AI Ecosystem – Yet Another Contender

Amazon makes a play for the AI ecosystem.

• Amazon has launched a new family of foundation models that I suspect are based on Anthropic’s technology that provides a starting point for those who want to create generative AI services in yet another play for the emerging AI ecosystem.
• At its annual developer conference, re:Invent, Amazon Web Services (AWS) launched Amazon Nova Micro (text-only), Amazon Nova Lite (low-cost image, video and text) and Amazon Nova Pro (greater capability and accuracy) with its largest and most capable model, Amazon Nova Premier, still in training.
• These models are exclusively available on Amazon Bedrock which is Amazon’s managed service that aims to make it easy for developers to create generative AI services without having to start from scratch.
• Bedrock is analogous to Nvidia’s Nemo, Microsoft’s AI Studio, Google’s AI Studio and so on and represents yet another sign that the AI industry is moving towards what RFM Research has defined as AI Ecosystem 2.0 (see here).
• AI Ecosystem 1.0 is still prevalent today and is where most developers bring their own or open-sourced foundation models like Meta’s Llama and create their services using Nvidia’s CUDA platform and the tools that come with it.
• CUDA is the de-facto industry standard platform for AI training, but as foundation models become more commoditised and the developer pool becomes less specialised, the focus is likely to move to developing on the foundation models themselves.
• The offerings are all quite different but essentially, they all boil down to a set of tools with which a developer can create a generative AI service without having to know anything about creating a foundation model or training on silicon.
• This makes generative AI much more accessible and while Nvidia already has 5m developers that know how to use its CUDA silicon development platform, many more developers want to create generative AI services.
• These developers are likely to migrate straight to these platforms meaning that the focus of development moves from silicon development to the foundation model.
• This is very similar to what happened in smartphones and RFM Research has concluded that this is where the battle for the AI ecosystem will be fought.
• Amazon’s models are the latest entry to what is becoming a long list of contenders and while Amazon is pretty late to this game, it does have a key advantage.
• This is the installed base of customers where AWS is the largest cloud provider in the world by some considerable margin.
• This gives it the option to offer generative AI creation services to existing customers at very attractive prices because, with its own silicon and infrastructure, it can eat some of the cost to compete on price.
• This is a classic Amazon tactic and I suspect that some of the commentary around how cost-effective these models are to run is in part Amazon offering low prices to gain market share.
• The big question mark is how good these models are and here I suspect that Amazon is taking no risks and has based its models on Anthropic technology.
• Amazon has not been very good at AI for years and I have long viewed its investment in Anthropic as a way to fix this problem.
• Hence, with a launch that represents Amazon’s offering to become a major player in the AI ecosystem, I don’t think that it can afford to take any risks.
• This combined with its checkered history in this area leads me to think that Nova is based on Claude from Anthropic making it even more likely that Amazon will end up acquiring the company.
• With $8bn+ now invested, I suspect that Amazon has effective control ensuring that Claude, Nova and all of Amazon’s other offerings will be optimised to run on its in-house silicon.
• This is how Amazon can reduce its dependence on Nvidia, but it will require its generative AI toolkit to become very popular.
• When it comes to engagement with developers, Nvidia still has the lead as it also has an offering in this space and has moved early to drive engagement with enterprises, but the battle is far from over.
• In the long run, I suspect that the AI ecosystem will boil down to 3-5 large foundation models upon which almost all AI is based but those 3-5 slots are going to be heavily contested.
• This is good for developers and the AI industry at large as fierce competition will mean that only the best products survive and Nvidia is on track to be one of those although it looks like it will not have the market to itself as it does now.
• However, this is going to take some time to evolve and so its commanding position looks safe for at least the current investment horizon of 12 to 18 months.