This is a brilliant articulation of the structural diffrences between cloud incumbents and pure-play AI firms. Your point about Amazon's retail DNA introducing it to low-margin, high-volumen operations really resonates as the critical differentiator. The real insight isn't just historical though: it illuminates why we're seeing such divergent investment patterns today, where hyperscalers can afford to lose on models while stil winning on infrastruckture, whereas OpenAI/Anthropic cannot. This framework explains the entire stack dynamics we're observing.
I would make the argument that datacenters are already commoditized (compute is pretty fungible from hyperscalers or neoclouds), and that the value comes from the platforms that hyperscalers operate. And if you take this view, both killer app and quietly embedding AI approaches lead back to reinforcing existing hyperscaler platforms.
I agree that for companies like OAI and Anthropic, developing a flagship app/product is existential.
But I think this is a bit too narrowly focused of a perspective for the hyperscalers? As Satya describes in his recent dwarkesh interview, they intend to seriously compete on models (one can debate how seriously this should be taken) but also on applications (e.g. github copilot), while acknowledging they can still win just on cloud services alone.
But simply because they could win on cloud while losing on models, I don't think makes hyperscalers any less serious competitors. For instance, Google is leading on models (Gemini), cloud (GCP), chips (TPU), applications (Waymo and gemini integrations) so they are rigorously competing on every part of the stack, even though they have a strong lead on cloud compute.
I generally view large tech bureaucracies much more idiosyncratically rather than determined through laws of markets e.g. Apple's failed Siri, Google's TPU success, Microsoft's early OAI investment, Meta's failure in Llama, are all the result of certain senior leaders in these companies.
The ironic thing to me is that despite all the Biden DOJ antitrust investigation into Nvidia and OpenAI, there is actually vigorous competition at every layer of the AI tech stack between startups and incumbents.
This is a brilliant articulation of the structural diffrences between cloud incumbents and pure-play AI firms. Your point about Amazon's retail DNA introducing it to low-margin, high-volumen operations really resonates as the critical differentiator. The real insight isn't just historical though: it illuminates why we're seeing such divergent investment patterns today, where hyperscalers can afford to lose on models while stil winning on infrastruckture, whereas OpenAI/Anthropic cannot. This framework explains the entire stack dynamics we're observing.
I would make the argument that datacenters are already commoditized (compute is pretty fungible from hyperscalers or neoclouds), and that the value comes from the platforms that hyperscalers operate. And if you take this view, both killer app and quietly embedding AI approaches lead back to reinforcing existing hyperscaler platforms.
I agree that for companies like OAI and Anthropic, developing a flagship app/product is existential.
But I think this is a bit too narrowly focused of a perspective for the hyperscalers? As Satya describes in his recent dwarkesh interview, they intend to seriously compete on models (one can debate how seriously this should be taken) but also on applications (e.g. github copilot), while acknowledging they can still win just on cloud services alone.
https://www.dwarkesh.com/p/satya-nadella-2
But simply because they could win on cloud while losing on models, I don't think makes hyperscalers any less serious competitors. For instance, Google is leading on models (Gemini), cloud (GCP), chips (TPU), applications (Waymo and gemini integrations) so they are rigorously competing on every part of the stack, even though they have a strong lead on cloud compute.
I generally view large tech bureaucracies much more idiosyncratically rather than determined through laws of markets e.g. Apple's failed Siri, Google's TPU success, Microsoft's early OAI investment, Meta's failure in Llama, are all the result of certain senior leaders in these companies.
The ironic thing to me is that despite all the Biden DOJ antitrust investigation into Nvidia and OpenAI, there is actually vigorous competition at every layer of the AI tech stack between startups and incumbents.