Amazon – No chops

Investment underlines Amazon’s AI weaknesses.

• Amazon is staking its claim in the world of generative AI but at the same time, the size of the bet it is making on Anthropic clearly signals that Amazon’s existing AI is second-rate and that something else is needed to allow it to play in the big leagues.
• Amazon is investing $1.25bn in Anthropic which will give it a minority stake in the company with the option to upsize that to $4bn at some point in the future.
• The terms of the deal are obscure, but it appears that Anthropic had the upper hand in its negotiations and that even after the next $3.75bn, Amazon will still not have control over the company.
• Anthropic also appears to be somewhat schizophrenic in its messaging as Amazon will now be Anthropic’s primary cloud supplier despite the existing relationship with Google.
• Anthropic has said that nothing has changed in its relationship with Google although when it raised money from Google it said that Google was the preferred cloud supplier.
• From Amazon’s perspective, this deal has two aims.
  • First, AI chops where I have been of the opinion for many years that Amazon is second or even third rate when it comes to AI.
  • One only has to look at how little improvement has been made in Alexa’s language ability over the last 6 years for evidence of just how weak it is.
  • Advertisements for products already purchased and AI wrongfully shutting down merchants for counterfeit goods are further evidence of how much Amazon has struggled.
  • With Anthropic’s AI under the hood, things might get somewhat better but Anthropic will still not be exclusive to Amazon leading me to think that even more money will be needed to secure Anthropic exclusively.
  • Second, orphan chips. Amazon has designed some chips for AI training that no one seems to want to use.
  • Trainium and Inferentia are pretty self-explanatory but critically they don’t run Cuda which is still what everyone seems to want to use.
  • Furthermore, Nvidia has a 10-year lead in developing tools to make its silicon easy to use for AI training and, with 85%+ market share, it is the industry standard.
  • By throwing a vast amount of money at Anthropic, Amazon has incentivised Anthropic to adapt its systems so that its generative AI can be trained on Trainium and executed on Inferentia.
  • This will be a cornerstone developer that Amazon hopes will attract others meaning that it won’t have to give a large portion of cloud profits over to Nvidia when it is forced to buy chips at 70% gross margins.
• This investment comes at a very interesting time as only last week Amazon announced the upgrade of Alexa to a generative AI model of some description.
• I had assumed that Amazon had something in-house but the demos were not great and so presumably, Alexa will now be powered by Anthropic once it has been implemented.
• This is a good deal for Anrthopic as Amazon has 1bn installed devices (excluding those consigned to draws) meaning that the reach of its AI services could now rival Google’s.
• I expect that this is just the beginning of a period of consolidation that will last beyond the current frenzy and excitement.
• This is because I suspect that the business model that all of these generative AI start-ups have raised money against will not stand up to the ravages of competition.
• Hence, the subscription price being forecast is probably much too high meaning that estimates and expectations are going to be missed.
• This will turn the tables on generative AI just as it did in autonomous driving meaning that capital will become much more scarce leading to down rounds and consolidation.
• This is the pin that I think will prick the generative AI bubble which will allow the survivors to get on with creating what is likely to be the real and very profitable products powered by generative AI.
• I continue to think that Amazon offers very little value to investors and that there are far more interesting opportunities to be had elsewhere.