Nvidia & Hugging Face – Health Insurance

Nvidia invests to keep its share and margins in rude health.

• Nvidia’s purchase of Hugging Face is about ensuring that no one becomes dominant in AI models and the services that are built on them, which will go a long way to maintaining its competitive edge in silicon.
• Nvidia has a long history of acting early and decisively to ensure the long-term prosperity of the company, and I think that this is what this acquisition is all about.
• Nvidia and Hugging Face have agreed on an $11.9bn acquisition, with a further $1bn being offered as an equity-linked retention bonus to Hugging Face staff, bringing the total to $12.9bn.
• Nvidia has committed to leaving Hugging Face exactly the way that it is, which will cause all sorts of sceptical opinions about the long-term independence of Hugging Face, just as many competitors to Nvidia are beginning to emerge.
• However, I think that it is in Nvidia’s best self-interest to maintain Hugging Face completely independent because this is how it can prevent market power at the top of the AI stack from someone like Anthropic from causing market share loss in silicon, where it makes all of its money.
• Historically, Nvidia’s market power comes from the CUDA development platform, which is an industry standard and its product cadence, where its silicon is routinely at least one generation ahead of everyone else.
• However, the CUDA lock-in is at its strongest for training, and is confined to the silicon layer, meaning that if industry development moves to models, then the lock-in weakens considerably.
• This is what RFM Research refers to as AI Ecosystem 2.0 (see here)
• Furthermore, CUDA is far less relevant when running inference as opposed to training, and so as the industry migrates towards inference and developing on pre-existing models, CUDA’s influence will wane.
• With 85% market share in AI silicon and 70%+ gross margins, Nvidia has a position to protect, and competitive pressure is ramping up substantially.
• This has been obvious for some time, and Nvidia has already made a number of moves to mitigate increasing competition and acquiring Hugging Face is the latest iteration.
• The knee-jerk reaction is to think that Nvidia will now create optimisations via Hugging Face that mean that its open models, such as Nemotron, work better, thereby creating an incentive to use Nvidia silicon but I think its intentions are far more subtle.
• The problem that Hugging Face has today is that it has investors that require a financial return, meaning that Hugging Face is constrained in terms of what it can spend and it needs to have a path to profitability,
• Now that it will be bankrolled by Nvidia, Hugging Face can invest to grow and consolidate its position as the go-to place for open-source AI models without having to worry about making money.
• As long as Hugging Face continues to dominate, no one will be able to corner the market for AI services, which is an area where Nvidia’s market position is far weaker than it is in silicon.
• For example, if OpenAI or Google were to become by far the dominant enabler of AI services, then the entire ecosystem could easily become optimised for their in-house silicon rather than Nvidia, which would destroy Nvidia’s ability to price its products at a premium, and it would very likely cause a vast amount of market share loss.
• Hence, for as long as open source provides a viable and thriving alternative, this future is much less likely to become a reality, as there will always be other places to go and no one platform will become dominant.
• A thriving Hugging Face reduces the risk of market power being exerted against it from another location in the AI stack, which is what I think is the most likely reason for the acquisition.
• The net result is that I don’t see hurdles to this acquisition and suspect that Nvidia will go all out to ensure that Hugging Face stays independent and continues to expand.
• The payback will be in greater durability of market share and profitability, which, given the size of the data centre systems business, $13bn is money well spent.
• I continue to be interested in Nvidia but have yet to take a position.