For most small businesses, AI arrived as a simple deal: about £20 a month per person, and you use it as much as you like. That deal is quietly changing.
What’s changing
OpenAI is moving to a metered allowance. According to an OpenAI help centre article reported at the end of October, from December its Plus, Pro and Business plans will share a single allowance measured in tokens, covering both ChatGPT and its Codex coding tool. When you hit the limit, you can wait, upgrade or buy more credits. OpenAI hasn’t yet said how big the allowances will be.
Usage is climbing fast. Figures from the spend-management firm Ramp show that in the week to 23 October, business token volume was 37.1% higher than a month earlier, while token spending was 8.9% higher. The price per token keeps falling, but people are using so much more that the total bill still goes up.
Even the biggest companies are rationing. Bloomberg reported that in October Apple cut the daily AI spending cap for some of its engineers from $300 to $200, with further cuts possible. If one of the richest companies in the world is watching its AI meter, small firms should be too.
Why this matters for a small business
A token is roughly a word or part of a word. Every question you ask, and every document the AI reads to answer it, uses tokens.
That’s the catch. The jobs that make AI really useful in a business are the ones that use the most tokens:
- reading a 40-page specification to find one tolerance
- comparing a new enquiry against years of past quotes
- digging through job history to see how a similar install went
- reading technical drawings and site photos
With a metered plan, the more useful AI becomes to your team, the more it costs. That leads to three problems.
- Budgets become guesswork. You can’t easily forecast a bill that depends on how many questions twelve people ask next month.
- People start rationing themselves. Once staff know there’s a meter running, they hesitate before asking. The tool you bought to save time ends up underused.
- Limits arrive at the worst moment. Hitting a cap halfway through a busy quoting week means waiting, upgrading in a rush, or buying credits at whatever the going rate is.
None of this makes cloud AI bad. But the “unlimited for a flat fee” era is ending, and it’s worth planning for that now rather than finding out from an invoice.
The alternative: own your AI
There’s another option that many small businesses haven’t considered: run the AI on your own premises, on hardware you own.
That’s what we build at Inferred. Cortex is a box that sits in your office and answers questions from your own files: drawings, specs, quotes and job history. In cost terms, that means:
- No meter. Ask ten questions a day or ten thousand. The cost is the same.
- A predictable price. You know what it costs up front. Support is optional, through Cortex Care, at a fixed monthly fee rather than a usage bill.
- No surprise limits. There’s no allowance to run out of on a Friday afternoon.
And because it runs inside your building, you also get the reason Cortex exists: your data never leaves the premises. It even works when the internet is down.
Is it right for everyone?
No, and we’d rather tell you that up front.
If you use AI now and then to draft the odd email, a cloud subscription is probably fine. The biggest cloud models are also still ahead on some general-purpose tasks.
Owning your AI starts to make sense when:
- several people in your team would use it every day
- the value is in your documents, not general knowledge
- those documents are commercially sensitive: pricing, client details, designs
- you want a cost you can put in a budget and forget about
We have worked the comparison out over three years, including where renting wins.
Five questions to ask before your next AI renewal
- How is our plan’s usage measured, and what happens when we hit the limit?
- What did we actually spend on AI last quarter, across every tool and every login?
- Which tasks would our team use AI for if cost weren’t a worry?
- Which of our files are we comfortable sending to a third party’s servers, and which aren’t we?
- If the price doubled next year, would we still be getting value?
If those answers make you uneasy, it’s worth a conversation.
Talk to us
Inferred builds AI that never leaves the building. If you’d like to see Cortex answer questions from your own files, get in touch for a short call. We’ll show you exactly what it would cost, with no meter running. Our prices are published on our pricing page.
Sources and scope
- Your Weekly AI Brief, week of 9 November 2026: the OpenAI plan change, citing an OpenAI help centre article first seen on 31 October. OpenAI has not published the allowance sizes.
- Ramp AI Index: token volume up 37.1% and token spending up 8.9% for the week ending 23 October 2026, against the same week a month earlier. Ramp’s figures come from businesses using its platform, mostly in the US. Also reported by Sam on AI, 9 November 2026.
- MacRumors, 4 November 2026, reporting Bloomberg: Apple’s cut to some engineers’ daily AI spending caps, from $300 to $200, with further cuts under consideration.
Plan details change often. Check your own provider’s current terms before you renew.