For those of us bottom feeding on cheap(er) AI services, we got a shock this week. DeepSeek increase the pricing of their model APIs in some cases by 14x. This means, of course, bills of $5 a month jump to $25, and $25 to $125, with no real change in functionality, just a straight increase. And, this is on the back of the peak time pricing, introduced a few weeks ago too.
Now I am not begrudging DeepSeek an increase their prices. The price was already low and by all accounts they do seem to have been struggling with capacity (and also looking for funding, so more revenue no doubt helps).
Even with the price increase they are still amongst the cheapest in the market.
But, this is happening elsewhere too. X.AI, in May this year, pushed users away from their extremely well priced “Fast” models to Grok 4.3, a 5x increase. It all points to a wider trend around AI pricing… it is going up.
The API economy
In our daily lives, most of us are using AI models now. If nothing more it is a much more efficient way to search for information (is it better or more accurate, that we can argue about another time). However if you are mainly using the browser or app based chat interfaces, with bundled pricing, you may be unaware of the burgeoning economy and infrastructure that is being built underneath. I know I was.
Beneath the surface of AI services is typically a myriad of APIs and API calls. These APIs are simply agreed data exchange formats that service can use to exchange information, to call out to other programmes and services. Some you pay for, some are free, but they are intertwined in modern AI infrastructure.
LLM Model, Web-search, Image creation, Email, Calendar, Weather, Daily Joke of the day… all can be linked to APIs… even the FCA is getting in on the act with the release of an API to access their Handbook.
Each of these have price points, data security considerations and with this complexity comes a requirement for some sort of active management.
- Do I need to be using the most expensive and capable model for simply moving a document?
- Do I use a particular image model to create a visual summary?
- Which gives me more accurate and current, up to date, information when searching?
- What data am I exchanging and where is it going?
Of course, it may be worth paying the price for a premium model, that does all this for you. I feel sure many will, and large providers are adding functionality at pace (eg GrokBot, by SpaceX launched this week… a bit like OpenClaw (now owned by OpenAI)).
However for those of us without oodles of cash to burn (ie many businesses and certainly me!) we will be left managing this complexity. Managing it well can have a dramatic difference to the cost profile (I was able to reduce cost by 80% with some optimisation).
Yet if monthly costs are now going to routinely reach £40 a month (or even £100+… I seem to be spending even more when you add in all AI services) per user, it does feel like the sun is setting on the era of cheap AI… winter is coming and we need to be prepared!
Going Local
So it is at this post we should ask the question… when do we move away from server based models to highly capable local, open source, models on our own, local, server.
These models are competitive, advanced and this is already possible (Ollama is super easy, and easiest, to try if you have not). However I have found without very a high end PC, memory and graphics card, running the most capable models are still slow in comparison to online versions.(ish)… and for higher end services such as web-search, image and video creation, in some cases not possible on my existing (dedicated) hardware.
This is really only a question of processing power, typical consumer and business PCs are just not fast enough. To build or buy, one that is currently costs around £5,000. Pricey, yes… but vs a steeping monthly AI bill… it is getting there for the small user.
I wonder, with the likely price escalation, as firms recover investment costs, if this is a trend will see as people react and adjust.
Nothing new?
I suppose in some ways this is nothing new.
Already in business we use different employees with different skill-sets (and price points) for different tasks. We match the resource to the task to optimise the cost profile, we give them the tools to communicate and measure performance for jobs done well… and we also look at building in-house, locally, vs using third party managed services (or not).
And, this returns to a recurring theme.
Despite all the excitement around new tech and AI, in some ways not much has changed. Much of the toolset we need to manage it we already have…
AI is not making good management, structure and process go away, in many ways it is making it even more important.
Something to think about back in the office this week. Have a good week, everyone.
[A quick event plug. We will be talking about many of these issues, including governance, standards and risks at our live events in Oct/Nov… more info here, come along if you are free -> Event page]

