Artificial Intelligence – Hot Summer

Valuations are still going up.

• Cohere, one of the current group of white-hot AI startups has raised money at $5.5bn confirming that the market’s appetite for overvalued paper remains unabated ensuring that when the correction comes, it is going to hurt some players.
• Cohere is a little bit different than many of its peers such as Anthropic or Mistral in that it focuses its models on the enterprise use case which is arguably the biggest use case that RFM has identified.
• This is because generative AI has the ability to ingest, categorise and cross-reference unstructured data like emails, transcripts, videos and images and then retrieve that data in an easy-to-command manner.
• This is a large step forward in a corporation’s ability to make use of the data that exists on its servers where historically it has never really been able to make any use of it.
• Furthermore, by addressing the enterprise, the use cases for Cohere’s models are more narrow meaning that the models can be smaller meaning that they are cheaper to build.
• However, they also have to be more specific to the task at hand, meaning that Cohere will need more models than Anthropic or OpenAI to address its customer base.
• Unsurprisingly, most of the money will be spent on “new models” meaning that it will be buying Nvidia silicon or renting it from Microsoft or Amazon although the company does intend to double its headcount to 500 this year.
• At $200K per head, this still only amounts to $50m, meaning that the main beneficiary of this raise will be Nvidia once again.
• This is important because when it comes to valuation, everyone looks at the meteoric increase in Nvidia’s share price but this is not where the situation seems out of line.
• What is crazy are the valuations attributed to the likes of OpenAI (~$90bn), Anthropic (~$20bn) and so on.
• Cohere is no different as at an enterprise value of $5bn (net of the $500m just raised) and an annual revenue run rate of $35m, the company is trading on 142x EV/Sales.
• Nvidia by comparison looks to be a bargain on an EV/Sales of 25x and given it is the main beneficiary of all the crazy funding rounds and spending, it is by far the safest play.
• Furthermore, it is not being forced to spend $40bn+ on capex to support all of the compute power being rolled out meaning that it is generating $15bn+ of cash flow every quarter.
• Nvidia will suffer when the correction comes but it is not going to crater like private AI valuations will which in relative terms makes it one of the lower-risk options in this sector.
• I suspect that it will be the private markets that end up being the catalyst for the correction as the hubris being touted to the press and the investment community does not match what is realistically achievable which I have argued many times.
• Consequently, when reality comes knocking, private valuations will fall which will be reported widely in the press which in turn will hammer sentiment as it has happened several times before.
• Generative AI has many lucrative use cases, but the technology is far from mature and costs far too much to implement which leads me to think that it will be some time before it really lives up to its economic potential.
• This is similar to the internet which is now a staple of daily life and worth trillions of dollars but in 2000 lost almost all of its market value before it began living up to its promise.
• I don’t think the collapse will be as bad as this, as there are real products now which generate revenues, but it is going to hurt a lot and put many companies out of business.
• Against this backdrop, I don’t want to go near a direct investment but if I had to make one, it would be Nvidia.
• However, I continue to prefer the adjacencies of inference at the edge and nuclear power.