Over the last few days I came across several pieces of news that, taken separately, seemed to have little to do with one another.
Nvidia agreed to buy Hugging Face for almost US$13 billion. Anthropic committed tens of billions to lock in compute. China is accelerating its own chips. And Deutsche Telekom, Orange, Vodafone, and Telefónica are reportedly studying a consortium to compete for European satellite spectrum and offer connectivity directly to the phone.
AI, chips, data centers, telecommunications. Different stories. But the more I looked at them, the more I seemed to see the same thing.
Everyone is trying to control a larger part of the system and depend a little less on someone else.
Nvidia already has the chips. Now it wants to move further up
Being honest, the one that stopped me the most was Hugging Face. I knew it by name, and I had landed there once or twice looking at published models, but I had never really sized it up. The simplest way to explain it is to think of something like GitHub, but for artificial intelligence. It is not exactly the same, but it helps.
That platform has models from Meta, Google, Microsoft, DeepSeek, Qwen, Nvidia, and thousands of other projects. According to Nvidia, it brings together more than 18 million developers, 3 million models, and more than 200,000 companies. And that is where the acquisition starts to look different.
Nvidia is still known for its GPUs. But it has been building a lot more around them for a while: CUDA, networking, servers, software, models, robotics, infrastructure for data centers. With Hugging Face it gets closer to the place where developers choose which models to use and build on top of them. It is not only buying technology. It is buying ecosystem and distribution. And maybe also insurance. Because many of its best customers are trying, precisely, to depend less on Nvidia.
Google has its TPUs. Amazon is developing Trainium. Microsoft has Maia. Meta has been building its own accelerators for years. China is building alternatives. It all makes sense: if you spend tens of billions a year on compute, you eventually ask yourself how much of that infrastructure you can control yourself.
Today, a few gigantic customers buying millions of GPUs is a spectacular business. But those same customers have the capital and the technical capacity to try to reduce their dependence. An ecosystem with millions of developers and thousands of open models is far more fragmented. And all of them need compute. That is why Hugging Face could be quite a bit more strategic than it looks.
Nvidia has insisted that the platform will remain open and that Nvidia hardware will not be required. That will also be interesting to watch. Part of Hugging Face’s value is that it is a relatively neutral meeting point. If it simply becomes “Nvidia’s model platform,” it could destroy exactly what made the acquisition attractive.
Anthropic is moving down toward infrastructure
When we think of Anthropic we think of Claude. Of models. Of Claude Code. Not of data centers.
But at the end of August it agreed to pay around US$45 billion for Nscale compute in West Virginia. A few days later, about US$35 billion with Lambda for infrastructure in Texas. That is close to US$80 billion in compute commitments in a matter of days.
When the numbers reach that level, it gets hard to keep thinking this race is only about who has the smartest model. Behind Claude you have to guarantee chips, electricity, data centers, cooling, networking, and capacity for millions of users. A good model without enough infrastructure is not worth much. And when the resource becomes strategic, you secure it years in advance.
The companies that started with models are moving down toward infrastructure. The ones that make infrastructure are moving up toward models and developers. The borders start to disappear.
China has a similar problem, but at a different scale
For years Nvidia dominated AI chips. U.S. export restrictions changed the incentives completely. If tomorrow you can lose a critical technology, developing an alternative stops being a business opportunity. It becomes a strategic necessity.
Huawei is the best-known name, but not the only one. Enflame, Moore Threads, MetaX, and Biren are also trying to compete. And not only in silicon: they also want to reduce CUDA, the software ecosystem around Nvidia’s GPUs.
According to Reuters Breakingviews, Nvidia’s share of the Chinese AI chip market has fallen to around 55%, from levels close to a monopoly. There is still an important gap. But the direction is clear: China wants to build its AI infrastructure without depending completely on U.S. technology.
The same pattern again. Control more. Depend less.
And then there are the telcos looking toward space
This last one caught my attention especially, maybe because it takes the same phenomenon out of AI and puts it in my world.
On September 7 it emerged that Deutsche Telekom, Orange, Vodafone, and Telefónica are in preliminary talks to create a consortium and compete for satellite spectrum that the European Union wants to reserve for European operators. The idea would be direct-to-mobile services: connecting the phone by satellite when there is no terrestrial coverage, and eventually competing with Starlink.
There is still no decision. But the fact that four operators that normally compete with one another are considering doing this together already seems interesting to me. Europe does not want an increasingly important layer of its communications to end up entirely in U.S. hands. And the telcos also do not want to discover, in a few years, that the satellite layer that complements their networks belongs to someone else.
I wrote something similar recently in another post. Telcos have spent years selling terrestrial connectivity: spectrum, towers, fiber, cable. But those borders are moving. Fiber, 5G, cloud, edge, satellites. At some point they stop being separate networks and become the same system. And then the question that, at least for me, sits underneath all of this appears: which parts of that system can you not afford not to control.
Maybe the next race is not only about building something better
For years the industry moved toward more specialization. One company made chips. Another, servers. Another, cloud. Another, software. Another, the network. Another, the applications. The logic was simple: you do not need to do everything if someone else can do it better and cheaper.
That logic is still valid. But I think another variable is being added to it: the risk of depending too much on someone else.
Nvidia: chips → systems → software → models → developers.
Anthropic: models → compute → infrastructure.
China: applications → models → chips → their own ecosystems.
The telcos: mobile → fiber → cloud → satellite.
I do not think every company will end up doing everything. That would be inefficient and, in many cases, impossible. There is an important difference between buying something from a supplier and depending on that supplier for your business to exist.
We keep looking at who has the best model, the fastest chip, or the best network. That matters. But underneath there is another race that I find more and more interesting: controlling the pieces that really matter.
Maybe these four stories are isolated moves. Or maybe we are entering a stage where the advantage will not only be having the best technology.
It will also be deciding which parts of it you cannot afford to leave completely in someone else’s hands.
Related reading:
- The Next Telco Business Is Not Selling More Speed
- Google Is Connecting Chile to Asia
- The Invisible Giants
✍️ Claudio from ViaMind
“Dare to imagine, create and transform.”
FEATURED POSTS
ABOUT ME
ViaMind Ecosystem
- ViaMind — The brand hub for ideas, technology and strategy.
- NeuraPRO — Practical software for managing real operations.
- ViaMind Radar — Less noise, more signal. AI-curated technology stories.