Interest in the tech industry is high over 2 mergers and acquisitions that have taken place recently in the AI market. One is Nvidia's $12.9 billion acquisition of Hugging Face. The other is payments platform Stripe's $7.5 billion purchase of AI model routing company OpenRouter.

Both deals have prompted many in the industry to ask, "Why did they buy it?" They have become a topic of discussion, and various experts have offered their own interpretations.

In this context, C.C. Gong (공), a principal at Menlo Ventures who oversees investments in consumer tech and AI, also shared his thoughts on the 2 deals on social media platform X (Twitter).

According to him, sentiment toward Hugging Face and OpenRouter was negative before the acquisitions. Hugging Face was seen as lacking profitability, while OpenRouter was tagged as having "no moat."

He did not appear to agree with that view. He took the opposite position. AI models themselves are becoming common, and both companies benefited from that, he said.

He said, "As the market fragments, the value of those that aggregate grows. Hugging Face did not need to build the best-performing model itself. It only needed to be the first place where each new model shows up. It started as a chatbot service, but has now become a model marketplace. Model weights are just files, so they can be stored anywhere, but Hugging Face added search, model cards, version control, standardised APIs and libraries. Above all, it built a community."

He added, "Hugging Face's growth secret was making it so that when a new open-source model comes out, it appears on Hugging Face right away. Developers knew they could always find the latest models there, and they kept coming back. As developers gathered, model developers ultimately had no choice but to upload their models there."

Nvidia controls computing, and Hugging Face has secured a point of contact where developers decide which model to run and where to run it. Combined, Nvidia can intervene on both sides.

He said, "With optimised runtimes, inference endpoints, deployment defaults, and tight integration with CUDA and Nvidia hardware, it can make Nvidia infrastructure the most convenient choice," adding that "Hugging Face is a tool to lift Nvidia's computing business growth rate."

OpenRouter has the same structure as Hugging Face. With prices, speed and availability varying across model providers, OpenRouter has focused on selecting and connecting the right models.

As millions of developers flocked to OpenRouter, model providers also began to pay attention. Prices and availability improved, and the value returned to developers increased. In matching supply and demand, it is closer to a marketplace structure than a simple router.

Gong said, "OpenRouter processes more than 10 trillion tokens a day, accumulating vast data on which developers use which models for which tasks. Even if you installed an open-source router separately at 10,000 companies, you could not gather this level of data," adding, "Once the scale of what you collect exceeds a certain level, what looked like a simple tool starts to move like a network. This network effect gives the acquirer a new distribution channel."

He explained that Hugging Face enables Nvidia to build a direct relationship with developers who choose models, going beyond a relationship of selling computing to them, while OpenRouter enables Stripe to build a relationship not after payment but with the flow of AI usage itself.

Gong said, "Neither Hugging Face nor OpenRouter was a hard product to build. But if a company that already controls an adjacent layer acquires it, the network value grows further, because the acquirer can turn that network into money from positions the acquired company could not monetise on its own," adding, "In AI, the scarce asset is not the model. The real power lies with whoever secures the distribution layer around the model. In the end, the moat that lasts is the network itself."