For customers of AI instruments, the dialogue usually focuses on the latest model of ChatGPT or Gemini or DeepResearch. However the level of contact of the AI business with customers is after all only one a part of the AI business as an entire. as an entire. Leonardo Gambacorta and Vatsala Shreeti lay it out in “The AI supply chain” (Financial institution of Worldwide Settlements Papers No. 154, March 2025). Right here is their model of the AI provide chain:
Primarily based on this construction, listed below are a number of the key participant within the general AI ecosystem:
In fact, these graphs are an summary fairly than an in depth presentation, however that’s usually a helpful approach of gaining perspective. Gambacorta and Shreeti present extra element within the textual content. For instance (citations and references to graphs omitted):
{Hardware}. Take into account a very powerful {hardware} for AI functions, particularly microprocessors like GPUs. Nvidia – headquartered in Santa Clara, California, in the US – serves a lot of the marketplace for GPUs, with its market share reported to be bigger than 90%. It has gross margins of over 70% and has seen its revenues improve by 405% between 2023 and 2024. Initially serving the online game market, Nvidia had a head begin in leveraging the parallel computing capability of its GPUs for AI fashions. Over time, it has built-up substantial mental property and a major popularity, solidifying its place because the market chief for GPUs. Aside from the GPUs themselves, Nvidia additionally produces complementary software program. Nvidia’s GPUs are available an unique bundle with CUDA, its parallel computing platform, which allows programmers and software program builders to simplify the method of utilizing GPUs and to reinforce their efficiency. CUDA has turn out to be the business commonplace for programmers and may solely be used with Nvidia’s GPUs. …
To make sure, a number of different corporations, together with startups and large techs, are additionally lively out there for AI {hardware}. Superior Micro Units (AMD), Intel and large techs like Microsoft, Google and Amazon are all producing AI microprocessors to compete with Nvidia’s GPUs, each for coaching AI fashions and for inference. Chinese language firms like Alibaba, Baidu and Huawei are additionally beginning to produce their very own microprocessors, particularly in mild of geopolitical constraints. …
Cloud computing layer. Globally, the cloud computing market is dominated by three massive tech firms: Amazon Net Providers (AWS) with a market share of 31%, Microsoft Azure with 24% and Google Cloud Platform with 11%. Within the European Union (EU), the estimated mixed market share of AWS and Azure in 2020 was over 80% and their revenue margins have been additionally reported to be excessive, at 30% and 38%, respectively. Within the case of the IaaS section – essentially the most related one for AI fashions the market is much more concentrated. In 2023, AWS, Microsoft Azure and Google Cloud Platform collectively accounted for almost 74% of the worldwide market. …
Coaching information. Thus far, frontier AI fashions have been skilled utilizing huge troves of publicly accessible information. Nonetheless, because the inventory of public information quickly declines, corporations are turning to different information sources. …
Basis fashions. At first look, the marketplace for basis fashions is dynamic and rife with opponents. There are over 300 basis fashions out there, offered by 14 totally different corporations. There are additionally competing enterprise fashions – whereas some corporations select to supply proprietary basis fashions (like OpenAI and Google DeepMind), others have adopted a comparatively extra open strategy (notably Meta with its open supply Llama fashions and, extra not too long ago, DeepSeek). … However, the marketplace for basis fashions is presently dominated by solely a handful of corporations like OpenAI, Google DeepMind, Anthropic and Meta. In 2023, regardless of quite a few competing basis fashions, OpenAI’s GPT-4 accounted for 69% of the marketplace for generative AI when it comes to world income. Given the dynamic nature of the market and the potential to understand efficiencies, the hierarchy might shift quickly. …
AI functions and user-facing layer. The final stage of the AI provide chain, the consumer dealing with layer, follows the playbook of digital platforms and cell functions. Because the “ChatGPT moment” of AI, functions constructed on high of basis fashions have been proliferating in varied sectors of the financial system together with well being, training, backend processing and compliance, software program growth and others. Nonetheless, and as with digital platforms, there could be a threat of “winner takes all” dynamics rising within the markets for AI functions. Whereas it’s a Herculean job to hint the marketplace for AI functions in each sector, the marketplace for chatbots could be instructive. … [D]espite a flurry of comparable interfaces, ChatGPT nonetheless accounted for 60% of the chatbot market (measured by the full variety of month-to-month visits) in 2024, highlighting the significance of being first to market.
One lesson from this overview is that if one believes that AI instruments are going to be essential drivers of productiveness going ahead, then a number one financial system like the US ought to be desirous about strengthening the whole ecosystem for growing and utilizing AI. A tougher lesson is that with a really quickly evolving know-how, the perfect approaches for corporations, customers, and policymakers are by no means apparent upfront. Companies that appear to have a dominant market place in early 2025 might not have one in six months or a 12 months or two, whether or not policymakers take any motion or not. Classes are discovered a few know-how as corporations and methods rise and fall, and to a considerable extent, policymakers ought to let the method play out so tha these classes could be discovered. Particularly, policymakers generally tend to take heed to the loudest complainers, fairly than giving equal weight to the beneficiaries who’re too busy constructing and utilizing these instruments to spend time lobbying about them.
