The competitive phase of a new technology could be short-livedWhat percentage of the booming market for generative AI will be absorbed by the largest tech companies? What will be left for all the other tech companies hoping to capitalize on the latest craze in the world of technology?
ChatGPT has only been around for less than five months, but these questions are already prominent as the largest tech companies compete to claim large portions.
Amazon is the latest company to announce its plans for generative AI. This was done through its Amazon Web Services cloud-computing arm. AWS announced last week that it would make available on its platform several large AI models including Anthropic’s large language model and Stable Diffusion, an open-source image-generating system.
AWS is trying to position its cloud as the leader in the new market for generative AI by hosting and delivering these independent AI services. AWS provides all the tools needed by developers to create, train, and deploy their own AI models. It also designs its own chips for training and running large machines-learning systems.
It’s not the only one. In the past month, Google has boasted about supercomputers that are built using its latest generation of chips called TPUs. These have reached breakthrough performance levels in training large AI model. Microsoft, too, has joined the rush of the largest tech companies in developing its own AI-specific chips. (The Information first reported on Microsoft’s plans.)
These moves show how far big tech companies have come in their efforts to control the entire “stack” of AI computing, that is the technology required to run and support the new computing workloads that are being demanded and to turn them into services that customers can use.
The chips at the bottom of the stack process huge amounts of data to train AI models. The other layers are the software and algorithms required to deploy and train the systems, as well as the large-scale vision and language models, which act as the base level of intelligence.
Amazon, Microsoft, and Google have already staked their claim on most of the lower level of this hierarchy of technologies, making it difficult for others to break in to a market that will require operating at a large scale with the lowest costs.
Elon Musk faces a steep uphill climb, even though he claims that his AI startup will be the “third force” against Google and Microsoft/OpenAI. Tesla, Musk’s electric car company has already developed an AI computer that can handle vision recognition. The irrepressible Musk said this week that selling the technology to others would be worth “hundreds and billions” one day. It will be difficult to catch up with tech giants who have spent years perfecting their technology.
It’s now a question of how far up the “stack”, the cloud companies will try to go, in order to claim more value from this new technology.
Achieving control over their large AI models, or, as in Microsoft’s example, a close partnership with OpenAI, seems to be a realistic goal for those who do not already possess it. The Foundation models are expensive to create and can be used for a variety of applications. This makes them an ideal first step for big tech companies with AI ambitions.
These large models are important to the strategic goals of companies and therefore they do not see them as profit centers in their own rights. Emad Mostaque is the head of Stability AI (the company behind Stable Diffusion) and he sees things that way. He warns that there will be a race to the bottom in pricing, as big tech companies fight to establish their AI systems and leave little room for others.
Mostaque instead is counting on two factors. Amazon will never try to replace rival AI models with its own. Instead, it will make money by hosting them in its cloud. Second, there will be plenty of room for differentiation among AI models and not all customers want to rely solely on opaque, giant systems operated by a few dominant tech companies. If he’s wrong, the early competitive phase of generative AI could be very short-lived.