The world of AI can look like a playground for only the big fish. It’s easy to feel that only businesses with deep pockets can afford it. If you’ve peeked into the world of AI in the past, you may have wondered why artificial intelligence came with such with a hefty price tag.
But here’s the good news: the price of AI is falling. 19 months ago, running GPT-4 came with a price tag of $36 per million tokens. That cost dropped to $4 per million tokens and below. Even better, thanks to the growing number of open-source options, efficiency breakthroughs, and increased competition, the costs are dropping across the board. With the UK government also embracing AI growth, it might be time for you to start considering how this technology can help you.
Tokens represent pieces of words, usually between 1 to 4 characters. The pricing is split between input tokens and output tokens. ‘Input tokens’ are the tokens you give to the model and ‘output tokens’ are the tokens the model produces in response.
As businesses grow, so does the likelihood of them adopting AI. According to the government in 2022, 68% of large businesses, 33% of medium businesses and 15% of small businesses are using at least one AI technology.
Today, small and medium-sized businesses can access AI solutions at a fraction of the cost, with free applications, open-source platforms, pre-trained models and usage-based price models lowering the entry barriers. So, buckle up as we open up AI’s pricing secrets and see what it can cost to make your digital dreams come true!
Where is the Cost?
AI development is grounded in a lot of choices and expensive resources.
What Type do You Want to Build
Artificial intelligence is the tree that branches into any technology that simulates human intelligence. AI voice assistants & chatbots use natural language to answer questions, sophisticated healthcare systems used to spot cancer in samples. All of these can be described as artificial intelligence. Each type of AI, however, comes with different levels of complexity, performance needs, and costs. Simply put, training and operating AI models that require lots of images costs more than one that only uses and outputs text.
Bottom of the Complexity scale
These are entry-level AI systems such as automatic data entry tools or basic chatbot applications. They typically rely on pre-built models or third-party platforms and don’t require much customisation. Mailchimp’s AI-Powered Marketing is a good example—offering basic automation and insights without deep learning capabilities. These solutions are often quick to implement and can cost between £5,000 and £50,000. Allowing for some integrations or branding at the top end of the scale.
Middle of the Complexity scale
This level includes more capable systems like customer service chatbots with natural language understanding, predictive analytics, or recommendation engines. They typically use machine learning trained on your own data to deliver personalised insights or responses. Costs here vary depending on the depth of functionality but generally range from £30,000 to £50,000, depending on the complexity of your goals.
Top of the Complexity scale
The most complex use cases of AI are where AI systems are performing very specialised work, such as image classification of unusual objects, or combining multiple AI tasks (text, voice, vision). These usually involve significant bespoke development, and training models on your data so that they can do these specialised tasks. A nice example of this level of complexity is in Spotify’s recommendation system, voice translation for podcasts, AI DJ feature, the natural language search function and Spotify Wrapped. The cost of these solutions varies depending on the use case but can often cost £100,000 to £1,000,000 or more.
How Big do You Want to Build it?
An important question in construction, and an important question for AI development. The effectiveness of AI hinges on the quality and quantity of data it’s trained on. Algorithms fed with ample, structured data generally perform better and cost less to implement. However, if your AI relies on unstructured data like emails or images, it may require extensive preprocessing and tagging to be usable, increasing both time and cost.
Cloud vs Hardware
Running AI models, particularly large ones, require some beefy hardware – current-generation servers with multiple GPUs. These can cost between £10,000 and £50,000, a bit of an eye-watering figure! Fortunately, many AI applications get by with cloud-based services or existing hardware, so you may not need to open up your wallet too wide here.
Cost of GPUs – Depending on the size of your model, for a small model the costs range from £500 – £1000. For a larger model they can range from £8000 – £15000.
Cloud computing – Renting cloud computing resources from Amazon Web Services (AWS), Microsoft Azure, or Google Cloud Platform (GCP) is alternative to buying hardware. For AWS, GPU instances range from £2 – £19 per hour.
Data management
An AI model is only as good as the data it’s trained on. And believe us, when it comes to data, you get what you pay for! Data collection, cleaning, and processing are essential for AI accuracy, with costs ranging from £5,000 to £30,000. Think of it as the “5-a-day” for AI—without quality data, your model’s performance will suffer. While the exact amount you’ll spend on data management depends on your needs, this expense is hard to avoid.
Integration Costs
If you’re building an AI that needs to work within existing software or systems, you may need additional funds for integration. Integration costs can range from £5,000 to £20,000 and may involve custom software development or middleware to ensure compatibility. For standalone applications, however, integration costs can often be minimal or entirely unnecessary, giving you the flexibility to skip this expense if it’s not relevant to your project.
These are the figures we’ve found from a lot of different providers, and as you can see development costs vary considerably. As we’ve discussed there’s a lot of factors affecting AI development cost. Unfortunately, all of these factors and the propensity for providers to overhype the benefits of AI development, while not considering their clients individual circumstances. Old legacy systems, shortage of technical talent and the potential risk to their reputation. This has led a lot of development delays. As a result, there’s been a bit of an exodus with companies abandoning their generative AI pilot projects. A 42% rise in abandoned projects compared to last year’s 17%.
Ongoing Maintenance
The cost of AI doesn’t end with deployment. To keep an AI system performing at its best, regular updates and maintenance are essential. Annual maintenance fees generally add up to 15–20% of the initial development cost, but this can vary depending on the complexity of your AI and update frequency.
Now that we’ve discussed the numerous price factors. Now you want to know, “what can AI do for me and where do I start?”. Depending on your confidence level and ability we’ll cover both open source and commercial solutions.
The Power of Open Source
If you’re dipping your toes into AI and simply exploring your options, open-source AI offers a powerful alternative to commercial options like Chat-GPT. Open-source AI is a bit of a communal treasure trove: it pulls together code, models, and datasets from developers around the world, all of which are freely shared under open-source licences. This means you can download, tweak, and build to fit your exact needs—all without spending a fortune.
Stable Diffusion
An image generation model that creates highly realistic images based on text or images. Unlike commercial tools it doesn’t have a web interface so you’ll have to either use third-party tools or run it locally on your own resources.
Meta Llama
Meta has developed a number of generative AI models available under open-source licenses – the Llama family of models. Meta provides a range of model sizes, from models you can run on single GPUs to large models which are competitive with state-of-the-art commercial models.
Bloom
Described as the biggest multi-lingual language model (176 billion parameters) as part of a global project. Created by Hugging Face, a great place to find open-source AI resources, Bloom is available for anyone to use if it isn’t used for harmful purposes. As described in the terms of the projects Responsible AI License.
Open-source Libraries/Frameworks
TensorFlow, PyTorch, BotKit, Microsoft Bot Framework and many more. Obviously, this is just a small compilation of the resources available.
Proprietary vs Open-source AI tools
We’ve gone over just a few of the options available for open-source AI tools now let’s compare open-source and proprietary pre-built AI tools.
Commercial and open-source AI tools can both do the same things but differ in important ways:
- First customisability, which goes to open-source as it is generally much easier to customise.
- Second is ease of use, this goes to the commercial AI tools because you won’t absolutely need AI proficient engineers to customise the software for your purpose.
- Third is the hardware requirements, for proprietary AI platforms there’s no infrastructure requirements which usually can’t be said for open-source. Requirements can include massive storage and specialised servers.
- Lastly is cost. It is impossible to say for certain which is more expensive. When it comes to commercial solutions, pricing models can vary significantly. For instance, OpenAI’s pricing is tiered mostly capping messages a user can send. The cost of open-source AI will likely have a relatively low set up cost with no licensing fees. However, the cost is dependent upon the cost of the decisions we’ve discussed above.
These are the figures we’ve found from a lot of different providers, and as you can see development costs vary considerably. As we’ve discussed there’s a lot of factors affecting AI development cost. Unfortunately, that’s led to a lot of providers overhyping the benefits of AI development.
There’s been a bit of an exodus with companies abandoning their generative – AI pilot projects. A 42% rise in abandoned projects compared to last years 17%.
Make the Most of Free AI
Free AI applications and open-source software provide a great start for those interested in AI without spending a penny. They’re perfect for trying out initial concepts and getting to know how AI systems work. By experimenting with how to use AI through these free options, you can determine if you’re ready to invest in something more powerful and custom purposed.
Design
- Canva – AI design suggestions/ background remover/ AI image generator
- DALL-E 2 – Art generation
Writing
- Grammarly – free plan now includes AI features
- Wordtune – helps rewrite your sentences
- Otter.ai – A speech-to-text transcription application
Education
- Jasper – Research assistant
- Quillbot – paraphrasing tool
- Quizlet – creates personal study plans
So, Is AI Worth the Cost?
While the upfront costs of AI can be steep, there are ways to mitigate costs and good reasons to make the investment. But it’s important to consider if your organisation is ready. In sectors like customer service, healthcare, and logistics, AI cuts down on admin time and automates time intensive tasks, reducing pressure on your human staff. For businesses eyeing AI as part of their growth strategy, a bit of due diligence on pricing can help with planning.
