered), current monthly cost in AUD, annualised cost, and how load-bearing it is to my work on a scale of critical, useful, or nice to have. Convert annual plans to a monthly figure so everything is comparable, and total the column at the bottom. If I have given you a metered account with no number, do not invent one: mark it "unknown, go and check" and tell me exactly where in that product's dashboard I would find it.
Then give me three short sections. First, the two or three line items where I am paying the most for the least, with your reasoning. Second, anything I am paying for twice, where two tools do substantially the same job. Third, the single biggest risk in this stack if prices rise, meaning the tool that is both expensive and hardest for me to leave.
Ask me up to four clarifying questions before you build the table if my list is missing something you need. Do not recommend that I cancel anything yet, that comes later. Flag every number you were not certain about rather than smoothing it over.This is the part people skip, and it is the part that changes decisions. A number that is fine at today's price and painful at double is a number you want to know about while you still have options.
Take the table you built and model what happens to my AI spend under three scenarios: prices stay flat, prices rise 2x, and prices rise 3x. Assume fixed subscription plans rise with the headline price and metered usage rises with the per-unit price while my volume stays the same.
Show me the monthly and annual total under each scenario, then tell me at which scenario my total crosses the point where I would genuinely have to change how I work. For each tool, say whether a price rise would be an inconvenience or a real problem for me, and be specific about why.
Finally, rank my tools by "pain per dollar if the price doubles", meaning how badly I would feel the increase relative to what I get from the tool. Put the ones I should be actively reducing my dependence on at the top of that list, and explain in one sentence each what my realistic alternative would be.Overlap is the cheapest saving available, because you lose nothing. Most stacks that grew organically over a year have two or three tools quietly doing the same job.
Go back through my stack and group the tools by the job they actually do for me, not by the category they market themselves as. Writing, image generation, video, transcription, research, scheduling, automation, and so on.
Wherever two or more tools land in the same group, tell me plainly which one to keep and which to cancel, and give me the reason in terms of my use case rather than general feature comparisons. If keeping both is genuinely justified, say that instead of forcing a cut, and explain what each one does that the other cannot.
For every cancellation you recommend, tell me three things: what I would lose, what I would need to set up in the tool I am keeping to cover the gap, and roughly how long that setup would take. Then give me the list in cancel-this-week order, cheapest and least disruptive first, so I can act on it today.The real cost of a price rise is not the money, it is being unable to leave. Portability is what turns a price hike into an annoyance rather than a crisis, and it is mostly a filing habit rather than a project.
For each tool in my stack, tell me how locked in I currently am, and treat "how hard is it to leave" as the only question that matters here.
For each one, list what of my work lives inside that product and only inside that product: prompts, saved instructions, documents, generated assets, workflow configuration, customer data. Then tell me exactly how I export each of those things, naming the menu or setting where the export lives, and what format it comes out in.
Rank the tools from most portable to most locked in. For the three most locked-in, give me a specific 30-minute task I can do this week to reduce that lock-in, and be concrete: the file I should create, where it should live, and what should go in it. Do not suggest switching tools, suggest making it possible to switch.Audit against real published prices, not vibes. These are Anthropic's own listed numbers, and both pages are worth opening yourself because they do change.