High AI training costs have been a significant barrier to AI adoption, preventing many companies from implementing AI technology. According to a 2017 Forrester Consulting report, 48% of companies cited high technology costs as a top reason for not implementing AI-driven solutions.
However, recent developments have shown that the cost of AI training is falling rapidly, and this trend is expected to continue in the future. According to the ARK Invest Big Ideas 2023 report, the cost of training a large language model similar to GPT-3 level performance has decreased from $4.6 million in 2020 to $450,000 in 2022, a decrease of 70% % per year.
Let’s further examine this trend of decreasing AI training costs and discuss the factors contributing to this decrease.
How have AI training costs changed over time?
According to the latest ARK Invest 2020 study, the cost of training deep learning models improves 50x faster than Moore’s Law. In fact, the costs associated with running an AI inference system have drastically reduced to almost negligible levels for numerous use cases.
In addition, training costs have fallen tenfold annually in recent years. For example, training an image classifier like ResNet-50 on a public cloud cost about $1,000 in 2017, but by 2019 the cost had dropped significantly to about $10.
These results align with a 2020 report by OpenAI, which found that the amount of computing power required to train an AI model to perform the same task has increased by a factor of 16 every 16 months since 2012 lost two.
Additionally, the ARK report highlights falling AI training costs. The report predicts that training costs for a GPT-3 level model will drop to $30 by 2030, compared to $450,000 in 2022.
Cost of Training Performance to GPT-3 Level – ARK Invest Big Ideas 2023
Factors contributing to decreasing AI training costs
Training AI models is becoming cheaper and easier as AI technologies continue to improve, making them more accessible to a wider range of organizations. Several factors, including hardware and software costs and cloud-based AI, have contributed to falling AI training costs.
Let’s examine these factors below.
1.Hardware
AI requires specialized, expensive, high-end hardware to process large amounts of data and calculations. Organizations like NVIDIA, IBM, and Google provide GPUs and TPUs to run HPC (high performance computing) workloads. High hardware costs make it difficult to democratize AI on a large scale.
However, as technology advances, hardware costs decrease. According to the ARK Invest 2023 report, Wright’s law predicts that the cost of producing AI relative compute units (RCU), i.e. the cost of AI training hardware, should fall by 57% annually, resulting in a 70% reduction in AI training costs % by 2030. as shown in the graphic below.
AI Training Hardware Cost – ARK Invest Big Ideas 2023
2.Software
AI software training costs can be reduced by 47% annually through increased efficiency and scalability. Software frameworks like TensorFlow and PyTorch enable developers to train complex deep learning models on distributed systems with high performance while saving time and resources.
In addition, large pre-trained models such as Inceptionv3 or ResNet and transfer learning techniques also help reduce costs by allowing developers to optimize existing models instead of training them from scratch.
AI Software Training Costs – ARK Invest Big Ideas 2023
3. Cloud-based artificial intelligence
Cloud-based AI training lowers costs by providing scalable computing resources on-demand. With the pay-as-you-go model, companies only pay for their computing resources. Also, cloud providers offer ready-made AI services that speed up AI training.
For example, Azure Machine Learning is a cloud-based predictive analytics service that enables rapid model development and deployment. It offers flexible computing resources and memory. Users can quickly scale to thousands of GPUs to increase their computing power. It allows users to work in preconfigured AI environments through their web browsers, eliminating the hassle of setup and installation.
The impact of falling AI training costs
The falling cost of AI training is having a significant impact on various industries and domains, leading to improved innovation and competitiveness.
Let’s discuss some of them below.
1. Mass Adoption of Sophisticated AI Chatbots
AI chatbots are on the rise due to falling AI costs. Especially after the development of OpenAI’s ChatGPT and GPT-4 (Generative Pre-Trained Transformer), the number of companies looking to develop AI chatbots with similar or better capabilities has increased significantly.
For example, ChatGPT amassed 1 million users five days after its release in November 2022. Although the cost of running the model at scale today is about $0.01 per query, Wright’s Law predicts that chatbot applications similar to ChatGPT will be much cheaper to deploy at scale by 2030 (an estimated $650 for the execution of one billion queries). with the potential to process 8.5 billion searches per day, which is equivalent to Google search.
Cost of running AI inference per billion queries – ARK Invest Big Ideas 2023
2. Increased use of generative AI
The falling cost of AI training has led to an increase in the development and implementation of generative AI technologies. In 2022, there has been a significant increase in Generative AI usage, driven by the introduction of innovative Generative AI tools such as DALL-E 2, Meta Make-A-Video, and Stable Diffusion. In 2023 we have already seen a groundbreaking model with GPT-4.
In addition to image and text generation, generative AI helps developers write code. Programs like GitHub Copilot can help complete a coding task in half the time.
Time to Get Programming Tasks Done – ARK Invest Big Ideas 2023
3. Better use of training data
Reduced AI training costs are expected to enable better use of training data for machine learning. For example, the ARK Invest 2023 report suggests that the cost of training a model with 57 times more parameters and 720 times more tokens than GPT-3 (175 billion parameters) is expected to reach $17 billion by 2030 will drop to $600,000.
Data availability and quality will be the main constraining factor for developing advanced machine learning models in this low-cost computing world. However, training models would develop the capacity to process an estimated 162 trillion words or 216 trillion tokens.
The future of AI looks very promising. To learn more about the latest trends and research in the field of artificial intelligence, visit Unite.ai.
